1 c Copyright (C) 1988-2018 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 * Pointers to Arrays:: Pointers to arrays with qualifiers work as expected.
50 * Initializers:: Non-constant initializers.
51 * Compound Literals:: Compound literals give structures, unions
53 * Designated Inits:: Labeling elements of initializers.
54 * Case Ranges:: `case 1 ... 9' and such.
55 * Cast to Union:: Casting to union type from any member of the union.
56 * Mixed Declarations:: Mixing declarations and code.
57 * Function Attributes:: Declaring that functions have no side effects,
58 or that they can never return.
59 * Variable Attributes:: Specifying attributes of variables.
60 * Type Attributes:: Specifying attributes of types.
61 * Label Attributes:: Specifying attributes on labels.
62 * Enumerator Attributes:: Specifying attributes on enumerators.
63 * Statement Attributes:: Specifying attributes on statements.
64 * Attribute Syntax:: Formal syntax for attributes.
65 * Function Prototypes:: Prototype declarations and old-style definitions.
66 * C++ Comments:: C++ comments are recognized.
67 * Dollar Signs:: Dollar sign is allowed in identifiers.
68 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
69 * Alignment:: Inquiring about the alignment of a type or variable.
70 * Inline:: Defining inline functions (as fast as macros).
71 * Volatiles:: What constitutes an access to a volatile object.
72 * Using Assembly Language with C:: Instructions and extensions for interfacing C with assembler.
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 * Integer Overflow Builtins:: Built-in functions to perform arithmetics and
83 arithmetic overflow checking.
84 * x86 specific memory model extensions for transactional memory:: x86 memory models.
85 * Object Size Checking:: Built-in functions for limited buffer overflow
87 * Pointer Bounds Checker builtins:: Built-in functions for Pointer Bounds Checker.
88 * Other Builtins:: Other built-in functions.
89 * Target Builtins:: Built-in functions specific to particular targets.
90 * Target Format Checks:: Format checks specific to particular targets.
91 * Pragmas:: Pragmas accepted by GCC.
92 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
93 * Thread-Local:: Per-thread variables.
94 * Binary constants:: Binary constants using the @samp{0b} prefix.
98 @section Statements and Declarations in Expressions
99 @cindex statements inside expressions
100 @cindex declarations inside expressions
101 @cindex expressions containing statements
102 @cindex macros, statements in expressions
104 @c the above section title wrapped and causes an underfull hbox.. i
105 @c changed it from "within" to "in". --mew 4feb93
106 A compound statement enclosed in parentheses may appear as an expression
107 in GNU C@. This allows you to use loops, switches, and local variables
108 within an expression.
110 Recall that a compound statement is a sequence of statements surrounded
111 by braces; in this construct, parentheses go around the braces. For
115 (@{ int y = foo (); int z;
122 is a valid (though slightly more complex than necessary) expression
123 for the absolute value of @code{foo ()}.
125 The last thing in the compound statement should be an expression
126 followed by a semicolon; the value of this subexpression serves as the
127 value of the entire construct. (If you use some other kind of statement
128 last within the braces, the construct has type @code{void}, and thus
129 effectively no value.)
131 This feature is especially useful in making macro definitions ``safe'' (so
132 that they evaluate each operand exactly once). For example, the
133 ``maximum'' function is commonly defined as a macro in standard C as
137 #define max(a,b) ((a) > (b) ? (a) : (b))
141 @cindex side effects, macro argument
142 But this definition computes either @var{a} or @var{b} twice, with bad
143 results if the operand has side effects. In GNU C, if you know the
144 type of the operands (here taken as @code{int}), you can define
145 the macro safely as follows:
148 #define maxint(a,b) \
149 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
152 Embedded statements are not allowed in constant expressions, such as
153 the value of an enumeration constant, the width of a bit-field, or
154 the initial value of a static variable.
156 If you don't know the type of the operand, you can still do this, but you
157 must use @code{typeof} or @code{__auto_type} (@pxref{Typeof}).
159 In G++, the result value of a statement expression undergoes array and
160 function pointer decay, and is returned by value to the enclosing
161 expression. For instance, if @code{A} is a class, then
170 constructs a temporary @code{A} object to hold the result of the
171 statement expression, and that is used to invoke @code{Foo}.
172 Therefore the @code{this} pointer observed by @code{Foo} is not the
175 In a statement expression, any temporaries created within a statement
176 are destroyed at that statement's end. This makes statement
177 expressions inside macros slightly different from function calls. In
178 the latter case temporaries introduced during argument evaluation are
179 destroyed at the end of the statement that includes the function
180 call. In the statement expression case they are destroyed during
181 the statement expression. For instance,
184 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
185 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
195 has different places where temporaries are destroyed. For the
196 @code{macro} case, the temporary @code{X} is destroyed just after
197 the initialization of @code{b}. In the @code{function} case that
198 temporary is destroyed when the function returns.
200 These considerations mean that it is probably a bad idea to use
201 statement expressions of this form in header files that are designed to
202 work with C++. (Note that some versions of the GNU C Library contained
203 header files using statement expressions that lead to precisely this
206 Jumping into a statement expression with @code{goto} or using a
207 @code{switch} statement outside the statement expression with a
208 @code{case} or @code{default} label inside the statement expression is
209 not permitted. Jumping into a statement expression with a computed
210 @code{goto} (@pxref{Labels as Values}) has undefined behavior.
211 Jumping out of a statement expression is permitted, but if the
212 statement expression is part of a larger expression then it is
213 unspecified which other subexpressions of that expression have been
214 evaluated except where the language definition requires certain
215 subexpressions to be evaluated before or after the statement
216 expression. In any case, as with a function call, the evaluation of a
217 statement expression is not interleaved with the evaluation of other
218 parts of the containing expression. For example,
221 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
225 calls @code{foo} and @code{bar1} and does not call @code{baz} but
226 may or may not call @code{bar2}. If @code{bar2} is called, it is
227 called after @code{foo} and before @code{bar1}.
230 @section Locally Declared Labels
232 @cindex macros, local labels
234 GCC allows you to declare @dfn{local labels} in any nested block
235 scope. A local label is just like an ordinary label, but you can
236 only reference it (with a @code{goto} statement, or by taking its
237 address) within the block in which it is declared.
239 A local label declaration looks like this:
242 __label__ @var{label};
249 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
252 Local label declarations must come at the beginning of the block,
253 before any ordinary declarations or statements.
255 The label declaration defines the label @emph{name}, but does not define
256 the label itself. You must do this in the usual way, with
257 @code{@var{label}:}, within the statements of the statement expression.
259 The local label feature is useful for complex macros. If a macro
260 contains nested loops, a @code{goto} can be useful for breaking out of
261 them. However, an ordinary label whose scope is the whole function
262 cannot be used: if the macro can be expanded several times in one
263 function, the label is multiply defined in that function. A
264 local label avoids this problem. For example:
267 #define SEARCH(value, array, target) \
270 typeof (target) _SEARCH_target = (target); \
271 typeof (*(array)) *_SEARCH_array = (array); \
274 for (i = 0; i < max; i++) \
275 for (j = 0; j < max; j++) \
276 if (_SEARCH_array[i][j] == _SEARCH_target) \
277 @{ (value) = i; goto found; @} \
283 This could also be written using a statement expression:
286 #define SEARCH(array, target) \
289 typeof (target) _SEARCH_target = (target); \
290 typeof (*(array)) *_SEARCH_array = (array); \
293 for (i = 0; i < max; i++) \
294 for (j = 0; j < max; j++) \
295 if (_SEARCH_array[i][j] == _SEARCH_target) \
296 @{ value = i; goto found; @} \
303 Local label declarations also make the labels they declare visible to
304 nested functions, if there are any. @xref{Nested Functions}, for details.
306 @node Labels as Values
307 @section Labels as Values
308 @cindex labels as values
309 @cindex computed gotos
310 @cindex goto with computed label
311 @cindex address of a label
313 You can get the address of a label defined in the current function
314 (or a containing function) with the unary operator @samp{&&}. The
315 value has type @code{void *}. This value is a constant and can be used
316 wherever a constant of that type is valid. For example:
324 To use these values, you need to be able to jump to one. This is done
325 with the computed goto statement@footnote{The analogous feature in
326 Fortran is called an assigned goto, but that name seems inappropriate in
327 C, where one can do more than simply store label addresses in label
328 variables.}, @code{goto *@var{exp};}. For example,
335 Any expression of type @code{void *} is allowed.
337 One way of using these constants is in initializing a static array that
338 serves as a jump table:
341 static void *array[] = @{ &&foo, &&bar, &&hack @};
345 Then you can select a label with indexing, like this:
352 Note that this does not check whether the subscript is in bounds---array
353 indexing in C never does that.
355 Such an array of label values serves a purpose much like that of the
356 @code{switch} statement. The @code{switch} statement is cleaner, so
357 use that rather than an array unless the problem does not fit a
358 @code{switch} statement very well.
360 Another use of label values is in an interpreter for threaded code.
361 The labels within the interpreter function can be stored in the
362 threaded code for super-fast dispatching.
364 You may not use this mechanism to jump to code in a different function.
365 If you do that, totally unpredictable things happen. The best way to
366 avoid this is to store the label address only in automatic variables and
367 never pass it as an argument.
369 An alternate way to write the above example is
372 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
374 goto *(&&foo + array[i]);
378 This is more friendly to code living in shared libraries, as it reduces
379 the number of dynamic relocations that are needed, and by consequence,
380 allows the data to be read-only.
381 This alternative with label differences is not supported for the AVR target,
382 please use the first approach for AVR programs.
384 The @code{&&foo} expressions for the same label might have different
385 values if the containing function is inlined or cloned. If a program
386 relies on them being always the same,
387 @code{__attribute__((__noinline__,__noclone__))} should be used to
388 prevent inlining and cloning. If @code{&&foo} is used in a static
389 variable initializer, inlining and cloning is forbidden.
391 @node Nested Functions
392 @section Nested Functions
393 @cindex nested functions
394 @cindex downward funargs
397 A @dfn{nested function} is a function defined inside another function.
398 Nested functions are supported as an extension in GNU C, but are not
399 supported by GNU C++.
401 The nested function's name is local to the block where it is defined.
402 For example, here we define a nested function named @code{square}, and
407 foo (double a, double b)
409 double square (double z) @{ return z * z; @}
411 return square (a) + square (b);
416 The nested function can access all the variables of the containing
417 function that are visible at the point of its definition. This is
418 called @dfn{lexical scoping}. For example, here we show a nested
419 function which uses an inherited variable named @code{offset}:
423 bar (int *array, int offset, int size)
425 int access (int *array, int index)
426 @{ return array[index + offset]; @}
429 for (i = 0; i < size; i++)
430 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
435 Nested function definitions are permitted within functions in the places
436 where variable definitions are allowed; that is, in any block, mixed
437 with the other declarations and statements in the block.
439 It is possible to call the nested function from outside the scope of its
440 name by storing its address or passing the address to another function:
443 hack (int *array, int size)
445 void store (int index, int value)
446 @{ array[index] = value; @}
448 intermediate (store, size);
452 Here, the function @code{intermediate} receives the address of
453 @code{store} as an argument. If @code{intermediate} calls @code{store},
454 the arguments given to @code{store} are used to store into @code{array}.
455 But this technique works only so long as the containing function
456 (@code{hack}, in this example) does not exit.
458 If you try to call the nested function through its address after the
459 containing function exits, all hell breaks loose. If you try
460 to call it after a containing scope level exits, and if it refers
461 to some of the variables that are no longer in scope, you may be lucky,
462 but it's not wise to take the risk. If, however, the nested function
463 does not refer to anything that has gone out of scope, you should be
466 GCC implements taking the address of a nested function using a technique
467 called @dfn{trampolines}. This technique was described in
468 @cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX
469 C++ Conference Proceedings, October 17-21, 1988).
471 A nested function can jump to a label inherited from a containing
472 function, provided the label is explicitly declared in the containing
473 function (@pxref{Local Labels}). Such a jump returns instantly to the
474 containing function, exiting the nested function that did the
475 @code{goto} and any intermediate functions as well. Here is an example:
479 bar (int *array, int offset, int size)
482 int access (int *array, int index)
486 return array[index + offset];
490 for (i = 0; i < size; i++)
491 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
495 /* @r{Control comes here from @code{access}
496 if it detects an error.} */
503 A nested function always has no linkage. Declaring one with
504 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
505 before its definition, use @code{auto} (which is otherwise meaningless
506 for function declarations).
509 bar (int *array, int offset, int size)
512 auto int access (int *, int);
514 int access (int *array, int index)
518 return array[index + offset];
524 @node Constructing Calls
525 @section Constructing Function Calls
526 @cindex constructing calls
527 @cindex forwarding calls
529 Using the built-in functions described below, you can record
530 the arguments a function received, and call another function
531 with the same arguments, without knowing the number or types
534 You can also record the return value of that function call,
535 and later return that value, without knowing what data type
536 the function tried to return (as long as your caller expects
539 However, these built-in functions may interact badly with some
540 sophisticated features or other extensions of the language. It
541 is, therefore, not recommended to use them outside very simple
542 functions acting as mere forwarders for their arguments.
544 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
545 This built-in function returns a pointer to data
546 describing how to perform a call with the same arguments as are passed
547 to the current function.
549 The function saves the arg pointer register, structure value address,
550 and all registers that might be used to pass arguments to a function
551 into a block of memory allocated on the stack. Then it returns the
552 address of that block.
555 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
556 This built-in function invokes @var{function}
557 with a copy of the parameters described by @var{arguments}
560 The value of @var{arguments} should be the value returned by
561 @code{__builtin_apply_args}. The argument @var{size} specifies the size
562 of the stack argument data, in bytes.
564 This function returns a pointer to data describing
565 how to return whatever value is returned by @var{function}. The data
566 is saved in a block of memory allocated on the stack.
568 It is not always simple to compute the proper value for @var{size}. The
569 value is used by @code{__builtin_apply} to compute the amount of data
570 that should be pushed on the stack and copied from the incoming argument
574 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
575 This built-in function returns the value described by @var{result} from
576 the containing function. You should specify, for @var{result}, a value
577 returned by @code{__builtin_apply}.
580 @deftypefn {Built-in Function} {} __builtin_va_arg_pack ()
581 This built-in function represents all anonymous arguments of an inline
582 function. It can be used only in inline functions that are always
583 inlined, never compiled as a separate function, such as those using
584 @code{__attribute__ ((__always_inline__))} or
585 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
586 It must be only passed as last argument to some other function
587 with variable arguments. This is useful for writing small wrapper
588 inlines for variable argument functions, when using preprocessor
589 macros is undesirable. For example:
591 extern int myprintf (FILE *f, const char *format, ...);
592 extern inline __attribute__ ((__gnu_inline__)) int
593 myprintf (FILE *f, const char *format, ...)
595 int r = fprintf (f, "myprintf: ");
598 int s = fprintf (f, format, __builtin_va_arg_pack ());
606 @deftypefn {Built-in Function} {size_t} __builtin_va_arg_pack_len ()
607 This built-in function returns the number of anonymous arguments of
608 an inline function. It can be used only in inline functions that
609 are always inlined, never compiled as a separate function, such
610 as those using @code{__attribute__ ((__always_inline__))} or
611 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
612 For example following does link- or run-time checking of open
613 arguments for optimized code:
616 extern inline __attribute__((__gnu_inline__)) int
617 myopen (const char *path, int oflag, ...)
619 if (__builtin_va_arg_pack_len () > 1)
620 warn_open_too_many_arguments ();
622 if (__builtin_constant_p (oflag))
624 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
626 warn_open_missing_mode ();
627 return __open_2 (path, oflag);
629 return open (path, oflag, __builtin_va_arg_pack ());
632 if (__builtin_va_arg_pack_len () < 1)
633 return __open_2 (path, oflag);
635 return open (path, oflag, __builtin_va_arg_pack ());
642 @section Referring to a Type with @code{typeof}
645 @cindex macros, types of arguments
647 Another way to refer to the type of an expression is with @code{typeof}.
648 The syntax of using of this keyword looks like @code{sizeof}, but the
649 construct acts semantically like a type name defined with @code{typedef}.
651 There are two ways of writing the argument to @code{typeof}: with an
652 expression or with a type. Here is an example with an expression:
659 This assumes that @code{x} is an array of pointers to functions;
660 the type described is that of the values of the functions.
662 Here is an example with a typename as the argument:
669 Here the type described is that of pointers to @code{int}.
671 If you are writing a header file that must work when included in ISO C
672 programs, write @code{__typeof__} instead of @code{typeof}.
673 @xref{Alternate Keywords}.
675 A @code{typeof} construct can be used anywhere a typedef name can be
676 used. For example, you can use it in a declaration, in a cast, or inside
677 of @code{sizeof} or @code{typeof}.
679 The operand of @code{typeof} is evaluated for its side effects if and
680 only if it is an expression of variably modified type or the name of
683 @code{typeof} is often useful in conjunction with
684 statement expressions (@pxref{Statement Exprs}).
685 Here is how the two together can
686 be used to define a safe ``maximum'' macro which operates on any
687 arithmetic type and evaluates each of its arguments exactly once:
691 (@{ typeof (a) _a = (a); \
692 typeof (b) _b = (b); \
693 _a > _b ? _a : _b; @})
696 @cindex underscores in variables in macros
697 @cindex @samp{_} in variables in macros
698 @cindex local variables in macros
699 @cindex variables, local, in macros
700 @cindex macros, local variables in
702 The reason for using names that start with underscores for the local
703 variables is to avoid conflicts with variable names that occur within the
704 expressions that are substituted for @code{a} and @code{b}. Eventually we
705 hope to design a new form of declaration syntax that allows you to declare
706 variables whose scopes start only after their initializers; this will be a
707 more reliable way to prevent such conflicts.
710 Some more examples of the use of @code{typeof}:
714 This declares @code{y} with the type of what @code{x} points to.
721 This declares @code{y} as an array of such values.
728 This declares @code{y} as an array of pointers to characters:
731 typeof (typeof (char *)[4]) y;
735 It is equivalent to the following traditional C declaration:
741 To see the meaning of the declaration using @code{typeof}, and why it
742 might be a useful way to write, rewrite it with these macros:
745 #define pointer(T) typeof(T *)
746 #define array(T, N) typeof(T [N])
750 Now the declaration can be rewritten this way:
753 array (pointer (char), 4) y;
757 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
758 pointers to @code{char}.
761 In GNU C, but not GNU C++, you may also declare the type of a variable
762 as @code{__auto_type}. In that case, the declaration must declare
763 only one variable, whose declarator must just be an identifier, the
764 declaration must be initialized, and the type of the variable is
765 determined by the initializer; the name of the variable is not in
766 scope until after the initializer. (In C++, you should use C++11
767 @code{auto} for this purpose.) Using @code{__auto_type}, the
768 ``maximum'' macro above could be written as:
772 (@{ __auto_type _a = (a); \
773 __auto_type _b = (b); \
774 _a > _b ? _a : _b; @})
777 Using @code{__auto_type} instead of @code{typeof} has two advantages:
780 @item Each argument to the macro appears only once in the expansion of
781 the macro. This prevents the size of the macro expansion growing
782 exponentially when calls to such macros are nested inside arguments of
785 @item If the argument to the macro has variably modified type, it is
786 evaluated only once when using @code{__auto_type}, but twice if
787 @code{typeof} is used.
791 @section Conditionals with Omitted Operands
792 @cindex conditional expressions, extensions
793 @cindex omitted middle-operands
794 @cindex middle-operands, omitted
795 @cindex extensions, @code{?:}
796 @cindex @code{?:} extensions
798 The middle operand in a conditional expression may be omitted. Then
799 if the first operand is nonzero, its value is the value of the conditional
802 Therefore, the expression
809 has the value of @code{x} if that is nonzero; otherwise, the value of
812 This example is perfectly equivalent to
818 @cindex side effect in @code{?:}
819 @cindex @code{?:} side effect
821 In this simple case, the ability to omit the middle operand is not
822 especially useful. When it becomes useful is when the first operand does,
823 or may (if it is a macro argument), contain a side effect. Then repeating
824 the operand in the middle would perform the side effect twice. Omitting
825 the middle operand uses the value already computed without the undesirable
826 effects of recomputing it.
829 @section 128-bit Integers
830 @cindex @code{__int128} data types
832 As an extension the integer scalar type @code{__int128} is supported for
833 targets which have an integer mode wide enough to hold 128 bits.
834 Simply write @code{__int128} for a signed 128-bit integer, or
835 @code{unsigned __int128} for an unsigned 128-bit integer. There is no
836 support in GCC for expressing an integer constant of type @code{__int128}
837 for targets with @code{long long} integer less than 128 bits wide.
840 @section Double-Word Integers
841 @cindex @code{long long} data types
842 @cindex double-word arithmetic
843 @cindex multiprecision arithmetic
844 @cindex @code{LL} integer suffix
845 @cindex @code{ULL} integer suffix
847 ISO C99 supports data types for integers that are at least 64 bits wide,
848 and as an extension GCC supports them in C90 mode and in C++.
849 Simply write @code{long long int} for a signed integer, or
850 @code{unsigned long long int} for an unsigned integer. To make an
851 integer constant of type @code{long long int}, add the suffix @samp{LL}
852 to the integer. To make an integer constant of type @code{unsigned long
853 long int}, add the suffix @samp{ULL} to the integer.
855 You can use these types in arithmetic like any other integer types.
856 Addition, subtraction, and bitwise boolean operations on these types
857 are open-coded on all types of machines. Multiplication is open-coded
858 if the machine supports a fullword-to-doubleword widening multiply
859 instruction. Division and shifts are open-coded only on machines that
860 provide special support. The operations that are not open-coded use
861 special library routines that come with GCC@.
863 There may be pitfalls when you use @code{long long} types for function
864 arguments without function prototypes. If a function
865 expects type @code{int} for its argument, and you pass a value of type
866 @code{long long int}, confusion results because the caller and the
867 subroutine disagree about the number of bytes for the argument.
868 Likewise, if the function expects @code{long long int} and you pass
869 @code{int}. The best way to avoid such problems is to use prototypes.
872 @section Complex Numbers
873 @cindex complex numbers
874 @cindex @code{_Complex} keyword
875 @cindex @code{__complex__} keyword
877 ISO C99 supports complex floating data types, and as an extension GCC
878 supports them in C90 mode and in C++. GCC also supports complex integer data
879 types which are not part of ISO C99. You can declare complex types
880 using the keyword @code{_Complex}. As an extension, the older GNU
881 keyword @code{__complex__} is also supported.
883 For example, @samp{_Complex double x;} declares @code{x} as a
884 variable whose real part and imaginary part are both of type
885 @code{double}. @samp{_Complex short int y;} declares @code{y} to
886 have real and imaginary parts of type @code{short int}; this is not
887 likely to be useful, but it shows that the set of complex types is
890 To write a constant with a complex data type, use the suffix @samp{i} or
891 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
892 has type @code{_Complex float} and @code{3i} has type
893 @code{_Complex int}. Such a constant always has a pure imaginary
894 value, but you can form any complex value you like by adding one to a
895 real constant. This is a GNU extension; if you have an ISO C99
896 conforming C library (such as the GNU C Library), and want to construct complex
897 constants of floating type, you should include @code{<complex.h>} and
898 use the macros @code{I} or @code{_Complex_I} instead.
900 The ISO C++14 library also defines the @samp{i} suffix, so C++14 code
901 that includes the @samp{<complex>} header cannot use @samp{i} for the
902 GNU extension. The @samp{j} suffix still has the GNU meaning.
904 @cindex @code{__real__} keyword
905 @cindex @code{__imag__} keyword
906 To extract the real part of a complex-valued expression @var{exp}, write
907 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
908 extract the imaginary part. This is a GNU extension; for values of
909 floating type, you should use the ISO C99 functions @code{crealf},
910 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
911 @code{cimagl}, declared in @code{<complex.h>} and also provided as
912 built-in functions by GCC@.
914 @cindex complex conjugation
915 The operator @samp{~} performs complex conjugation when used on a value
916 with a complex type. This is a GNU extension; for values of
917 floating type, you should use the ISO C99 functions @code{conjf},
918 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
919 provided as built-in functions by GCC@.
921 GCC can allocate complex automatic variables in a noncontiguous
922 fashion; it's even possible for the real part to be in a register while
923 the imaginary part is on the stack (or vice versa). Only the DWARF
924 debug info format can represent this, so use of DWARF is recommended.
925 If you are using the stabs debug info format, GCC describes a noncontiguous
926 complex variable as if it were two separate variables of noncomplex type.
927 If the variable's actual name is @code{foo}, the two fictitious
928 variables are named @code{foo$real} and @code{foo$imag}. You can
929 examine and set these two fictitious variables with your debugger.
932 @section Additional Floating Types
933 @cindex additional floating types
934 @cindex @code{_Float@var{n}} data types
935 @cindex @code{_Float@var{n}x} data types
936 @cindex @code{__float80} data type
937 @cindex @code{__float128} data type
938 @cindex @code{__ibm128} data type
939 @cindex @code{w} floating point suffix
940 @cindex @code{q} floating point suffix
941 @cindex @code{W} floating point suffix
942 @cindex @code{Q} floating point suffix
944 ISO/IEC TS 18661-3:2015 defines C support for additional floating
945 types @code{_Float@var{n}} and @code{_Float@var{n}x}, and GCC supports
946 these type names; the set of types supported depends on the target
947 architecture. These types are not supported when compiling C++.
948 Constants with these types use suffixes @code{f@var{n}} or
949 @code{F@var{n}} and @code{f@var{n}x} or @code{F@var{n}x}. These type
950 names can be used together with @code{_Complex} to declare complex
953 As an extension, GNU C and GNU C++ support additional floating
954 types, which are not supported by all targets.
956 @item @code{__float128} is available on i386, x86_64, IA-64, and
957 hppa HP-UX, as well as on PowerPC GNU/Linux targets that enable
958 the vector scalar (VSX) instruction set. @code{__float128} supports
959 the 128-bit floating type. On i386, x86_64, PowerPC, and IA-64
960 other than HP-UX, @code{__float128} is an alias for @code{_Float128}.
961 On hppa and IA-64 HP-UX, @code{__float128} is an alias for @code{long
964 @item @code{__float80} is available on the i386, x86_64, and IA-64
965 targets, and supports the 80-bit (@code{XFmode}) floating type. It is
966 an alias for the type name @code{_Float64x} on these targets.
968 @item @code{__ibm128} is available on PowerPC targets, and provides
969 access to the IBM extended double format which is the current format
970 used for @code{long double}. When @code{long double} transitions to
971 @code{__float128} on PowerPC in the future, @code{__ibm128} will remain
972 for use in conversions between the two types.
975 Support for these additional types includes the arithmetic operators:
976 add, subtract, multiply, divide; unary arithmetic operators;
977 relational operators; equality operators; and conversions to and from
978 integer and other floating types. Use a suffix @samp{w} or @samp{W}
979 in a literal constant of type @code{__float80} or type
980 @code{__ibm128}. Use a suffix @samp{q} or @samp{Q} for @code{_float128}.
982 In order to use @code{_Float128}, @code{__float128}, and @code{__ibm128}
983 on PowerPC Linux systems, you must use the @option{-mfloat128} option. It is
984 expected in future versions of GCC that @code{_Float128} and @code{__float128}
985 will be enabled automatically.
987 The @code{_Float128} type is supported on all systems where
988 @code{__float128} is supported or where @code{long double} has the
989 IEEE binary128 format. The @code{_Float64x} type is supported on all
990 systems where @code{__float128} is supported. The @code{_Float32}
991 type is supported on all systems supporting IEEE binary32; the
992 @code{_Float64} and @code{_Float32x} types are supported on all systems
993 supporting IEEE binary64. The @code{_Float16} type is supported on AArch64
994 systems by default, and on ARM systems when the IEEE format for 16-bit
995 floating-point types is selected with @option{-mfp16-format=ieee}.
996 GCC does not currently support @code{_Float128x} on any systems.
998 On the i386, x86_64, IA-64, and HP-UX targets, you can declare complex
999 types using the corresponding internal complex type, @code{XCmode} for
1000 @code{__float80} type and @code{TCmode} for @code{__float128} type:
1003 typedef _Complex float __attribute__((mode(TC))) _Complex128;
1004 typedef _Complex float __attribute__((mode(XC))) _Complex80;
1007 On the PowerPC Linux VSX targets, you can declare complex types using
1008 the corresponding internal complex type, @code{KCmode} for
1009 @code{__float128} type and @code{ICmode} for @code{__ibm128} type:
1012 typedef _Complex float __attribute__((mode(KC))) _Complex_float128;
1013 typedef _Complex float __attribute__((mode(IC))) _Complex_ibm128;
1016 @node Half-Precision
1017 @section Half-Precision Floating Point
1018 @cindex half-precision floating point
1019 @cindex @code{__fp16} data type
1021 On ARM and AArch64 targets, GCC supports half-precision (16-bit) floating
1022 point via the @code{__fp16} type defined in the ARM C Language Extensions.
1023 On ARM systems, you must enable this type explicitly with the
1024 @option{-mfp16-format} command-line option in order to use it.
1026 ARM targets support two incompatible representations for half-precision
1027 floating-point values. You must choose one of the representations and
1028 use it consistently in your program.
1030 Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format.
1031 This format can represent normalized values in the range of @math{2^{-14}} to 65504.
1032 There are 11 bits of significand precision, approximately 3
1035 Specifying @option{-mfp16-format=alternative} selects the ARM
1036 alternative format. This representation is similar to the IEEE
1037 format, but does not support infinities or NaNs. Instead, the range
1038 of exponents is extended, so that this format can represent normalized
1039 values in the range of @math{2^{-14}} to 131008.
1041 The GCC port for AArch64 only supports the IEEE 754-2008 format, and does
1042 not require use of the @option{-mfp16-format} command-line option.
1044 The @code{__fp16} type may only be used as an argument to intrinsics defined
1045 in @code{<arm_fp16.h>}, or as a storage format. For purposes of
1046 arithmetic and other operations, @code{__fp16} values in C or C++
1047 expressions are automatically promoted to @code{float}.
1049 The ARM target provides hardware support for conversions between
1050 @code{__fp16} and @code{float} values
1051 as an extension to VFP and NEON (Advanced SIMD), and from ARMv8-A provides
1052 hardware support for conversions between @code{__fp16} and @code{double}
1053 values. GCC generates code using these hardware instructions if you
1054 compile with options to select an FPU that provides them;
1055 for example, @option{-mfpu=neon-fp16 -mfloat-abi=softfp},
1056 in addition to the @option{-mfp16-format} option to select
1057 a half-precision format.
1059 Language-level support for the @code{__fp16} data type is
1060 independent of whether GCC generates code using hardware floating-point
1061 instructions. In cases where hardware support is not specified, GCC
1062 implements conversions between @code{__fp16} and other types as library
1065 It is recommended that portable code use the @code{_Float16} type defined
1066 by ISO/IEC TS 18661-3:2015. @xref{Floating Types}.
1069 @section Decimal Floating Types
1070 @cindex decimal floating types
1071 @cindex @code{_Decimal32} data type
1072 @cindex @code{_Decimal64} data type
1073 @cindex @code{_Decimal128} data type
1074 @cindex @code{df} integer suffix
1075 @cindex @code{dd} integer suffix
1076 @cindex @code{dl} integer suffix
1077 @cindex @code{DF} integer suffix
1078 @cindex @code{DD} integer suffix
1079 @cindex @code{DL} integer suffix
1081 As an extension, GNU C supports decimal floating types as
1082 defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal
1083 floating types in GCC will evolve as the draft technical report changes.
1084 Calling conventions for any target might also change. Not all targets
1085 support decimal floating types.
1087 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
1088 @code{_Decimal128}. They use a radix of ten, unlike the floating types
1089 @code{float}, @code{double}, and @code{long double} whose radix is not
1090 specified by the C standard but is usually two.
1092 Support for decimal floating types includes the arithmetic operators
1093 add, subtract, multiply, divide; unary arithmetic operators;
1094 relational operators; equality operators; and conversions to and from
1095 integer and other floating types. Use a suffix @samp{df} or
1096 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
1097 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
1100 GCC support of decimal float as specified by the draft technical report
1105 When the value of a decimal floating type cannot be represented in the
1106 integer type to which it is being converted, the result is undefined
1107 rather than the result value specified by the draft technical report.
1110 GCC does not provide the C library functionality associated with
1111 @file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and
1112 @file{wchar.h}, which must come from a separate C library implementation.
1113 Because of this the GNU C compiler does not define macro
1114 @code{__STDC_DEC_FP__} to indicate that the implementation conforms to
1115 the technical report.
1118 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
1119 are supported by the DWARF debug information format.
1125 ISO C99 supports floating-point numbers written not only in the usual
1126 decimal notation, such as @code{1.55e1}, but also numbers such as
1127 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1128 supports this in C90 mode (except in some cases when strictly
1129 conforming) and in C++. In that format the
1130 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1131 mandatory. The exponent is a decimal number that indicates the power of
1132 2 by which the significant part is multiplied. Thus @samp{0x1.f} is
1139 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1140 is the same as @code{1.55e1}.
1142 Unlike for floating-point numbers in the decimal notation the exponent
1143 is always required in the hexadecimal notation. Otherwise the compiler
1144 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1145 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1146 extension for floating-point constants of type @code{float}.
1149 @section Fixed-Point Types
1150 @cindex fixed-point types
1151 @cindex @code{_Fract} data type
1152 @cindex @code{_Accum} data type
1153 @cindex @code{_Sat} data type
1154 @cindex @code{hr} fixed-suffix
1155 @cindex @code{r} fixed-suffix
1156 @cindex @code{lr} fixed-suffix
1157 @cindex @code{llr} fixed-suffix
1158 @cindex @code{uhr} fixed-suffix
1159 @cindex @code{ur} fixed-suffix
1160 @cindex @code{ulr} fixed-suffix
1161 @cindex @code{ullr} fixed-suffix
1162 @cindex @code{hk} fixed-suffix
1163 @cindex @code{k} fixed-suffix
1164 @cindex @code{lk} fixed-suffix
1165 @cindex @code{llk} fixed-suffix
1166 @cindex @code{uhk} fixed-suffix
1167 @cindex @code{uk} fixed-suffix
1168 @cindex @code{ulk} fixed-suffix
1169 @cindex @code{ullk} fixed-suffix
1170 @cindex @code{HR} fixed-suffix
1171 @cindex @code{R} fixed-suffix
1172 @cindex @code{LR} fixed-suffix
1173 @cindex @code{LLR} fixed-suffix
1174 @cindex @code{UHR} fixed-suffix
1175 @cindex @code{UR} fixed-suffix
1176 @cindex @code{ULR} fixed-suffix
1177 @cindex @code{ULLR} fixed-suffix
1178 @cindex @code{HK} fixed-suffix
1179 @cindex @code{K} fixed-suffix
1180 @cindex @code{LK} fixed-suffix
1181 @cindex @code{LLK} fixed-suffix
1182 @cindex @code{UHK} fixed-suffix
1183 @cindex @code{UK} fixed-suffix
1184 @cindex @code{ULK} fixed-suffix
1185 @cindex @code{ULLK} fixed-suffix
1187 As an extension, GNU C supports fixed-point types as
1188 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1189 types in GCC will evolve as the draft technical report changes.
1190 Calling conventions for any target might also change. Not all targets
1191 support fixed-point types.
1193 The fixed-point types are
1194 @code{short _Fract},
1197 @code{long long _Fract},
1198 @code{unsigned short _Fract},
1199 @code{unsigned _Fract},
1200 @code{unsigned long _Fract},
1201 @code{unsigned long long _Fract},
1202 @code{_Sat short _Fract},
1204 @code{_Sat long _Fract},
1205 @code{_Sat long long _Fract},
1206 @code{_Sat unsigned short _Fract},
1207 @code{_Sat unsigned _Fract},
1208 @code{_Sat unsigned long _Fract},
1209 @code{_Sat unsigned long long _Fract},
1210 @code{short _Accum},
1213 @code{long long _Accum},
1214 @code{unsigned short _Accum},
1215 @code{unsigned _Accum},
1216 @code{unsigned long _Accum},
1217 @code{unsigned long long _Accum},
1218 @code{_Sat short _Accum},
1220 @code{_Sat long _Accum},
1221 @code{_Sat long long _Accum},
1222 @code{_Sat unsigned short _Accum},
1223 @code{_Sat unsigned _Accum},
1224 @code{_Sat unsigned long _Accum},
1225 @code{_Sat unsigned long long _Accum}.
1227 Fixed-point data values contain fractional and optional integral parts.
1228 The format of fixed-point data varies and depends on the target machine.
1230 Support for fixed-point types includes:
1233 prefix and postfix increment and decrement operators (@code{++}, @code{--})
1235 unary arithmetic operators (@code{+}, @code{-}, @code{!})
1237 binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/})
1239 binary shift operators (@code{<<}, @code{>>})
1241 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>})
1243 equality operators (@code{==}, @code{!=})
1245 assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=},
1246 @code{<<=}, @code{>>=})
1248 conversions to and from integer, floating-point, or fixed-point types
1251 Use a suffix in a fixed-point literal constant:
1253 @item @samp{hr} or @samp{HR} for @code{short _Fract} and
1254 @code{_Sat short _Fract}
1255 @item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract}
1256 @item @samp{lr} or @samp{LR} for @code{long _Fract} and
1257 @code{_Sat long _Fract}
1258 @item @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1259 @code{_Sat long long _Fract}
1260 @item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1261 @code{_Sat unsigned short _Fract}
1262 @item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1263 @code{_Sat unsigned _Fract}
1264 @item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1265 @code{_Sat unsigned long _Fract}
1266 @item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1267 and @code{_Sat unsigned long long _Fract}
1268 @item @samp{hk} or @samp{HK} for @code{short _Accum} and
1269 @code{_Sat short _Accum}
1270 @item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum}
1271 @item @samp{lk} or @samp{LK} for @code{long _Accum} and
1272 @code{_Sat long _Accum}
1273 @item @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1274 @code{_Sat long long _Accum}
1275 @item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1276 @code{_Sat unsigned short _Accum}
1277 @item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1278 @code{_Sat unsigned _Accum}
1279 @item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1280 @code{_Sat unsigned long _Accum}
1281 @item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1282 and @code{_Sat unsigned long long _Accum}
1285 GCC support of fixed-point types as specified by the draft technical report
1290 Pragmas to control overflow and rounding behaviors are not implemented.
1293 Fixed-point types are supported by the DWARF debug information format.
1295 @node Named Address Spaces
1296 @section Named Address Spaces
1297 @cindex Named Address Spaces
1299 As an extension, GNU C supports named address spaces as
1300 defined in the N1275 draft of ISO/IEC DTR 18037. Support for named
1301 address spaces in GCC will evolve as the draft technical report
1302 changes. Calling conventions for any target might also change. At
1303 present, only the AVR, SPU, M32C, RL78, and x86 targets support
1304 address spaces other than the generic address space.
1306 Address space identifiers may be used exactly like any other C type
1307 qualifier (e.g., @code{const} or @code{volatile}). See the N1275
1308 document for more details.
1310 @anchor{AVR Named Address Spaces}
1311 @subsection AVR Named Address Spaces
1313 On the AVR target, there are several address spaces that can be used
1314 in order to put read-only data into the flash memory and access that
1315 data by means of the special instructions @code{LPM} or @code{ELPM}
1316 needed to read from flash.
1318 Devices belonging to @code{avrtiny} and @code{avrxmega3} can access
1319 flash memory by means of @code{LD*} instructions because the flash
1320 memory is mapped into the RAM address space. There is @emph{no need}
1321 for language extensions like @code{__flash} or attribute
1322 @ref{AVR Variable Attributes,,@code{progmem}}.
1323 The default linker description files for these devices cater for that
1324 feature and @code{.rodata} stays in flash: The compiler just generates
1325 @code{LD*} instructions, and the linker script adds core specific
1326 offsets to all @code{.rodata} symbols: @code{0x4000} in the case of
1327 @code{avrtiny} and @code{0x8000} in the case of @code{avrxmega3}.
1328 See @ref{AVR Options} for a list of respective devices.
1330 For devices not in @code{avrtiny} or @code{avrxmega3},
1331 any data including read-only data is located in RAM (the generic
1332 address space) because flash memory is not visible in the RAM address
1333 space. In order to locate read-only data in flash memory @emph{and}
1334 to generate the right instructions to access this data without
1335 using (inline) assembler code, special address spaces are needed.
1339 @cindex @code{__flash} AVR Named Address Spaces
1340 The @code{__flash} qualifier locates data in the
1341 @code{.progmem.data} section. Data is read using the @code{LPM}
1342 instruction. Pointers to this address space are 16 bits wide.
1349 @cindex @code{__flash1} AVR Named Address Spaces
1350 @cindex @code{__flash2} AVR Named Address Spaces
1351 @cindex @code{__flash3} AVR Named Address Spaces
1352 @cindex @code{__flash4} AVR Named Address Spaces
1353 @cindex @code{__flash5} AVR Named Address Spaces
1354 These are 16-bit address spaces locating data in section
1355 @code{.progmem@var{N}.data} where @var{N} refers to
1356 address space @code{__flash@var{N}}.
1357 The compiler sets the @code{RAMPZ} segment register appropriately
1358 before reading data by means of the @code{ELPM} instruction.
1361 @cindex @code{__memx} AVR Named Address Spaces
1362 This is a 24-bit address space that linearizes flash and RAM:
1363 If the high bit of the address is set, data is read from
1364 RAM using the lower two bytes as RAM address.
1365 If the high bit of the address is clear, data is read from flash
1366 with @code{RAMPZ} set according to the high byte of the address.
1367 @xref{AVR Built-in Functions,,@code{__builtin_avr_flash_segment}}.
1369 Objects in this address space are located in @code{.progmemx.data}.
1375 char my_read (const __flash char ** p)
1377 /* p is a pointer to RAM that points to a pointer to flash.
1378 The first indirection of p reads that flash pointer
1379 from RAM and the second indirection reads a char from this
1385 /* Locate array[] in flash memory */
1386 const __flash int array[] = @{ 3, 5, 7, 11, 13, 17, 19 @};
1392 /* Return 17 by reading from flash memory */
1393 return array[array[i]];
1398 For each named address space supported by avr-gcc there is an equally
1399 named but uppercase built-in macro defined.
1400 The purpose is to facilitate testing if respective address space
1401 support is available or not:
1405 const __flash int var = 1;
1412 #include <avr/pgmspace.h> /* From AVR-LibC */
1414 const int var PROGMEM = 1;
1418 return (int) pgm_read_word (&var);
1420 #endif /* __FLASH */
1424 Notice that attribute @ref{AVR Variable Attributes,,@code{progmem}}
1425 locates data in flash but
1426 accesses to these data read from generic address space, i.e.@:
1428 so that you need special accessors like @code{pgm_read_byte}
1429 from @w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC}}
1430 together with attribute @code{progmem}.
1433 @b{Limitations and caveats}
1437 Reading across the 64@tie{}KiB section boundary of
1438 the @code{__flash} or @code{__flash@var{N}} address spaces
1439 shows undefined behavior. The only address space that
1440 supports reading across the 64@tie{}KiB flash segment boundaries is
1444 If you use one of the @code{__flash@var{N}} address spaces
1445 you must arrange your linker script to locate the
1446 @code{.progmem@var{N}.data} sections according to your needs.
1449 Any data or pointers to the non-generic address spaces must
1450 be qualified as @code{const}, i.e.@: as read-only data.
1451 This still applies if the data in one of these address
1452 spaces like software version number or calibration lookup table are intended to
1453 be changed after load time by, say, a boot loader. In this case
1454 the right qualification is @code{const} @code{volatile} so that the compiler
1455 must not optimize away known values or insert them
1456 as immediates into operands of instructions.
1459 The following code initializes a variable @code{pfoo}
1460 located in static storage with a 24-bit address:
1462 extern const __memx char foo;
1463 const __memx void *pfoo = &foo;
1467 On the reduced Tiny devices like ATtiny40, no address spaces are supported.
1468 Just use vanilla C / C++ code without overhead as outlined above.
1469 Attribute @code{progmem} is supported but works differently,
1470 see @ref{AVR Variable Attributes}.
1474 @subsection M32C Named Address Spaces
1475 @cindex @code{__far} M32C Named Address Spaces
1477 On the M32C target, with the R8C and M16C CPU variants, variables
1478 qualified with @code{__far} are accessed using 32-bit addresses in
1479 order to access memory beyond the first 64@tie{}Ki bytes. If
1480 @code{__far} is used with the M32CM or M32C CPU variants, it has no
1483 @subsection RL78 Named Address Spaces
1484 @cindex @code{__far} RL78 Named Address Spaces
1486 On the RL78 target, variables qualified with @code{__far} are accessed
1487 with 32-bit pointers (20-bit addresses) rather than the default 16-bit
1488 addresses. Non-far variables are assumed to appear in the topmost
1489 64@tie{}KiB of the address space.
1491 @subsection SPU Named Address Spaces
1492 @cindex @code{__ea} SPU Named Address Spaces
1494 On the SPU target variables may be declared as
1495 belonging to another address space by qualifying the type with the
1496 @code{__ea} address space identifier:
1503 The compiler generates special code to access the variable @code{i}.
1504 It may use runtime library
1505 support, or generate special machine instructions to access that address
1508 @subsection x86 Named Address Spaces
1509 @cindex x86 named address spaces
1511 On the x86 target, variables may be declared as being relative
1512 to the @code{%fs} or @code{%gs} segments.
1517 @cindex @code{__seg_fs} x86 named address space
1518 @cindex @code{__seg_gs} x86 named address space
1519 The object is accessed with the respective segment override prefix.
1521 The respective segment base must be set via some method specific to
1522 the operating system. Rather than require an expensive system call
1523 to retrieve the segment base, these address spaces are not considered
1524 to be subspaces of the generic (flat) address space. This means that
1525 explicit casts are required to convert pointers between these address
1526 spaces and the generic address space. In practice the application
1527 should cast to @code{uintptr_t} and apply the segment base offset
1528 that it installed previously.
1530 The preprocessor symbols @code{__SEG_FS} and @code{__SEG_GS} are
1531 defined when these address spaces are supported.
1535 @section Arrays of Length Zero
1536 @cindex arrays of length zero
1537 @cindex zero-length arrays
1538 @cindex length-zero arrays
1539 @cindex flexible array members
1541 Zero-length arrays are allowed in GNU C@. They are very useful as the
1542 last element of a structure that is really a header for a variable-length
1551 struct line *thisline = (struct line *)
1552 malloc (sizeof (struct line) + this_length);
1553 thisline->length = this_length;
1556 In ISO C90, you would have to give @code{contents} a length of 1, which
1557 means either you waste space or complicate the argument to @code{malloc}.
1559 In ISO C99, you would use a @dfn{flexible array member}, which is
1560 slightly different in syntax and semantics:
1564 Flexible array members are written as @code{contents[]} without
1568 Flexible array members have incomplete type, and so the @code{sizeof}
1569 operator may not be applied. As a quirk of the original implementation
1570 of zero-length arrays, @code{sizeof} evaluates to zero.
1573 Flexible array members may only appear as the last member of a
1574 @code{struct} that is otherwise non-empty.
1577 A structure containing a flexible array member, or a union containing
1578 such a structure (possibly recursively), may not be a member of a
1579 structure or an element of an array. (However, these uses are
1580 permitted by GCC as extensions.)
1583 Non-empty initialization of zero-length
1584 arrays is treated like any case where there are more initializer
1585 elements than the array holds, in that a suitable warning about ``excess
1586 elements in array'' is given, and the excess elements (all of them, in
1587 this case) are ignored.
1589 GCC allows static initialization of flexible array members.
1590 This is equivalent to defining a new structure containing the original
1591 structure followed by an array of sufficient size to contain the data.
1592 E.g.@: in the following, @code{f1} is constructed as if it were declared
1598 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1601 struct f1 f1; int data[3];
1602 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1606 The convenience of this extension is that @code{f1} has the desired
1607 type, eliminating the need to consistently refer to @code{f2.f1}.
1609 This has symmetry with normal static arrays, in that an array of
1610 unknown size is also written with @code{[]}.
1612 Of course, this extension only makes sense if the extra data comes at
1613 the end of a top-level object, as otherwise we would be overwriting
1614 data at subsequent offsets. To avoid undue complication and confusion
1615 with initialization of deeply nested arrays, we simply disallow any
1616 non-empty initialization except when the structure is the top-level
1617 object. For example:
1620 struct foo @{ int x; int y[]; @};
1621 struct bar @{ struct foo z; @};
1623 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1624 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1625 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1626 struct foo d[1] = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1629 @node Empty Structures
1630 @section Structures with No Members
1631 @cindex empty structures
1632 @cindex zero-size structures
1634 GCC permits a C structure to have no members:
1641 The structure has size zero. In C++, empty structures are part
1642 of the language. G++ treats empty structures as if they had a single
1643 member of type @code{char}.
1645 @node Variable Length
1646 @section Arrays of Variable Length
1647 @cindex variable-length arrays
1648 @cindex arrays of variable length
1651 Variable-length automatic arrays are allowed in ISO C99, and as an
1652 extension GCC accepts them in C90 mode and in C++. These arrays are
1653 declared like any other automatic arrays, but with a length that is not
1654 a constant expression. The storage is allocated at the point of
1655 declaration and deallocated when the block scope containing the declaration
1661 concat_fopen (char *s1, char *s2, char *mode)
1663 char str[strlen (s1) + strlen (s2) + 1];
1666 return fopen (str, mode);
1670 @cindex scope of a variable length array
1671 @cindex variable-length array scope
1672 @cindex deallocating variable length arrays
1673 Jumping or breaking out of the scope of the array name deallocates the
1674 storage. Jumping into the scope is not allowed; you get an error
1677 @cindex variable-length array in a structure
1678 As an extension, GCC accepts variable-length arrays as a member of
1679 a structure or a union. For example:
1685 struct S @{ int x[n]; @};
1689 @cindex @code{alloca} vs variable-length arrays
1690 You can use the function @code{alloca} to get an effect much like
1691 variable-length arrays. The function @code{alloca} is available in
1692 many other C implementations (but not in all). On the other hand,
1693 variable-length arrays are more elegant.
1695 There are other differences between these two methods. Space allocated
1696 with @code{alloca} exists until the containing @emph{function} returns.
1697 The space for a variable-length array is deallocated as soon as the array
1698 name's scope ends, unless you also use @code{alloca} in this scope.
1700 You can also use variable-length arrays as arguments to functions:
1704 tester (int len, char data[len][len])
1710 The length of an array is computed once when the storage is allocated
1711 and is remembered for the scope of the array in case you access it with
1714 If you want to pass the array first and the length afterward, you can
1715 use a forward declaration in the parameter list---another GNU extension.
1719 tester (int len; char data[len][len], int len)
1725 @cindex parameter forward declaration
1726 The @samp{int len} before the semicolon is a @dfn{parameter forward
1727 declaration}, and it serves the purpose of making the name @code{len}
1728 known when the declaration of @code{data} is parsed.
1730 You can write any number of such parameter forward declarations in the
1731 parameter list. They can be separated by commas or semicolons, but the
1732 last one must end with a semicolon, which is followed by the ``real''
1733 parameter declarations. Each forward declaration must match a ``real''
1734 declaration in parameter name and data type. ISO C99 does not support
1735 parameter forward declarations.
1737 @node Variadic Macros
1738 @section Macros with a Variable Number of Arguments.
1739 @cindex variable number of arguments
1740 @cindex macro with variable arguments
1741 @cindex rest argument (in macro)
1742 @cindex variadic macros
1744 In the ISO C standard of 1999, a macro can be declared to accept a
1745 variable number of arguments much as a function can. The syntax for
1746 defining the macro is similar to that of a function. Here is an
1750 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1754 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1755 such a macro, it represents the zero or more tokens until the closing
1756 parenthesis that ends the invocation, including any commas. This set of
1757 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1758 wherever it appears. See the CPP manual for more information.
1760 GCC has long supported variadic macros, and used a different syntax that
1761 allowed you to give a name to the variable arguments just like any other
1762 argument. Here is an example:
1765 #define debug(format, args...) fprintf (stderr, format, args)
1769 This is in all ways equivalent to the ISO C example above, but arguably
1770 more readable and descriptive.
1772 GNU CPP has two further variadic macro extensions, and permits them to
1773 be used with either of the above forms of macro definition.
1775 In standard C, you are not allowed to leave the variable argument out
1776 entirely; but you are allowed to pass an empty argument. For example,
1777 this invocation is invalid in ISO C, because there is no comma after
1784 GNU CPP permits you to completely omit the variable arguments in this
1785 way. In the above examples, the compiler would complain, though since
1786 the expansion of the macro still has the extra comma after the format
1789 To help solve this problem, CPP behaves specially for variable arguments
1790 used with the token paste operator, @samp{##}. If instead you write
1793 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1797 and if the variable arguments are omitted or empty, the @samp{##}
1798 operator causes the preprocessor to remove the comma before it. If you
1799 do provide some variable arguments in your macro invocation, GNU CPP
1800 does not complain about the paste operation and instead places the
1801 variable arguments after the comma. Just like any other pasted macro
1802 argument, these arguments are not macro expanded.
1804 @node Escaped Newlines
1805 @section Slightly Looser Rules for Escaped Newlines
1806 @cindex escaped newlines
1807 @cindex newlines (escaped)
1809 The preprocessor treatment of escaped newlines is more relaxed
1810 than that specified by the C90 standard, which requires the newline
1811 to immediately follow a backslash.
1812 GCC's implementation allows whitespace in the form
1813 of spaces, horizontal and vertical tabs, and form feeds between the
1814 backslash and the subsequent newline. The preprocessor issues a
1815 warning, but treats it as a valid escaped newline and combines the two
1816 lines to form a single logical line. This works within comments and
1817 tokens, as well as between tokens. Comments are @emph{not} treated as
1818 whitespace for the purposes of this relaxation, since they have not
1819 yet been replaced with spaces.
1822 @section Non-Lvalue Arrays May Have Subscripts
1823 @cindex subscripting
1824 @cindex arrays, non-lvalue
1826 @cindex subscripting and function values
1827 In ISO C99, arrays that are not lvalues still decay to pointers, and
1828 may be subscripted, although they may not be modified or used after
1829 the next sequence point and the unary @samp{&} operator may not be
1830 applied to them. As an extension, GNU C allows such arrays to be
1831 subscripted in C90 mode, though otherwise they do not decay to
1832 pointers outside C99 mode. For example,
1833 this is valid in GNU C though not valid in C90:
1837 struct foo @{int a[4];@};
1843 return f().a[index];
1849 @section Arithmetic on @code{void}- and Function-Pointers
1850 @cindex void pointers, arithmetic
1851 @cindex void, size of pointer to
1852 @cindex function pointers, arithmetic
1853 @cindex function, size of pointer to
1855 In GNU C, addition and subtraction operations are supported on pointers to
1856 @code{void} and on pointers to functions. This is done by treating the
1857 size of a @code{void} or of a function as 1.
1859 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1860 and on function types, and returns 1.
1862 @opindex Wpointer-arith
1863 The option @option{-Wpointer-arith} requests a warning if these extensions
1866 @node Pointers to Arrays
1867 @section Pointers to Arrays with Qualifiers Work as Expected
1868 @cindex pointers to arrays
1869 @cindex const qualifier
1871 In GNU C, pointers to arrays with qualifiers work similar to pointers
1872 to other qualified types. For example, a value of type @code{int (*)[5]}
1873 can be used to initialize a variable of type @code{const int (*)[5]}.
1874 These types are incompatible in ISO C because the @code{const} qualifier
1875 is formally attached to the element type of the array and not the
1880 transpose (int N, int M, double out[M][N], const double in[N][M]);
1884 transpose(3, 2, y, x);
1888 @section Non-Constant Initializers
1889 @cindex initializers, non-constant
1890 @cindex non-constant initializers
1892 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1893 automatic variable are not required to be constant expressions in GNU C@.
1894 Here is an example of an initializer with run-time varying elements:
1897 foo (float f, float g)
1899 float beat_freqs[2] = @{ f-g, f+g @};
1904 @node Compound Literals
1905 @section Compound Literals
1906 @cindex constructor expressions
1907 @cindex initializations in expressions
1908 @cindex structures, constructor expression
1909 @cindex expressions, constructor
1910 @cindex compound literals
1911 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1913 A compound literal looks like a cast of a brace-enclosed aggregate
1914 initializer list. Its value is an object of the type specified in
1915 the cast, containing the elements specified in the initializer.
1916 Unlike the result of a cast, a compound literal is an lvalue. ISO
1917 C99 and later support compound literals. As an extension, GCC
1918 supports compound literals also in C90 mode and in C++, although
1919 as explained below, the C++ semantics are somewhat different.
1921 Usually, the specified type of a compound literal is a structure. Assume
1922 that @code{struct foo} and @code{structure} are declared as shown:
1925 struct foo @{int a; char b[2];@} structure;
1929 Here is an example of constructing a @code{struct foo} with a compound literal:
1932 structure = ((struct foo) @{x + y, 'a', 0@});
1936 This is equivalent to writing the following:
1940 struct foo temp = @{x + y, 'a', 0@};
1945 You can also construct an array, though this is dangerous in C++, as
1946 explained below. If all the elements of the compound literal are
1947 (made up of) simple constant expressions suitable for use in
1948 initializers of objects of static storage duration, then the compound
1949 literal can be coerced to a pointer to its first element and used in
1950 such an initializer, as shown here:
1953 char **foo = (char *[]) @{ "x", "y", "z" @};
1956 Compound literals for scalar types and union types are also allowed. In
1957 the following example the variable @code{i} is initialized to the value
1958 @code{2}, the result of incrementing the unnamed object created by
1959 the compound literal.
1962 int i = ++(int) @{ 1 @};
1965 As a GNU extension, GCC allows initialization of objects with static storage
1966 duration by compound literals (which is not possible in ISO C99 because
1967 the initializer is not a constant).
1968 It is handled as if the object were initialized only with the brace-enclosed
1969 list if the types of the compound literal and the object match.
1970 The elements of the compound literal must be constant.
1971 If the object being initialized has array type of unknown size, the size is
1972 determined by the size of the compound literal.
1975 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1976 static int y[] = (int []) @{1, 2, 3@};
1977 static int z[] = (int [3]) @{1@};
1981 The above lines are equivalent to the following:
1983 static struct foo x = @{1, 'a', 'b'@};
1984 static int y[] = @{1, 2, 3@};
1985 static int z[] = @{1, 0, 0@};
1988 In C, a compound literal designates an unnamed object with static or
1989 automatic storage duration. In C++, a compound literal designates a
1990 temporary object that only lives until the end of its full-expression.
1991 As a result, well-defined C code that takes the address of a subobject
1992 of a compound literal can be undefined in C++, so G++ rejects
1993 the conversion of a temporary array to a pointer. For instance, if
1994 the array compound literal example above appeared inside a function,
1995 any subsequent use of @code{foo} in C++ would have undefined behavior
1996 because the lifetime of the array ends after the declaration of @code{foo}.
1998 As an optimization, G++ sometimes gives array compound literals longer
1999 lifetimes: when the array either appears outside a function or has
2000 a @code{const}-qualified type. If @code{foo} and its initializer had
2001 elements of type @code{char *const} rather than @code{char *}, or if
2002 @code{foo} were a global variable, the array would have static storage
2003 duration. But it is probably safest just to avoid the use of array
2004 compound literals in C++ code.
2006 @node Designated Inits
2007 @section Designated Initializers
2008 @cindex initializers with labeled elements
2009 @cindex labeled elements in initializers
2010 @cindex case labels in initializers
2011 @cindex designated initializers
2013 Standard C90 requires the elements of an initializer to appear in a fixed
2014 order, the same as the order of the elements in the array or structure
2017 In ISO C99 you can give the elements in any order, specifying the array
2018 indices or structure field names they apply to, and GNU C allows this as
2019 an extension in C90 mode as well. This extension is not
2020 implemented in GNU C++.
2022 To specify an array index, write
2023 @samp{[@var{index}] =} before the element value. For example,
2026 int a[6] = @{ [4] = 29, [2] = 15 @};
2033 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
2037 The index values must be constant expressions, even if the array being
2038 initialized is automatic.
2040 An alternative syntax for this that has been obsolete since GCC 2.5 but
2041 GCC still accepts is to write @samp{[@var{index}]} before the element
2042 value, with no @samp{=}.
2044 To initialize a range of elements to the same value, write
2045 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
2046 extension. For example,
2049 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
2053 If the value in it has side effects, the side effects happen only once,
2054 not for each initialized field by the range initializer.
2057 Note that the length of the array is the highest value specified
2060 In a structure initializer, specify the name of a field to initialize
2061 with @samp{.@var{fieldname} =} before the element value. For example,
2062 given the following structure,
2065 struct point @{ int x, y; @};
2069 the following initialization
2072 struct point p = @{ .y = yvalue, .x = xvalue @};
2079 struct point p = @{ xvalue, yvalue @};
2082 Another syntax that has the same meaning, obsolete since GCC 2.5, is
2083 @samp{@var{fieldname}:}, as shown here:
2086 struct point p = @{ y: yvalue, x: xvalue @};
2089 Omitted field members are implicitly initialized the same as objects
2090 that have static storage duration.
2093 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
2094 @dfn{designator}. You can also use a designator (or the obsolete colon
2095 syntax) when initializing a union, to specify which element of the union
2096 should be used. For example,
2099 union foo @{ int i; double d; @};
2101 union foo f = @{ .d = 4 @};
2105 converts 4 to a @code{double} to store it in the union using
2106 the second element. By contrast, casting 4 to type @code{union foo}
2107 stores it into the union as the integer @code{i}, since it is
2108 an integer. @xref{Cast to Union}.
2110 You can combine this technique of naming elements with ordinary C
2111 initialization of successive elements. Each initializer element that
2112 does not have a designator applies to the next consecutive element of the
2113 array or structure. For example,
2116 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
2123 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
2126 Labeling the elements of an array initializer is especially useful
2127 when the indices are characters or belong to an @code{enum} type.
2132 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
2133 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
2136 @cindex designator lists
2137 You can also write a series of @samp{.@var{fieldname}} and
2138 @samp{[@var{index}]} designators before an @samp{=} to specify a
2139 nested subobject to initialize; the list is taken relative to the
2140 subobject corresponding to the closest surrounding brace pair. For
2141 example, with the @samp{struct point} declaration above:
2144 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
2148 If the same field is initialized multiple times, it has the value from
2149 the last initialization. If any such overridden initialization has
2150 side effect, it is unspecified whether the side effect happens or not.
2151 Currently, GCC discards them and issues a warning.
2154 @section Case Ranges
2156 @cindex ranges in case statements
2158 You can specify a range of consecutive values in a single @code{case} label,
2162 case @var{low} ... @var{high}:
2166 This has the same effect as the proper number of individual @code{case}
2167 labels, one for each integer value from @var{low} to @var{high}, inclusive.
2169 This feature is especially useful for ranges of ASCII character codes:
2175 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
2176 it may be parsed wrong when you use it with integer values. For example,
2191 @section Cast to a Union Type
2192 @cindex cast to a union
2193 @cindex union, casting to a
2195 A cast to union type looks similar to other casts, except that the type
2196 specified is a union type. You can specify the type either with the
2197 @code{union} keyword or with a @code{typedef} name that refers to
2198 a union. A cast to a union actually creates a compound literal and
2199 yields an lvalue, not an rvalue like true casts do.
2200 @xref{Compound Literals}.
2202 The types that may be cast to the union type are those of the members
2203 of the union. Thus, given the following union and variables:
2206 union foo @{ int i; double d; @};
2212 both @code{x} and @code{y} can be cast to type @code{union foo}.
2214 Using the cast as the right-hand side of an assignment to a variable of
2215 union type is equivalent to storing in a member of the union:
2220 u = (union foo) x @equiv{} u.i = x
2221 u = (union foo) y @equiv{} u.d = y
2224 You can also use the union cast as a function argument:
2227 void hack (union foo);
2229 hack ((union foo) x);
2232 @node Mixed Declarations
2233 @section Mixed Declarations and Code
2234 @cindex mixed declarations and code
2235 @cindex declarations, mixed with code
2236 @cindex code, mixed with declarations
2238 ISO C99 and ISO C++ allow declarations and code to be freely mixed
2239 within compound statements. As an extension, GNU C also allows this in
2240 C90 mode. For example, you could do:
2249 Each identifier is visible from where it is declared until the end of
2250 the enclosing block.
2252 @node Function Attributes
2253 @section Declaring Attributes of Functions
2254 @cindex function attributes
2255 @cindex declaring attributes of functions
2256 @cindex @code{volatile} applied to function
2257 @cindex @code{const} applied to function
2259 In GNU C, you can use function attributes to declare certain things
2260 about functions called in your program which help the compiler
2261 optimize calls and check your code more carefully. For example, you
2262 can use attributes to declare that a function never returns
2263 (@code{noreturn}), returns a value depending only on its arguments
2264 (@code{pure}), or has @code{printf}-style arguments (@code{format}).
2266 You can also use attributes to control memory placement, code
2267 generation options or call/return conventions within the function
2268 being annotated. Many of these attributes are target-specific. For
2269 example, many targets support attributes for defining interrupt
2270 handler functions, which typically must follow special register usage
2271 and return conventions.
2273 Function attributes are introduced by the @code{__attribute__} keyword
2274 on a declaration, followed by an attribute specification inside double
2275 parentheses. You can specify multiple attributes in a declaration by
2276 separating them by commas within the double parentheses or by
2277 immediately following an attribute declaration with another attribute
2278 declaration. @xref{Attribute Syntax}, for the exact rules on attribute
2279 syntax and placement. Compatible attribute specifications on distinct
2280 declarations of the same function are merged. An attribute specification
2281 that is not compatible with attributes already applied to a declaration
2282 of the same function is ignored with a warning.
2284 GCC also supports attributes on
2285 variable declarations (@pxref{Variable Attributes}),
2286 labels (@pxref{Label Attributes}),
2287 enumerators (@pxref{Enumerator Attributes}),
2288 statements (@pxref{Statement Attributes}),
2289 and types (@pxref{Type Attributes}).
2291 There is some overlap between the purposes of attributes and pragmas
2292 (@pxref{Pragmas,,Pragmas Accepted by GCC}). It has been
2293 found convenient to use @code{__attribute__} to achieve a natural
2294 attachment of attributes to their corresponding declarations, whereas
2295 @code{#pragma} is of use for compatibility with other compilers
2296 or constructs that do not naturally form part of the grammar.
2298 In addition to the attributes documented here,
2299 GCC plugins may provide their own attributes.
2302 * Common Function Attributes::
2303 * AArch64 Function Attributes::
2304 * ARC Function Attributes::
2305 * ARM Function Attributes::
2306 * AVR Function Attributes::
2307 * Blackfin Function Attributes::
2308 * CR16 Function Attributes::
2309 * Epiphany Function Attributes::
2310 * H8/300 Function Attributes::
2311 * IA-64 Function Attributes::
2312 * M32C Function Attributes::
2313 * M32R/D Function Attributes::
2314 * m68k Function Attributes::
2315 * MCORE Function Attributes::
2316 * MeP Function Attributes::
2317 * MicroBlaze Function Attributes::
2318 * Microsoft Windows Function Attributes::
2319 * MIPS Function Attributes::
2320 * MSP430 Function Attributes::
2321 * NDS32 Function Attributes::
2322 * Nios II Function Attributes::
2323 * Nvidia PTX Function Attributes::
2324 * PowerPC Function Attributes::
2325 * RISC-V Function Attributes::
2326 * RL78 Function Attributes::
2327 * RX Function Attributes::
2328 * S/390 Function Attributes::
2329 * SH Function Attributes::
2330 * SPU Function Attributes::
2331 * Symbian OS Function Attributes::
2332 * V850 Function Attributes::
2333 * Visium Function Attributes::
2334 * x86 Function Attributes::
2335 * Xstormy16 Function Attributes::
2338 @node Common Function Attributes
2339 @subsection Common Function Attributes
2341 The following attributes are supported on most targets.
2344 @c Keep this table alphabetized by attribute name. Treat _ as space.
2346 @item alias ("@var{target}")
2347 @cindex @code{alias} function attribute
2348 The @code{alias} attribute causes the declaration to be emitted as an
2349 alias for another symbol, which must be specified. For instance,
2352 void __f () @{ /* @r{Do something.} */; @}
2353 void f () __attribute__ ((weak, alias ("__f")));
2357 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
2358 mangled name for the target must be used. It is an error if @samp{__f}
2359 is not defined in the same translation unit.
2361 This attribute requires assembler and object file support,
2362 and may not be available on all targets.
2364 @item aligned (@var{alignment})
2365 @cindex @code{aligned} function attribute
2366 This attribute specifies a minimum alignment for the function,
2369 You cannot use this attribute to decrease the alignment of a function,
2370 only to increase it. However, when you explicitly specify a function
2371 alignment this overrides the effect of the
2372 @option{-falign-functions} (@pxref{Optimize Options}) option for this
2375 Note that the effectiveness of @code{aligned} attributes may be
2376 limited by inherent limitations in your linker. On many systems, the
2377 linker is only able to arrange for functions to be aligned up to a
2378 certain maximum alignment. (For some linkers, the maximum supported
2379 alignment may be very very small.) See your linker documentation for
2380 further information.
2382 The @code{aligned} attribute can also be used for variables and fields
2383 (@pxref{Variable Attributes}.)
2386 @cindex @code{alloc_align} function attribute
2387 The @code{alloc_align} attribute is used to tell the compiler that the
2388 function return value points to memory, where the returned pointer minimum
2389 alignment is given by one of the functions parameters. GCC uses this
2390 information to improve pointer alignment analysis.
2392 The function parameter denoting the allocated alignment is specified by
2393 one integer argument, whose number is the argument of the attribute.
2394 Argument numbering starts at one.
2399 void* my_memalign(size_t, size_t) __attribute__((alloc_align(1)))
2403 declares that @code{my_memalign} returns memory with minimum alignment
2404 given by parameter 1.
2407 @cindex @code{alloc_size} function attribute
2408 The @code{alloc_size} attribute is used to tell the compiler that the
2409 function return value points to memory, where the size is given by
2410 one or two of the functions parameters. GCC uses this
2411 information to improve the correctness of @code{__builtin_object_size}.
2413 The function parameter(s) denoting the allocated size are specified by
2414 one or two integer arguments supplied to the attribute. The allocated size
2415 is either the value of the single function argument specified or the product
2416 of the two function arguments specified. Argument numbering starts at
2422 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
2423 void* my_realloc(void*, size_t) __attribute__((alloc_size(2)))
2427 declares that @code{my_calloc} returns memory of the size given by
2428 the product of parameter 1 and 2 and that @code{my_realloc} returns memory
2429 of the size given by parameter 2.
2432 @cindex @code{always_inline} function attribute
2433 Generally, functions are not inlined unless optimization is specified.
2434 For functions declared inline, this attribute inlines the function
2435 independent of any restrictions that otherwise apply to inlining.
2436 Failure to inline such a function is diagnosed as an error.
2437 Note that if such a function is called indirectly the compiler may
2438 or may not inline it depending on optimization level and a failure
2439 to inline an indirect call may or may not be diagnosed.
2442 @cindex @code{artificial} function attribute
2443 This attribute is useful for small inline wrappers that if possible
2444 should appear during debugging as a unit. Depending on the debug
2445 info format it either means marking the function as artificial
2446 or using the caller location for all instructions within the inlined
2449 @item assume_aligned
2450 @cindex @code{assume_aligned} function attribute
2451 The @code{assume_aligned} attribute is used to tell the compiler that the
2452 function return value points to memory, where the returned pointer minimum
2453 alignment is given by the first argument.
2454 If the attribute has two arguments, the second argument is misalignment offset.
2459 void* my_alloc1(size_t) __attribute__((assume_aligned(16)))
2460 void* my_alloc2(size_t) __attribute__((assume_aligned(32, 8)))
2464 declares that @code{my_alloc1} returns 16-byte aligned pointer and
2465 that @code{my_alloc2} returns a pointer whose value modulo 32 is equal
2468 @item bnd_instrument
2469 @cindex @code{bnd_instrument} function attribute
2470 The @code{bnd_instrument} attribute on functions is used to inform the
2471 compiler that the function should be instrumented when compiled
2472 with the @option{-fchkp-instrument-marked-only} option.
2475 @cindex @code{bnd_legacy} function attribute
2476 @cindex Pointer Bounds Checker attributes
2477 The @code{bnd_legacy} attribute on functions is used to inform the
2478 compiler that the function should not be instrumented when compiled
2479 with the @option{-fcheck-pointer-bounds} option.
2482 @cindex @code{cold} function attribute
2483 The @code{cold} attribute on functions is used to inform the compiler that
2484 the function is unlikely to be executed. The function is optimized for
2485 size rather than speed and on many targets it is placed into a special
2486 subsection of the text section so all cold functions appear close together,
2487 improving code locality of non-cold parts of program. The paths leading
2488 to calls of cold functions within code are marked as unlikely by the branch
2489 prediction mechanism. It is thus useful to mark functions used to handle
2490 unlikely conditions, such as @code{perror}, as cold to improve optimization
2491 of hot functions that do call marked functions in rare occasions.
2493 When profile feedback is available, via @option{-fprofile-use}, cold functions
2494 are automatically detected and this attribute is ignored.
2497 @cindex @code{const} function attribute
2498 @cindex functions that have no side effects
2499 Many functions do not examine any values except their arguments, and
2500 have no effects except to return a value. Calls to such functions lend
2501 themselves to optimization such as common subexpression elimination.
2502 The @code{const} attribute imposes greater restrictions on a function's
2503 definition than the similar @code{pure} attribute below because it prohibits
2504 the function from reading global variables. Consequently, the presence of
2505 the attribute on a function declaration allows GCC to emit more efficient
2506 code for some calls to the function. Decorating the same function with
2507 both the @code{const} and the @code{pure} attribute is diagnosed.
2509 @cindex pointer arguments
2510 Note that a function that has pointer arguments and examines the data
2511 pointed to must @emph{not} be declared @code{const}. Likewise, a
2512 function that calls a non-@code{const} function usually must not be
2513 @code{const}. Because a @code{const} function cannot have any side
2514 effects it does not make sense for such a function to return @code{void}.
2515 Declaring such a function is diagnosed.
2519 @itemx constructor (@var{priority})
2520 @itemx destructor (@var{priority})
2521 @cindex @code{constructor} function attribute
2522 @cindex @code{destructor} function attribute
2523 The @code{constructor} attribute causes the function to be called
2524 automatically before execution enters @code{main ()}. Similarly, the
2525 @code{destructor} attribute causes the function to be called
2526 automatically after @code{main ()} completes or @code{exit ()} is
2527 called. Functions with these attributes are useful for
2528 initializing data that is used implicitly during the execution of
2531 You may provide an optional integer priority to control the order in
2532 which constructor and destructor functions are run. A constructor
2533 with a smaller priority number runs before a constructor with a larger
2534 priority number; the opposite relationship holds for destructors. So,
2535 if you have a constructor that allocates a resource and a destructor
2536 that deallocates the same resource, both functions typically have the
2537 same priority. The priorities for constructor and destructor
2538 functions are the same as those specified for namespace-scope C++
2539 objects (@pxref{C++ Attributes}). However, at present, the order in which
2540 constructors for C++ objects with static storage duration and functions
2541 decorated with attribute @code{constructor} are invoked is unspecified.
2542 In mixed declarations, attribute @code{init_priority} can be used to
2543 impose a specific ordering.
2546 @itemx deprecated (@var{msg})
2547 @cindex @code{deprecated} function attribute
2548 The @code{deprecated} attribute results in a warning if the function
2549 is used anywhere in the source file. This is useful when identifying
2550 functions that are expected to be removed in a future version of a
2551 program. The warning also includes the location of the declaration
2552 of the deprecated function, to enable users to easily find further
2553 information about why the function is deprecated, or what they should
2554 do instead. Note that the warnings only occurs for uses:
2557 int old_fn () __attribute__ ((deprecated));
2559 int (*fn_ptr)() = old_fn;
2563 results in a warning on line 3 but not line 2. The optional @var{msg}
2564 argument, which must be a string, is printed in the warning if
2567 The @code{deprecated} attribute can also be used for variables and
2568 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2570 @item error ("@var{message}")
2571 @itemx warning ("@var{message}")
2572 @cindex @code{error} function attribute
2573 @cindex @code{warning} function attribute
2574 If the @code{error} or @code{warning} attribute
2575 is used on a function declaration and a call to such a function
2576 is not eliminated through dead code elimination or other optimizations,
2577 an error or warning (respectively) that includes @var{message} is diagnosed.
2579 for compile-time checking, especially together with @code{__builtin_constant_p}
2580 and inline functions where checking the inline function arguments is not
2581 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
2583 While it is possible to leave the function undefined and thus invoke
2584 a link failure (to define the function with
2585 a message in @code{.gnu.warning*} section),
2586 when using these attributes the problem is diagnosed
2587 earlier and with exact location of the call even in presence of inline
2588 functions or when not emitting debugging information.
2590 @item externally_visible
2591 @cindex @code{externally_visible} function attribute
2592 This attribute, attached to a global variable or function, nullifies
2593 the effect of the @option{-fwhole-program} command-line option, so the
2594 object remains visible outside the current compilation unit.
2596 If @option{-fwhole-program} is used together with @option{-flto} and
2597 @command{gold} is used as the linker plugin,
2598 @code{externally_visible} attributes are automatically added to functions
2599 (not variable yet due to a current @command{gold} issue)
2600 that are accessed outside of LTO objects according to resolution file
2601 produced by @command{gold}.
2602 For other linkers that cannot generate resolution file,
2603 explicit @code{externally_visible} attributes are still necessary.
2606 @cindex @code{flatten} function attribute
2607 Generally, inlining into a function is limited. For a function marked with
2608 this attribute, every call inside this function is inlined, if possible.
2609 Whether the function itself is considered for inlining depends on its size and
2610 the current inlining parameters.
2612 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2613 @cindex @code{format} function attribute
2614 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
2616 The @code{format} attribute specifies that a function takes @code{printf},
2617 @code{scanf}, @code{strftime} or @code{strfmon} style arguments that
2618 should be type-checked against a format string. For example, the
2623 my_printf (void *my_object, const char *my_format, ...)
2624 __attribute__ ((format (printf, 2, 3)));
2628 causes the compiler to check the arguments in calls to @code{my_printf}
2629 for consistency with the @code{printf} style format string argument
2632 The parameter @var{archetype} determines how the format string is
2633 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2634 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2635 @code{strfmon}. (You can also use @code{__printf__},
2636 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2637 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2638 @code{ms_strftime} are also present.
2639 @var{archetype} values such as @code{printf} refer to the formats accepted
2640 by the system's C runtime library,
2641 while values prefixed with @samp{gnu_} always refer
2642 to the formats accepted by the GNU C Library. On Microsoft Windows
2643 targets, values prefixed with @samp{ms_} refer to the formats accepted by the
2644 @file{msvcrt.dll} library.
2645 The parameter @var{string-index}
2646 specifies which argument is the format string argument (starting
2647 from 1), while @var{first-to-check} is the number of the first
2648 argument to check against the format string. For functions
2649 where the arguments are not available to be checked (such as
2650 @code{vprintf}), specify the third parameter as zero. In this case the
2651 compiler only checks the format string for consistency. For
2652 @code{strftime} formats, the third parameter is required to be zero.
2653 Since non-static C++ methods have an implicit @code{this} argument, the
2654 arguments of such methods should be counted from two, not one, when
2655 giving values for @var{string-index} and @var{first-to-check}.
2657 In the example above, the format string (@code{my_format}) is the second
2658 argument of the function @code{my_print}, and the arguments to check
2659 start with the third argument, so the correct parameters for the format
2660 attribute are 2 and 3.
2662 @opindex ffreestanding
2663 @opindex fno-builtin
2664 The @code{format} attribute allows you to identify your own functions
2665 that take format strings as arguments, so that GCC can check the
2666 calls to these functions for errors. The compiler always (unless
2667 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2668 for the standard library functions @code{printf}, @code{fprintf},
2669 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2670 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2671 warnings are requested (using @option{-Wformat}), so there is no need to
2672 modify the header file @file{stdio.h}. In C99 mode, the functions
2673 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2674 @code{vsscanf} are also checked. Except in strictly conforming C
2675 standard modes, the X/Open function @code{strfmon} is also checked as
2676 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2677 @xref{C Dialect Options,,Options Controlling C Dialect}.
2679 For Objective-C dialects, @code{NSString} (or @code{__NSString__}) is
2680 recognized in the same context. Declarations including these format attributes
2681 are parsed for correct syntax, however the result of checking of such format
2682 strings is not yet defined, and is not carried out by this version of the
2685 The target may also provide additional types of format checks.
2686 @xref{Target Format Checks,,Format Checks Specific to Particular
2689 @item format_arg (@var{string-index})
2690 @cindex @code{format_arg} function attribute
2691 @opindex Wformat-nonliteral
2692 The @code{format_arg} attribute specifies that a function takes a format
2693 string for a @code{printf}, @code{scanf}, @code{strftime} or
2694 @code{strfmon} style function and modifies it (for example, to translate
2695 it into another language), so the result can be passed to a
2696 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2697 function (with the remaining arguments to the format function the same
2698 as they would have been for the unmodified string). For example, the
2703 my_dgettext (char *my_domain, const char *my_format)
2704 __attribute__ ((format_arg (2)));
2708 causes the compiler to check the arguments in calls to a @code{printf},
2709 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2710 format string argument is a call to the @code{my_dgettext} function, for
2711 consistency with the format string argument @code{my_format}. If the
2712 @code{format_arg} attribute had not been specified, all the compiler
2713 could tell in such calls to format functions would be that the format
2714 string argument is not constant; this would generate a warning when
2715 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2716 without the attribute.
2718 The parameter @var{string-index} specifies which argument is the format
2719 string argument (starting from one). Since non-static C++ methods have
2720 an implicit @code{this} argument, the arguments of such methods should
2721 be counted from two.
2723 The @code{format_arg} attribute allows you to identify your own
2724 functions that modify format strings, so that GCC can check the
2725 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2726 type function whose operands are a call to one of your own function.
2727 The compiler always treats @code{gettext}, @code{dgettext}, and
2728 @code{dcgettext} in this manner except when strict ISO C support is
2729 requested by @option{-ansi} or an appropriate @option{-std} option, or
2730 @option{-ffreestanding} or @option{-fno-builtin}
2731 is used. @xref{C Dialect Options,,Options
2732 Controlling C Dialect}.
2734 For Objective-C dialects, the @code{format-arg} attribute may refer to an
2735 @code{NSString} reference for compatibility with the @code{format} attribute
2738 The target may also allow additional types in @code{format-arg} attributes.
2739 @xref{Target Format Checks,,Format Checks Specific to Particular
2743 @cindex @code{gnu_inline} function attribute
2744 This attribute should be used with a function that is also declared
2745 with the @code{inline} keyword. It directs GCC to treat the function
2746 as if it were defined in gnu90 mode even when compiling in C99 or
2749 If the function is declared @code{extern}, then this definition of the
2750 function is used only for inlining. In no case is the function
2751 compiled as a standalone function, not even if you take its address
2752 explicitly. Such an address becomes an external reference, as if you
2753 had only declared the function, and had not defined it. This has
2754 almost the effect of a macro. The way to use this is to put a
2755 function definition in a header file with this attribute, and put
2756 another copy of the function, without @code{extern}, in a library
2757 file. The definition in the header file causes most calls to the
2758 function to be inlined. If any uses of the function remain, they
2759 refer to the single copy in the library. Note that the two
2760 definitions of the functions need not be precisely the same, although
2761 if they do not have the same effect your program may behave oddly.
2763 In C, if the function is neither @code{extern} nor @code{static}, then
2764 the function is compiled as a standalone function, as well as being
2765 inlined where possible.
2767 This is how GCC traditionally handled functions declared
2768 @code{inline}. Since ISO C99 specifies a different semantics for
2769 @code{inline}, this function attribute is provided as a transition
2770 measure and as a useful feature in its own right. This attribute is
2771 available in GCC 4.1.3 and later. It is available if either of the
2772 preprocessor macros @code{__GNUC_GNU_INLINE__} or
2773 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
2774 Function is As Fast As a Macro}.
2776 In C++, this attribute does not depend on @code{extern} in any way,
2777 but it still requires the @code{inline} keyword to enable its special
2781 @cindex @code{hot} function attribute
2782 The @code{hot} attribute on a function is used to inform the compiler that
2783 the function is a hot spot of the compiled program. The function is
2784 optimized more aggressively and on many targets it is placed into a special
2785 subsection of the text section so all hot functions appear close together,
2788 When profile feedback is available, via @option{-fprofile-use}, hot functions
2789 are automatically detected and this attribute is ignored.
2791 @item ifunc ("@var{resolver}")
2792 @cindex @code{ifunc} function attribute
2793 @cindex indirect functions
2794 @cindex functions that are dynamically resolved
2795 The @code{ifunc} attribute is used to mark a function as an indirect
2796 function using the STT_GNU_IFUNC symbol type extension to the ELF
2797 standard. This allows the resolution of the symbol value to be
2798 determined dynamically at load time, and an optimized version of the
2799 routine to be selected for the particular processor or other system
2800 characteristics determined then. To use this attribute, first define
2801 the implementation functions available, and a resolver function that
2802 returns a pointer to the selected implementation function. The
2803 implementation functions' declarations must match the API of the
2804 function being implemented. The resolver should be declared to
2805 be a function taking no arguments and returning a pointer to
2806 a function of the same type as the implementation. For example:
2809 void *my_memcpy (void *dst, const void *src, size_t len)
2815 static void * (*resolve_memcpy (void))(void *, const void *, size_t)
2817 return my_memcpy; // we will just always select this routine
2822 The exported header file declaring the function the user calls would
2826 extern void *memcpy (void *, const void *, size_t);
2830 allowing the user to call @code{memcpy} as a regular function, unaware of
2831 the actual implementation. Finally, the indirect function needs to be
2832 defined in the same translation unit as the resolver function:
2835 void *memcpy (void *, const void *, size_t)
2836 __attribute__ ((ifunc ("resolve_memcpy")));
2839 In C++, the @code{ifunc} attribute takes a string that is the mangled name
2840 of the resolver function. A C++ resolver for a non-static member function
2841 of class @code{C} should be declared to return a pointer to a non-member
2842 function taking pointer to @code{C} as the first argument, followed by
2843 the same arguments as of the implementation function. G++ checks
2844 the signatures of the two functions and issues
2845 a @option{-Wattribute-alias} warning for mismatches. To suppress a warning
2846 for the necessary cast from a pointer to the implementation member function
2847 to the type of the corresponding non-member function use
2848 the @option{-Wno-pmf-conversions} option. For example:
2854 int debug_impl (int);
2855 int optimized_impl (int);
2857 typedef int Func (S*, int);
2859 static Func* resolver ();
2862 int interface (int);
2865 int S::debug_impl (int) @{ /* @r{@dots{}} */ @}
2866 int S::optimized_impl (int) @{ /* @r{@dots{}} */ @}
2868 S::Func* S::resolver ()
2870 int (S::*pimpl) (int)
2871 = getenv ("DEBUG") ? &S::debug_impl : &S::optimized_impl;
2873 // Cast triggers -Wno-pmf-conversions.
2874 return reinterpret_cast<Func*>(pimpl);
2877 int S::interface (int) __attribute__ ((ifunc ("_ZN1S8resolverEv")));
2880 Indirect functions cannot be weak. Binutils version 2.20.1 or higher
2881 and GNU C Library version 2.11.1 are required to use this feature.
2884 @itemx interrupt_handler
2885 Many GCC back ends support attributes to indicate that a function is
2886 an interrupt handler, which tells the compiler to generate function
2887 entry and exit sequences that differ from those from regular
2888 functions. The exact syntax and behavior are target-specific;
2889 refer to the following subsections for details.
2892 @cindex @code{leaf} function attribute
2893 Calls to external functions with this attribute must return to the
2894 current compilation unit only by return or by exception handling. In
2895 particular, a leaf function is not allowed to invoke callback functions
2896 passed to it from the current compilation unit, directly call functions
2897 exported by the unit, or @code{longjmp} into the unit. Leaf functions
2898 might still call functions from other compilation units and thus they
2899 are not necessarily leaf in the sense that they contain no function
2902 The attribute is intended for library functions to improve dataflow
2903 analysis. The compiler takes the hint that any data not escaping the
2904 current compilation unit cannot be used or modified by the leaf
2905 function. For example, the @code{sin} function is a leaf function, but
2906 @code{qsort} is not.
2908 Note that leaf functions might indirectly run a signal handler defined
2909 in the current compilation unit that uses static variables. Similarly,
2910 when lazy symbol resolution is in effect, leaf functions might invoke
2911 indirect functions whose resolver function or implementation function is
2912 defined in the current compilation unit and uses static variables. There
2913 is no standard-compliant way to write such a signal handler, resolver
2914 function, or implementation function, and the best that you can do is to
2915 remove the @code{leaf} attribute or mark all such static variables
2916 @code{volatile}. Lastly, for ELF-based systems that support symbol
2917 interposition, care should be taken that functions defined in the
2918 current compilation unit do not unexpectedly interpose other symbols
2919 based on the defined standards mode and defined feature test macros;
2920 otherwise an inadvertent callback would be added.
2922 The attribute has no effect on functions defined within the current
2923 compilation unit. This is to allow easy merging of multiple compilation
2924 units into one, for example, by using the link-time optimization. For
2925 this reason the attribute is not allowed on types to annotate indirect
2929 @cindex @code{malloc} function attribute
2930 @cindex functions that behave like malloc
2931 This tells the compiler that a function is @code{malloc}-like, i.e.,
2932 that the pointer @var{P} returned by the function cannot alias any
2933 other pointer valid when the function returns, and moreover no
2934 pointers to valid objects occur in any storage addressed by @var{P}.
2936 Using this attribute can improve optimization. Functions like
2937 @code{malloc} and @code{calloc} have this property because they return
2938 a pointer to uninitialized or zeroed-out storage. However, functions
2939 like @code{realloc} do not have this property, as they can return a
2940 pointer to storage containing pointers.
2943 @cindex @code{no_icf} function attribute
2944 This function attribute prevents a functions from being merged with another
2945 semantically equivalent function.
2947 @item no_instrument_function
2948 @cindex @code{no_instrument_function} function attribute
2949 @opindex finstrument-functions
2950 If @option{-finstrument-functions} is given, profiling function calls are
2951 generated at entry and exit of most user-compiled functions.
2952 Functions with this attribute are not so instrumented.
2954 @item no_profile_instrument_function
2955 @cindex @code{no_profile_instrument_function} function attribute
2956 The @code{no_profile_instrument_function} attribute on functions is used
2957 to inform the compiler that it should not process any profile feedback based
2958 optimization code instrumentation.
2961 @cindex @code{no_reorder} function attribute
2962 Do not reorder functions or variables marked @code{no_reorder}
2963 against each other or top level assembler statements the executable.
2964 The actual order in the program will depend on the linker command
2965 line. Static variables marked like this are also not removed.
2966 This has a similar effect
2967 as the @option{-fno-toplevel-reorder} option, but only applies to the
2970 @item no_sanitize ("@var{sanitize_option}")
2971 @cindex @code{no_sanitize} function attribute
2972 The @code{no_sanitize} attribute on functions is used
2973 to inform the compiler that it should not do sanitization of all options
2974 mentioned in @var{sanitize_option}. A list of values acceptable by
2975 @option{-fsanitize} option can be provided.
2978 void __attribute__ ((no_sanitize ("alignment", "object-size")))
2979 f () @{ /* @r{Do something.} */; @}
2980 void __attribute__ ((no_sanitize ("alignment,object-size")))
2981 g () @{ /* @r{Do something.} */; @}
2984 @item no_sanitize_address
2985 @itemx no_address_safety_analysis
2986 @cindex @code{no_sanitize_address} function attribute
2987 The @code{no_sanitize_address} attribute on functions is used
2988 to inform the compiler that it should not instrument memory accesses
2989 in the function when compiling with the @option{-fsanitize=address} option.
2990 The @code{no_address_safety_analysis} is a deprecated alias of the
2991 @code{no_sanitize_address} attribute, new code should use
2992 @code{no_sanitize_address}.
2994 @item no_sanitize_thread
2995 @cindex @code{no_sanitize_thread} function attribute
2996 The @code{no_sanitize_thread} attribute on functions is used
2997 to inform the compiler that it should not instrument memory accesses
2998 in the function when compiling with the @option{-fsanitize=thread} option.
3000 @item no_sanitize_undefined
3001 @cindex @code{no_sanitize_undefined} function attribute
3002 The @code{no_sanitize_undefined} attribute on functions is used
3003 to inform the compiler that it should not check for undefined behavior
3004 in the function when compiling with the @option{-fsanitize=undefined} option.
3006 @item no_split_stack
3007 @cindex @code{no_split_stack} function attribute
3008 @opindex fsplit-stack
3009 If @option{-fsplit-stack} is given, functions have a small
3010 prologue which decides whether to split the stack. Functions with the
3011 @code{no_split_stack} attribute do not have that prologue, and thus
3012 may run with only a small amount of stack space available.
3014 @item no_stack_limit
3015 @cindex @code{no_stack_limit} function attribute
3016 This attribute locally overrides the @option{-fstack-limit-register}
3017 and @option{-fstack-limit-symbol} command-line options; it has the effect
3018 of disabling stack limit checking in the function it applies to.
3021 @cindex @code{noclone} function attribute
3022 This function attribute prevents a function from being considered for
3023 cloning---a mechanism that produces specialized copies of functions
3024 and which is (currently) performed by interprocedural constant
3028 @cindex @code{noinline} function attribute
3029 This function attribute prevents a function from being considered for
3031 @c Don't enumerate the optimizations by name here; we try to be
3032 @c future-compatible with this mechanism.
3033 If the function does not have side effects, there are optimizations
3034 other than inlining that cause function calls to be optimized away,
3035 although the function call is live. To keep such calls from being
3042 (@pxref{Extended Asm}) in the called function, to serve as a special
3046 @cindex @code{noipa} function attribute
3047 Disable interprocedural optimizations between the function with this
3048 attribute and its callers, as if the body of the function is not available
3049 when optimizing callers and the callers are unavailable when optimizing
3050 the body. This attribute implies @code{noinline}, @code{noclone} and
3051 @code{no_icf} attributes. However, this attribute is not equivalent
3052 to a combination of other attributes, because its purpose is to suppress
3053 existing and future optimizations employing interprocedural analysis,
3054 including those that do not have an attribute suitable for disabling
3055 them individually. This attribute is supported mainly for the purpose
3056 of testing the compiler.
3058 @item nonnull (@var{arg-index}, @dots{})
3059 @cindex @code{nonnull} function attribute
3060 @cindex functions with non-null pointer arguments
3061 The @code{nonnull} attribute specifies that some function parameters should
3062 be non-null pointers. For instance, the declaration:
3066 my_memcpy (void *dest, const void *src, size_t len)
3067 __attribute__((nonnull (1, 2)));
3071 causes the compiler to check that, in calls to @code{my_memcpy},
3072 arguments @var{dest} and @var{src} are non-null. If the compiler
3073 determines that a null pointer is passed in an argument slot marked
3074 as non-null, and the @option{-Wnonnull} option is enabled, a warning
3075 is issued. The compiler may also choose to make optimizations based
3076 on the knowledge that certain function arguments will never be null.
3078 If no argument index list is given to the @code{nonnull} attribute,
3079 all pointer arguments are marked as non-null. To illustrate, the
3080 following declaration is equivalent to the previous example:
3084 my_memcpy (void *dest, const void *src, size_t len)
3085 __attribute__((nonnull));
3089 @cindex @code{noplt} function attribute
3090 The @code{noplt} attribute is the counterpart to option @option{-fno-plt}.
3091 Calls to functions marked with this attribute in position-independent code
3096 /* Externally defined function foo. */
3097 int foo () __attribute__ ((noplt));
3100 main (/* @r{@dots{}} */)
3109 The @code{noplt} attribute on function @code{foo}
3110 tells the compiler to assume that
3111 the function @code{foo} is externally defined and that the call to
3112 @code{foo} must avoid the PLT
3113 in position-independent code.
3115 In position-dependent code, a few targets also convert calls to
3116 functions that are marked to not use the PLT to use the GOT instead.
3119 @cindex @code{noreturn} function attribute
3120 @cindex functions that never return
3121 A few standard library functions, such as @code{abort} and @code{exit},
3122 cannot return. GCC knows this automatically. Some programs define
3123 their own functions that never return. You can declare them
3124 @code{noreturn} to tell the compiler this fact. For example,
3128 void fatal () __attribute__ ((noreturn));
3131 fatal (/* @r{@dots{}} */)
3133 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
3139 The @code{noreturn} keyword tells the compiler to assume that
3140 @code{fatal} cannot return. It can then optimize without regard to what
3141 would happen if @code{fatal} ever did return. This makes slightly
3142 better code. More importantly, it helps avoid spurious warnings of
3143 uninitialized variables.
3145 The @code{noreturn} keyword does not affect the exceptional path when that
3146 applies: a @code{noreturn}-marked function may still return to the caller
3147 by throwing an exception or calling @code{longjmp}.
3149 Do not assume that registers saved by the calling function are
3150 restored before calling the @code{noreturn} function.
3152 It does not make sense for a @code{noreturn} function to have a return
3153 type other than @code{void}.
3156 @cindex @code{nothrow} function attribute
3157 The @code{nothrow} attribute is used to inform the compiler that a
3158 function cannot throw an exception. For example, most functions in
3159 the standard C library can be guaranteed not to throw an exception
3160 with the notable exceptions of @code{qsort} and @code{bsearch} that
3161 take function pointer arguments.
3164 @cindex @code{optimize} function attribute
3165 The @code{optimize} attribute is used to specify that a function is to
3166 be compiled with different optimization options than specified on the
3167 command line. Arguments can either be numbers or strings. Numbers
3168 are assumed to be an optimization level. Strings that begin with
3169 @code{O} are assumed to be an optimization option, while other options
3170 are assumed to be used with a @code{-f} prefix. You can also use the
3171 @samp{#pragma GCC optimize} pragma to set the optimization options
3172 that affect more than one function.
3173 @xref{Function Specific Option Pragmas}, for details about the
3174 @samp{#pragma GCC optimize} pragma.
3176 This attribute should be used for debugging purposes only. It is not
3177 suitable in production code.
3179 @item patchable_function_entry
3180 @cindex @code{patchable_function_entry} function attribute
3181 @cindex extra NOP instructions at the function entry point
3182 In case the target's text segment can be made writable at run time by
3183 any means, padding the function entry with a number of NOPs can be
3184 used to provide a universal tool for instrumentation.
3186 The @code{patchable_function_entry} function attribute can be used to
3187 change the number of NOPs to any desired value. The two-value syntax
3188 is the same as for the command-line switch
3189 @option{-fpatchable-function-entry=N,M}, generating @var{N} NOPs, with
3190 the function entry point before the @var{M}th NOP instruction.
3191 @var{M} defaults to 0 if omitted e.g. function entry point is before
3194 If patchable function entries are enabled globally using the command-line
3195 option @option{-fpatchable-function-entry=N,M}, then you must disable
3196 instrumentation on all functions that are part of the instrumentation
3197 framework with the attribute @code{patchable_function_entry (0)}
3198 to prevent recursion.
3201 @cindex @code{pure} function attribute
3202 @cindex functions that have no side effects
3203 Many functions have no effects except the return value and their
3204 return value depends only on the parameters and/or global variables.
3205 Calls to such functions can be subject
3206 to common subexpression elimination and loop optimization just as an
3207 arithmetic operator would be. These functions should be declared
3208 with the attribute @code{pure}. For example,
3211 int square (int) __attribute__ ((pure));
3215 says that the hypothetical function @code{square} is safe to call
3216 fewer times than the program says.
3218 Some common examples of pure functions are @code{strlen} or @code{memcmp}.
3219 Interesting non-pure functions are functions with infinite loops or those
3220 depending on volatile memory or other system resource, that may change between
3221 two consecutive calls (such as @code{feof} in a multithreading environment).
3223 The @code{pure} attribute imposes similar but looser restrictions on
3224 a function's defintion than the @code{const} attribute: it allows the
3225 function to read global variables. Decorating the same function with
3226 both the @code{pure} and the @code{const} attribute is diagnosed.
3227 Because a @code{pure} function cannot have any side effects it does not
3228 make sense for such a function to return @code{void}. Declaring such
3229 a function is diagnosed.
3231 @item returns_nonnull
3232 @cindex @code{returns_nonnull} function attribute
3233 The @code{returns_nonnull} attribute specifies that the function
3234 return value should be a non-null pointer. For instance, the declaration:
3238 mymalloc (size_t len) __attribute__((returns_nonnull));
3242 lets the compiler optimize callers based on the knowledge
3243 that the return value will never be null.
3246 @cindex @code{returns_twice} function attribute
3247 @cindex functions that return more than once
3248 The @code{returns_twice} attribute tells the compiler that a function may
3249 return more than one time. The compiler ensures that all registers
3250 are dead before calling such a function and emits a warning about
3251 the variables that may be clobbered after the second return from the
3252 function. Examples of such functions are @code{setjmp} and @code{vfork}.
3253 The @code{longjmp}-like counterpart of such function, if any, might need
3254 to be marked with the @code{noreturn} attribute.
3256 @item section ("@var{section-name}")
3257 @cindex @code{section} function attribute
3258 @cindex functions in arbitrary sections
3259 Normally, the compiler places the code it generates in the @code{text} section.
3260 Sometimes, however, you need additional sections, or you need certain
3261 particular functions to appear in special sections. The @code{section}
3262 attribute specifies that a function lives in a particular section.
3263 For example, the declaration:
3266 extern void foobar (void) __attribute__ ((section ("bar")));
3270 puts the function @code{foobar} in the @code{bar} section.
3272 Some file formats do not support arbitrary sections so the @code{section}
3273 attribute is not available on all platforms.
3274 If you need to map the entire contents of a module to a particular
3275 section, consider using the facilities of the linker instead.
3278 @cindex @code{sentinel} function attribute
3279 This function attribute ensures that a parameter in a function call is
3280 an explicit @code{NULL}. The attribute is only valid on variadic
3281 functions. By default, the sentinel is located at position zero, the
3282 last parameter of the function call. If an optional integer position
3283 argument P is supplied to the attribute, the sentinel must be located at
3284 position P counting backwards from the end of the argument list.
3287 __attribute__ ((sentinel))
3289 __attribute__ ((sentinel(0)))
3292 The attribute is automatically set with a position of 0 for the built-in
3293 functions @code{execl} and @code{execlp}. The built-in function
3294 @code{execle} has the attribute set with a position of 1.
3296 A valid @code{NULL} in this context is defined as zero with any pointer
3297 type. If your system defines the @code{NULL} macro with an integer type
3298 then you need to add an explicit cast. GCC replaces @code{stddef.h}
3299 with a copy that redefines NULL appropriately.
3301 The warnings for missing or incorrect sentinels are enabled with
3305 @itemx simd("@var{mask}")
3306 @cindex @code{simd} function attribute
3307 This attribute enables creation of one or more function versions that
3308 can process multiple arguments using SIMD instructions from a
3309 single invocation. Specifying this attribute allows compiler to
3310 assume that such versions are available at link time (provided
3311 in the same or another translation unit). Generated versions are
3312 target-dependent and described in the corresponding Vector ABI document. For
3313 x86_64 target this document can be found
3314 @w{@uref{https://sourceware.org/glibc/wiki/libmvec?action=AttachFile&do=view&target=VectorABI.txt,here}}.
3316 The optional argument @var{mask} may have the value
3317 @code{notinbranch} or @code{inbranch},
3318 and instructs the compiler to generate non-masked or masked
3319 clones correspondingly. By default, all clones are generated.
3321 If the attribute is specified and @code{#pragma omp declare simd} is
3322 present on a declaration and the @option{-fopenmp} or @option{-fopenmp-simd}
3323 switch is specified, then the attribute is ignored.
3326 @cindex @code{stack_protect} function attribute
3327 This attribute adds stack protection code to the function if
3328 flags @option{-fstack-protector}, @option{-fstack-protector-strong}
3329 or @option{-fstack-protector-explicit} are set.
3331 @item target (@var{options})
3332 @cindex @code{target} function attribute
3333 Multiple target back ends implement the @code{target} attribute
3334 to specify that a function is to
3335 be compiled with different target options than specified on the
3336 command line. This can be used for instance to have functions
3337 compiled with a different ISA (instruction set architecture) than the
3338 default. You can also use the @samp{#pragma GCC target} pragma to set
3339 more than one function to be compiled with specific target options.
3340 @xref{Function Specific Option Pragmas}, for details about the
3341 @samp{#pragma GCC target} pragma.
3343 For instance, on an x86, you could declare one function with the
3344 @code{target("sse4.1,arch=core2")} attribute and another with
3345 @code{target("sse4a,arch=amdfam10")}. This is equivalent to
3346 compiling the first function with @option{-msse4.1} and
3347 @option{-march=core2} options, and the second function with
3348 @option{-msse4a} and @option{-march=amdfam10} options. It is up to you
3349 to make sure that a function is only invoked on a machine that
3350 supports the particular ISA it is compiled for (for example by using
3351 @code{cpuid} on x86 to determine what feature bits and architecture
3355 int core2_func (void) __attribute__ ((__target__ ("arch=core2")));
3356 int sse3_func (void) __attribute__ ((__target__ ("sse3")));
3359 You can either use multiple
3360 strings separated by commas to specify multiple options,
3361 or separate the options with a comma (@samp{,}) within a single string.
3363 The options supported are specific to each target; refer to @ref{x86
3364 Function Attributes}, @ref{PowerPC Function Attributes},
3365 @ref{ARM Function Attributes}, @ref{AArch64 Function Attributes},
3366 @ref{Nios II Function Attributes}, and @ref{S/390 Function Attributes}
3369 @item target_clones (@var{options})
3370 @cindex @code{target_clones} function attribute
3371 The @code{target_clones} attribute is used to specify that a function
3372 be cloned into multiple versions compiled with different target options
3373 than specified on the command line. The supported options and restrictions
3374 are the same as for @code{target} attribute.
3376 For instance, on an x86, you could compile a function with
3377 @code{target_clones("sse4.1,avx")}. GCC creates two function clones,
3378 one compiled with @option{-msse4.1} and another with @option{-mavx}.
3380 On a PowerPC, you can compile a function with
3381 @code{target_clones("cpu=power9,default")}. GCC will create two
3382 function clones, one compiled with @option{-mcpu=power9} and another
3383 with the default options. GCC must be configured to use GLIBC 2.23 or
3384 newer in order to use the @code{target_clones} attribute.
3386 It also creates a resolver function (see
3387 the @code{ifunc} attribute above) that dynamically selects a clone
3388 suitable for current architecture. The resolver is created only if there
3389 is a usage of a function with @code{target_clones} attribute.
3392 @cindex @code{unused} function attribute
3393 This attribute, attached to a function, means that the function is meant
3394 to be possibly unused. GCC does not produce a warning for this
3398 @cindex @code{used} function attribute
3399 This attribute, attached to a function, means that code must be emitted
3400 for the function even if it appears that the function is not referenced.
3401 This is useful, for example, when the function is referenced only in
3404 When applied to a member function of a C++ class template, the
3405 attribute also means that the function is instantiated if the
3406 class itself is instantiated.
3408 @item visibility ("@var{visibility_type}")
3409 @cindex @code{visibility} function attribute
3410 This attribute affects the linkage of the declaration to which it is attached.
3411 It can be applied to variables (@pxref{Common Variable Attributes}) and types
3412 (@pxref{Common Type Attributes}) as well as functions.
3414 There are four supported @var{visibility_type} values: default,
3415 hidden, protected or internal visibility.
3418 void __attribute__ ((visibility ("protected")))
3419 f () @{ /* @r{Do something.} */; @}
3420 int i __attribute__ ((visibility ("hidden")));
3423 The possible values of @var{visibility_type} correspond to the
3424 visibility settings in the ELF gABI.
3427 @c keep this list of visibilities in alphabetical order.
3430 Default visibility is the normal case for the object file format.
3431 This value is available for the visibility attribute to override other
3432 options that may change the assumed visibility of entities.
3434 On ELF, default visibility means that the declaration is visible to other
3435 modules and, in shared libraries, means that the declared entity may be
3438 On Darwin, default visibility means that the declaration is visible to
3441 Default visibility corresponds to ``external linkage'' in the language.
3444 Hidden visibility indicates that the entity declared has a new
3445 form of linkage, which we call ``hidden linkage''. Two
3446 declarations of an object with hidden linkage refer to the same object
3447 if they are in the same shared object.
3450 Internal visibility is like hidden visibility, but with additional
3451 processor specific semantics. Unless otherwise specified by the
3452 psABI, GCC defines internal visibility to mean that a function is
3453 @emph{never} called from another module. Compare this with hidden
3454 functions which, while they cannot be referenced directly by other
3455 modules, can be referenced indirectly via function pointers. By
3456 indicating that a function cannot be called from outside the module,
3457 GCC may for instance omit the load of a PIC register since it is known
3458 that the calling function loaded the correct value.
3461 Protected visibility is like default visibility except that it
3462 indicates that references within the defining module bind to the
3463 definition in that module. That is, the declared entity cannot be
3464 overridden by another module.
3468 All visibilities are supported on many, but not all, ELF targets
3469 (supported when the assembler supports the @samp{.visibility}
3470 pseudo-op). Default visibility is supported everywhere. Hidden
3471 visibility is supported on Darwin targets.
3473 The visibility attribute should be applied only to declarations that
3474 would otherwise have external linkage. The attribute should be applied
3475 consistently, so that the same entity should not be declared with
3476 different settings of the attribute.
3478 In C++, the visibility attribute applies to types as well as functions
3479 and objects, because in C++ types have linkage. A class must not have
3480 greater visibility than its non-static data member types and bases,
3481 and class members default to the visibility of their class. Also, a
3482 declaration without explicit visibility is limited to the visibility
3485 In C++, you can mark member functions and static member variables of a
3486 class with the visibility attribute. This is useful if you know a
3487 particular method or static member variable should only be used from
3488 one shared object; then you can mark it hidden while the rest of the
3489 class has default visibility. Care must be taken to avoid breaking
3490 the One Definition Rule; for example, it is usually not useful to mark
3491 an inline method as hidden without marking the whole class as hidden.
3493 A C++ namespace declaration can also have the visibility attribute.
3496 namespace nspace1 __attribute__ ((visibility ("protected")))
3497 @{ /* @r{Do something.} */; @}
3500 This attribute applies only to the particular namespace body, not to
3501 other definitions of the same namespace; it is equivalent to using
3502 @samp{#pragma GCC visibility} before and after the namespace
3503 definition (@pxref{Visibility Pragmas}).
3505 In C++, if a template argument has limited visibility, this
3506 restriction is implicitly propagated to the template instantiation.
3507 Otherwise, template instantiations and specializations default to the
3508 visibility of their template.
3510 If both the template and enclosing class have explicit visibility, the
3511 visibility from the template is used.
3513 @item warn_unused_result
3514 @cindex @code{warn_unused_result} function attribute
3515 The @code{warn_unused_result} attribute causes a warning to be emitted
3516 if a caller of the function with this attribute does not use its
3517 return value. This is useful for functions where not checking
3518 the result is either a security problem or always a bug, such as
3522 int fn () __attribute__ ((warn_unused_result));
3525 if (fn () < 0) return -1;
3532 results in warning on line 5.
3535 @cindex @code{weak} function attribute
3536 The @code{weak} attribute causes the declaration to be emitted as a weak
3537 symbol rather than a global. This is primarily useful in defining
3538 library functions that can be overridden in user code, though it can
3539 also be used with non-function declarations. Weak symbols are supported
3540 for ELF targets, and also for a.out targets when using the GNU assembler
3544 @itemx weakref ("@var{target}")
3545 @cindex @code{weakref} function attribute
3546 The @code{weakref} attribute marks a declaration as a weak reference.
3547 Without arguments, it should be accompanied by an @code{alias} attribute
3548 naming the target symbol. Optionally, the @var{target} may be given as
3549 an argument to @code{weakref} itself. In either case, @code{weakref}
3550 implicitly marks the declaration as @code{weak}. Without a
3551 @var{target}, given as an argument to @code{weakref} or to @code{alias},
3552 @code{weakref} is equivalent to @code{weak}.
3555 static int x() __attribute__ ((weakref ("y")));
3556 /* is equivalent to... */
3557 static int x() __attribute__ ((weak, weakref, alias ("y")));
3559 static int x() __attribute__ ((weakref));
3560 static int x() __attribute__ ((alias ("y")));
3563 A weak reference is an alias that does not by itself require a
3564 definition to be given for the target symbol. If the target symbol is
3565 only referenced through weak references, then it becomes a @code{weak}
3566 undefined symbol. If it is directly referenced, however, then such
3567 strong references prevail, and a definition is required for the
3568 symbol, not necessarily in the same translation unit.
3570 The effect is equivalent to moving all references to the alias to a
3571 separate translation unit, renaming the alias to the aliased symbol,
3572 declaring it as weak, compiling the two separate translation units and
3573 performing a reloadable link on them.
3575 At present, a declaration to which @code{weakref} is attached can
3576 only be @code{static}.
3581 @c This is the end of the target-independent attribute table
3583 @node AArch64 Function Attributes
3584 @subsection AArch64 Function Attributes
3586 The following target-specific function attributes are available for the
3587 AArch64 target. For the most part, these options mirror the behavior of
3588 similar command-line options (@pxref{AArch64 Options}), but on a
3592 @item general-regs-only
3593 @cindex @code{general-regs-only} function attribute, AArch64
3594 Indicates that no floating-point or Advanced SIMD registers should be
3595 used when generating code for this function. If the function explicitly
3596 uses floating-point code, then the compiler gives an error. This is
3597 the same behavior as that of the command-line option
3598 @option{-mgeneral-regs-only}.
3600 @item fix-cortex-a53-835769
3601 @cindex @code{fix-cortex-a53-835769} function attribute, AArch64
3602 Indicates that the workaround for the Cortex-A53 erratum 835769 should be
3603 applied to this function. To explicitly disable the workaround for this
3604 function specify the negated form: @code{no-fix-cortex-a53-835769}.
3605 This corresponds to the behavior of the command line options
3606 @option{-mfix-cortex-a53-835769} and @option{-mno-fix-cortex-a53-835769}.
3609 @cindex @code{cmodel=} function attribute, AArch64
3610 Indicates that code should be generated for a particular code model for
3611 this function. The behavior and permissible arguments are the same as
3612 for the command line option @option{-mcmodel=}.
3615 @itemx no-strict-align
3616 @cindex @code{strict-align} function attribute, AArch64
3617 @code{strict-align} indicates that the compiler should not assume that unaligned
3618 memory references are handled by the system. To allow the compiler to assume
3619 that aligned memory references are handled by the system, the inverse attribute
3620 @code{no-strict-align} can be specified. The behavior is same as for the
3621 command-line option @option{-mstrict-align} and @option{-mno-strict-align}.
3623 @item omit-leaf-frame-pointer
3624 @cindex @code{omit-leaf-frame-pointer} function attribute, AArch64
3625 Indicates that the frame pointer should be omitted for a leaf function call.
3626 To keep the frame pointer, the inverse attribute
3627 @code{no-omit-leaf-frame-pointer} can be specified. These attributes have
3628 the same behavior as the command-line options @option{-momit-leaf-frame-pointer}
3629 and @option{-mno-omit-leaf-frame-pointer}.
3632 @cindex @code{tls-dialect=} function attribute, AArch64
3633 Specifies the TLS dialect to use for this function. The behavior and
3634 permissible arguments are the same as for the command-line option
3635 @option{-mtls-dialect=}.
3638 @cindex @code{arch=} function attribute, AArch64
3639 Specifies the architecture version and architectural extensions to use
3640 for this function. The behavior and permissible arguments are the same as
3641 for the @option{-march=} command-line option.
3644 @cindex @code{tune=} function attribute, AArch64
3645 Specifies the core for which to tune the performance of this function.
3646 The behavior and permissible arguments are the same as for the @option{-mtune=}
3647 command-line option.
3650 @cindex @code{cpu=} function attribute, AArch64
3651 Specifies the core for which to tune the performance of this function and also
3652 whose architectural features to use. The behavior and valid arguments are the
3653 same as for the @option{-mcpu=} command-line option.
3655 @item sign-return-address
3656 @cindex @code{sign-return-address} function attribute, AArch64
3657 Select the function scope on which return address signing will be applied. The
3658 behavior and permissible arguments are the same as for the command-line option
3659 @option{-msign-return-address=}. The default value is @code{none}.
3663 The above target attributes can be specified as follows:
3666 __attribute__((target("@var{attr-string}")))
3674 where @code{@var{attr-string}} is one of the attribute strings specified above.
3676 Additionally, the architectural extension string may be specified on its
3677 own. This can be used to turn on and off particular architectural extensions
3678 without having to specify a particular architecture version or core. Example:
3681 __attribute__((target("+crc+nocrypto")))
3689 In this example @code{target("+crc+nocrypto")} enables the @code{crc}
3690 extension and disables the @code{crypto} extension for the function @code{foo}
3691 without modifying an existing @option{-march=} or @option{-mcpu} option.
3693 Multiple target function attributes can be specified by separating them with
3694 a comma. For example:
3696 __attribute__((target("arch=armv8-a+crc+crypto,tune=cortex-a53")))
3704 is valid and compiles function @code{foo} for ARMv8-A with @code{crc}
3705 and @code{crypto} extensions and tunes it for @code{cortex-a53}.
3707 @subsubsection Inlining rules
3708 Specifying target attributes on individual functions or performing link-time
3709 optimization across translation units compiled with different target options
3710 can affect function inlining rules:
3712 In particular, a caller function can inline a callee function only if the
3713 architectural features available to the callee are a subset of the features
3714 available to the caller.
3715 For example: A function @code{foo} compiled with @option{-march=armv8-a+crc},
3716 or tagged with the equivalent @code{arch=armv8-a+crc} attribute,
3717 can inline a function @code{bar} compiled with @option{-march=armv8-a+nocrc}
3718 because the all the architectural features that function @code{bar} requires
3719 are available to function @code{foo}. Conversely, function @code{bar} cannot
3720 inline function @code{foo}.
3722 Additionally inlining a function compiled with @option{-mstrict-align} into a
3723 function compiled without @code{-mstrict-align} is not allowed.
3724 However, inlining a function compiled without @option{-mstrict-align} into a
3725 function compiled with @option{-mstrict-align} is allowed.
3727 Note that CPU tuning options and attributes such as the @option{-mcpu=},
3728 @option{-mtune=} do not inhibit inlining unless the CPU specified by the
3729 @option{-mcpu=} option or the @code{cpu=} attribute conflicts with the
3730 architectural feature rules specified above.
3732 @node ARC Function Attributes
3733 @subsection ARC Function Attributes
3735 These function attributes are supported by the ARC back end:
3739 @cindex @code{interrupt} function attribute, ARC
3740 Use this attribute to indicate
3741 that the specified function is an interrupt handler. The compiler generates
3742 function entry and exit sequences suitable for use in an interrupt handler
3743 when this attribute is present.
3745 On the ARC, you must specify the kind of interrupt to be handled
3746 in a parameter to the interrupt attribute like this:
3749 void f () __attribute__ ((interrupt ("ilink1")));
3752 Permissible values for this parameter are: @w{@code{ilink1}} and
3758 @cindex @code{long_call} function attribute, ARC
3759 @cindex @code{medium_call} function attribute, ARC
3760 @cindex @code{short_call} function attribute, ARC
3761 @cindex indirect calls, ARC
3762 These attributes specify how a particular function is called.
3763 These attributes override the
3764 @option{-mlong-calls} and @option{-mmedium-calls} (@pxref{ARC Options})
3765 command-line switches and @code{#pragma long_calls} settings.
3767 For ARC, a function marked with the @code{long_call} attribute is
3768 always called using register-indirect jump-and-link instructions,
3769 thereby enabling the called function to be placed anywhere within the
3770 32-bit address space. A function marked with the @code{medium_call}
3771 attribute will always be close enough to be called with an unconditional
3772 branch-and-link instruction, which has a 25-bit offset from
3773 the call site. A function marked with the @code{short_call}
3774 attribute will always be close enough to be called with a conditional
3775 branch-and-link instruction, which has a 21-bit offset from
3779 @cindex @code{jli_always} function attribute, ARC
3780 Forces a particular function to be called using @code{jli}
3781 instruction. The @code{jli} instruction makes use of a table stored
3782 into @code{.jlitab} section, which holds the location of the functions
3783 which are addressed using this instruction.
3786 @cindex @code{jli_fixed} function attribute, ARC
3787 Identical like the above one, but the location of the function in the
3788 @code{jli} table is known and given as an attribute parameter.
3791 @cindex @code{secure_call} function attribute, ARC
3792 This attribute allows one to mark secure-code functions that are
3793 callable from normal mode. The location of the secure call function
3794 into the @code{sjli} table needs to be passed as argument.
3798 @node ARM Function Attributes
3799 @subsection ARM Function Attributes
3801 These function attributes are supported for ARM targets:
3805 @cindex @code{interrupt} function attribute, ARM
3806 Use this attribute to indicate
3807 that the specified function is an interrupt handler. The compiler generates
3808 function entry and exit sequences suitable for use in an interrupt handler
3809 when this attribute is present.
3811 You can specify the kind of interrupt to be handled by
3812 adding an optional parameter to the interrupt attribute like this:
3815 void f () __attribute__ ((interrupt ("IRQ")));
3819 Permissible values for this parameter are: @code{IRQ}, @code{FIQ},
3820 @code{SWI}, @code{ABORT} and @code{UNDEF}.
3822 On ARMv7-M the interrupt type is ignored, and the attribute means the function
3823 may be called with a word-aligned stack pointer.
3826 @cindex @code{isr} function attribute, ARM
3827 Use this attribute on ARM to write Interrupt Service Routines. This is an
3828 alias to the @code{interrupt} attribute above.
3832 @cindex @code{long_call} function attribute, ARM
3833 @cindex @code{short_call} function attribute, ARM
3834 @cindex indirect calls, ARM
3835 These attributes specify how a particular function is called.
3836 These attributes override the
3837 @option{-mlong-calls} (@pxref{ARM Options})
3838 command-line switch and @code{#pragma long_calls} settings. For ARM, the
3839 @code{long_call} attribute indicates that the function might be far
3840 away from the call site and require a different (more expensive)
3841 calling sequence. The @code{short_call} attribute always places
3842 the offset to the function from the call site into the @samp{BL}
3843 instruction directly.
3846 @cindex @code{naked} function attribute, ARM
3847 This attribute allows the compiler to construct the
3848 requisite function declaration, while allowing the body of the
3849 function to be assembly code. The specified function will not have
3850 prologue/epilogue sequences generated by the compiler. Only basic
3851 @code{asm} statements can safely be included in naked functions
3852 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
3853 basic @code{asm} and C code may appear to work, they cannot be
3854 depended upon to work reliably and are not supported.
3857 @cindex @code{pcs} function attribute, ARM
3859 The @code{pcs} attribute can be used to control the calling convention
3860 used for a function on ARM. The attribute takes an argument that specifies
3861 the calling convention to use.
3863 When compiling using the AAPCS ABI (or a variant of it) then valid
3864 values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In
3865 order to use a variant other than @code{"aapcs"} then the compiler must
3866 be permitted to use the appropriate co-processor registers (i.e., the
3867 VFP registers must be available in order to use @code{"aapcs-vfp"}).
3871 /* Argument passed in r0, and result returned in r0+r1. */
3872 double f2d (float) __attribute__((pcs("aapcs")));
3875 Variadic functions always use the @code{"aapcs"} calling convention and
3876 the compiler rejects attempts to specify an alternative.
3878 @item target (@var{options})
3879 @cindex @code{target} function attribute
3880 As discussed in @ref{Common Function Attributes}, this attribute
3881 allows specification of target-specific compilation options.
3883 On ARM, the following options are allowed:
3887 @cindex @code{target("thumb")} function attribute, ARM
3888 Force code generation in the Thumb (T16/T32) ISA, depending on the
3892 @cindex @code{target("arm")} function attribute, ARM
3893 Force code generation in the ARM (A32) ISA.
3895 Functions from different modes can be inlined in the caller's mode.
3898 @cindex @code{target("fpu=")} function attribute, ARM
3899 Specifies the fpu for which to tune the performance of this function.
3900 The behavior and permissible arguments are the same as for the @option{-mfpu=}
3901 command-line option.
3904 @cindex @code{arch=} function attribute, ARM
3905 Specifies the architecture version and architectural extensions to use
3906 for this function. The behavior and permissible arguments are the same as
3907 for the @option{-march=} command-line option.
3909 The above target attributes can be specified as follows:
3912 __attribute__((target("arch=armv8-a+crc")))
3920 Additionally, the architectural extension string may be specified on its
3921 own. This can be used to turn on and off particular architectural extensions
3922 without having to specify a particular architecture version or core. Example:
3925 __attribute__((target("+crc+nocrypto")))
3933 In this example @code{target("+crc+nocrypto")} enables the @code{crc}
3934 extension and disables the @code{crypto} extension for the function @code{foo}
3935 without modifying an existing @option{-march=} or @option{-mcpu} option.
3941 @node AVR Function Attributes
3942 @subsection AVR Function Attributes
3944 These function attributes are supported by the AVR back end:
3948 @cindex @code{interrupt} function attribute, AVR
3949 Use this attribute to indicate
3950 that the specified function is an interrupt handler. The compiler generates
3951 function entry and exit sequences suitable for use in an interrupt handler
3952 when this attribute is present.
3954 On the AVR, the hardware globally disables interrupts when an
3955 interrupt is executed. The first instruction of an interrupt handler
3956 declared with this attribute is a @code{SEI} instruction to
3957 re-enable interrupts. See also the @code{signal} function attribute
3958 that does not insert a @code{SEI} instruction. If both @code{signal} and
3959 @code{interrupt} are specified for the same function, @code{signal}
3960 is silently ignored.
3963 @cindex @code{naked} function attribute, AVR
3964 This attribute allows the compiler to construct the
3965 requisite function declaration, while allowing the body of the
3966 function to be assembly code. The specified function will not have
3967 prologue/epilogue sequences generated by the compiler. Only basic
3968 @code{asm} statements can safely be included in naked functions
3969 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
3970 basic @code{asm} and C code may appear to work, they cannot be
3971 depended upon to work reliably and are not supported.
3974 @cindex @code{no_gccisr} function attribute, AVR
3975 Do not use @code{__gcc_isr} pseudo instructions in a function with
3976 the @code{interrupt} or @code{signal} attribute aka. interrupt
3977 service routine (ISR).
3978 Use this attribute if the preamble of the ISR prologue should always read
3982 in __tmp_reg__, __SREG__
3986 and accordingly for the postamble of the epilogue --- no matter whether
3987 the mentioned registers are actually used in the ISR or not.
3988 Situations where you might want to use this attribute include:
3991 Code that (effectively) clobbers bits of @code{SREG} other than the
3992 @code{I}-flag by writing to the memory location of @code{SREG}.
3994 Code that uses inline assembler to jump to a different function which
3995 expects (parts of) the prologue code as outlined above to be present.
3997 To disable @code{__gcc_isr} generation for the whole compilation unit,
3998 there is option @option{-mno-gas-isr-prologues}, @pxref{AVR Options}.
4002 @cindex @code{OS_main} function attribute, AVR
4003 @cindex @code{OS_task} function attribute, AVR
4004 On AVR, functions with the @code{OS_main} or @code{OS_task} attribute
4005 do not save/restore any call-saved register in their prologue/epilogue.
4007 The @code{OS_main} attribute can be used when there @emph{is
4008 guarantee} that interrupts are disabled at the time when the function
4009 is entered. This saves resources when the stack pointer has to be
4010 changed to set up a frame for local variables.
4012 The @code{OS_task} attribute can be used when there is @emph{no
4013 guarantee} that interrupts are disabled at that time when the function
4014 is entered like for, e@.g@. task functions in a multi-threading operating
4015 system. In that case, changing the stack pointer register is
4016 guarded by save/clear/restore of the global interrupt enable flag.
4018 The differences to the @code{naked} function attribute are:
4020 @item @code{naked} functions do not have a return instruction whereas
4021 @code{OS_main} and @code{OS_task} functions have a @code{RET} or
4022 @code{RETI} return instruction.
4023 @item @code{naked} functions do not set up a frame for local variables
4024 or a frame pointer whereas @code{OS_main} and @code{OS_task} do this
4029 @cindex @code{signal} function attribute, AVR
4030 Use this attribute on the AVR to indicate that the specified
4031 function is an interrupt handler. The compiler generates function
4032 entry and exit sequences suitable for use in an interrupt handler when this
4033 attribute is present.
4035 See also the @code{interrupt} function attribute.
4037 The AVR hardware globally disables interrupts when an interrupt is executed.
4038 Interrupt handler functions defined with the @code{signal} attribute
4039 do not re-enable interrupts. It is save to enable interrupts in a
4040 @code{signal} handler. This ``save'' only applies to the code
4041 generated by the compiler and not to the IRQ layout of the
4042 application which is responsibility of the application.
4044 If both @code{signal} and @code{interrupt} are specified for the same
4045 function, @code{signal} is silently ignored.
4048 @node Blackfin Function Attributes
4049 @subsection Blackfin Function Attributes
4051 These function attributes are supported by the Blackfin back end:
4055 @item exception_handler
4056 @cindex @code{exception_handler} function attribute
4057 @cindex exception handler functions, Blackfin
4058 Use this attribute on the Blackfin to indicate that the specified function
4059 is an exception handler. The compiler generates function entry and
4060 exit sequences suitable for use in an exception handler when this
4061 attribute is present.
4063 @item interrupt_handler
4064 @cindex @code{interrupt_handler} function attribute, Blackfin
4065 Use this attribute to
4066 indicate that the specified function is an interrupt handler. The compiler
4067 generates function entry and exit sequences suitable for use in an
4068 interrupt handler when this attribute is present.
4071 @cindex @code{kspisusp} function attribute, Blackfin
4072 @cindex User stack pointer in interrupts on the Blackfin
4073 When used together with @code{interrupt_handler}, @code{exception_handler}
4074 or @code{nmi_handler}, code is generated to load the stack pointer
4075 from the USP register in the function prologue.
4078 @cindex @code{l1_text} function attribute, Blackfin
4079 This attribute specifies a function to be placed into L1 Instruction
4080 SRAM@. The function is put into a specific section named @code{.l1.text}.
4081 With @option{-mfdpic}, function calls with a such function as the callee
4082 or caller uses inlined PLT.
4085 @cindex @code{l2} function attribute, Blackfin
4086 This attribute specifies a function to be placed into L2
4087 SRAM. The function is put into a specific section named
4088 @code{.l2.text}. With @option{-mfdpic}, callers of such functions use
4093 @cindex indirect calls, Blackfin
4094 @cindex @code{longcall} function attribute, Blackfin
4095 @cindex @code{shortcall} function attribute, Blackfin
4096 The @code{longcall} attribute
4097 indicates that the function might be far away from the call site and
4098 require a different (more expensive) calling sequence. The
4099 @code{shortcall} attribute indicates that the function is always close
4100 enough for the shorter calling sequence to be used. These attributes
4101 override the @option{-mlongcall} switch.
4104 @cindex @code{nesting} function attribute, Blackfin
4105 @cindex Allow nesting in an interrupt handler on the Blackfin processor
4106 Use this attribute together with @code{interrupt_handler},
4107 @code{exception_handler} or @code{nmi_handler} to indicate that the function
4108 entry code should enable nested interrupts or exceptions.
4111 @cindex @code{nmi_handler} function attribute, Blackfin
4112 @cindex NMI handler functions on the Blackfin processor
4113 Use this attribute on the Blackfin to indicate that the specified function
4114 is an NMI handler. The compiler generates function entry and
4115 exit sequences suitable for use in an NMI handler when this
4116 attribute is present.
4119 @cindex @code{saveall} function attribute, Blackfin
4120 @cindex save all registers on the Blackfin
4121 Use this attribute to indicate that
4122 all registers except the stack pointer should be saved in the prologue
4123 regardless of whether they are used or not.
4126 @node CR16 Function Attributes
4127 @subsection CR16 Function Attributes
4129 These function attributes are supported by the CR16 back end:
4133 @cindex @code{interrupt} function attribute, CR16
4134 Use this attribute to indicate
4135 that the specified function is an interrupt handler. The compiler generates
4136 function entry and exit sequences suitable for use in an interrupt handler
4137 when this attribute is present.
4140 @node Epiphany Function Attributes
4141 @subsection Epiphany Function Attributes
4143 These function attributes are supported by the Epiphany back end:
4147 @cindex @code{disinterrupt} function attribute, Epiphany
4148 This attribute causes the compiler to emit
4149 instructions to disable interrupts for the duration of the given
4152 @item forwarder_section
4153 @cindex @code{forwarder_section} function attribute, Epiphany
4154 This attribute modifies the behavior of an interrupt handler.
4155 The interrupt handler may be in external memory which cannot be
4156 reached by a branch instruction, so generate a local memory trampoline
4157 to transfer control. The single parameter identifies the section where
4158 the trampoline is placed.
4161 @cindex @code{interrupt} function attribute, Epiphany
4162 Use this attribute to indicate
4163 that the specified function is an interrupt handler. The compiler generates
4164 function entry and exit sequences suitable for use in an interrupt handler
4165 when this attribute is present. It may also generate
4166 a special section with code to initialize the interrupt vector table.
4168 On Epiphany targets one or more optional parameters can be added like this:
4171 void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler ();
4174 Permissible values for these parameters are: @w{@code{reset}},
4175 @w{@code{software_exception}}, @w{@code{page_miss}},
4176 @w{@code{timer0}}, @w{@code{timer1}}, @w{@code{message}},
4177 @w{@code{dma0}}, @w{@code{dma1}}, @w{@code{wand}} and @w{@code{swi}}.
4178 Multiple parameters indicate that multiple entries in the interrupt
4179 vector table should be initialized for this function, i.e.@: for each
4180 parameter @w{@var{name}}, a jump to the function is emitted in
4181 the section @w{ivt_entry_@var{name}}. The parameter(s) may be omitted
4182 entirely, in which case no interrupt vector table entry is provided.
4184 Note that interrupts are enabled inside the function
4185 unless the @code{disinterrupt} attribute is also specified.
4187 The following examples are all valid uses of these attributes on
4190 void __attribute__ ((interrupt)) universal_handler ();
4191 void __attribute__ ((interrupt ("dma1"))) dma1_handler ();
4192 void __attribute__ ((interrupt ("dma0, dma1")))
4193 universal_dma_handler ();
4194 void __attribute__ ((interrupt ("timer0"), disinterrupt))
4195 fast_timer_handler ();
4196 void __attribute__ ((interrupt ("dma0, dma1"),
4197 forwarder_section ("tramp")))
4198 external_dma_handler ();
4203 @cindex @code{long_call} function attribute, Epiphany
4204 @cindex @code{short_call} function attribute, Epiphany
4205 @cindex indirect calls, Epiphany
4206 These attributes specify how a particular function is called.
4207 These attributes override the
4208 @option{-mlong-calls} (@pxref{Adapteva Epiphany Options})
4209 command-line switch and @code{#pragma long_calls} settings.
4213 @node H8/300 Function Attributes
4214 @subsection H8/300 Function Attributes
4216 These function attributes are available for H8/300 targets:
4219 @item function_vector
4220 @cindex @code{function_vector} function attribute, H8/300
4221 Use this attribute on the H8/300, H8/300H, and H8S to indicate
4222 that the specified function should be called through the function vector.
4223 Calling a function through the function vector reduces code size; however,
4224 the function vector has a limited size (maximum 128 entries on the H8/300
4225 and 64 entries on the H8/300H and H8S)
4226 and shares space with the interrupt vector.
4228 @item interrupt_handler
4229 @cindex @code{interrupt_handler} function attribute, H8/300
4230 Use this attribute on the H8/300, H8/300H, and H8S to
4231 indicate that the specified function is an interrupt handler. The compiler
4232 generates function entry and exit sequences suitable for use in an
4233 interrupt handler when this attribute is present.
4236 @cindex @code{saveall} function attribute, H8/300
4237 @cindex save all registers on the H8/300, H8/300H, and H8S
4238 Use this attribute on the H8/300, H8/300H, and H8S to indicate that
4239 all registers except the stack pointer should be saved in the prologue
4240 regardless of whether they are used or not.
4243 @node IA-64 Function Attributes
4244 @subsection IA-64 Function Attributes
4246 These function attributes are supported on IA-64 targets:
4249 @item syscall_linkage
4250 @cindex @code{syscall_linkage} function attribute, IA-64
4251 This attribute is used to modify the IA-64 calling convention by marking
4252 all input registers as live at all function exits. This makes it possible
4253 to restart a system call after an interrupt without having to save/restore
4254 the input registers. This also prevents kernel data from leaking into
4258 @cindex @code{version_id} function attribute, IA-64
4259 This IA-64 HP-UX attribute, attached to a global variable or function, renames a
4260 symbol to contain a version string, thus allowing for function level
4261 versioning. HP-UX system header files may use function level versioning
4262 for some system calls.
4265 extern int foo () __attribute__((version_id ("20040821")));
4269 Calls to @code{foo} are mapped to calls to @code{foo@{20040821@}}.
4272 @node M32C Function Attributes
4273 @subsection M32C Function Attributes
4275 These function attributes are supported by the M32C back end:
4279 @cindex @code{bank_switch} function attribute, M32C
4280 When added to an interrupt handler with the M32C port, causes the
4281 prologue and epilogue to use bank switching to preserve the registers
4282 rather than saving them on the stack.
4284 @item fast_interrupt
4285 @cindex @code{fast_interrupt} function attribute, M32C
4286 Use this attribute on the M32C port to indicate that the specified
4287 function is a fast interrupt handler. This is just like the
4288 @code{interrupt} attribute, except that @code{freit} is used to return
4289 instead of @code{reit}.
4291 @item function_vector
4292 @cindex @code{function_vector} function attribute, M16C/M32C
4293 On M16C/M32C targets, the @code{function_vector} attribute declares a
4294 special page subroutine call function. Use of this attribute reduces
4295 the code size by 2 bytes for each call generated to the
4296 subroutine. The argument to the attribute is the vector number entry
4297 from the special page vector table which contains the 16 low-order
4298 bits of the subroutine's entry address. Each vector table has special
4299 page number (18 to 255) that is used in @code{jsrs} instructions.
4300 Jump addresses of the routines are generated by adding 0x0F0000 (in
4301 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the
4302 2-byte addresses set in the vector table. Therefore you need to ensure
4303 that all the special page vector routines should get mapped within the
4304 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
4307 In the following example 2 bytes are saved for each call to
4308 function @code{foo}.
4311 void foo (void) __attribute__((function_vector(0x18)));
4322 If functions are defined in one file and are called in another file,
4323 then be sure to write this declaration in both files.
4325 This attribute is ignored for R8C target.
4328 @cindex @code{interrupt} function attribute, M32C
4329 Use this attribute to indicate
4330 that the specified function is an interrupt handler. The compiler generates
4331 function entry and exit sequences suitable for use in an interrupt handler
4332 when this attribute is present.
4335 @node M32R/D Function Attributes
4336 @subsection M32R/D Function Attributes
4338 These function attributes are supported by the M32R/D back end:
4342 @cindex @code{interrupt} function attribute, M32R/D
4343 Use this attribute to indicate
4344 that the specified function is an interrupt handler. The compiler generates
4345 function entry and exit sequences suitable for use in an interrupt handler
4346 when this attribute is present.
4348 @item model (@var{model-name})
4349 @cindex @code{model} function attribute, M32R/D
4350 @cindex function addressability on the M32R/D
4352 On the M32R/D, use this attribute to set the addressability of an
4353 object, and of the code generated for a function. The identifier
4354 @var{model-name} is one of @code{small}, @code{medium}, or
4355 @code{large}, representing each of the code models.
4357 Small model objects live in the lower 16MB of memory (so that their
4358 addresses can be loaded with the @code{ld24} instruction), and are
4359 callable with the @code{bl} instruction.
4361 Medium model objects may live anywhere in the 32-bit address space (the
4362 compiler generates @code{seth/add3} instructions to load their addresses),
4363 and are callable with the @code{bl} instruction.
4365 Large model objects may live anywhere in the 32-bit address space (the
4366 compiler generates @code{seth/add3} instructions to load their addresses),
4367 and may not be reachable with the @code{bl} instruction (the compiler
4368 generates the much slower @code{seth/add3/jl} instruction sequence).
4371 @node m68k Function Attributes
4372 @subsection m68k Function Attributes
4374 These function attributes are supported by the m68k back end:
4378 @itemx interrupt_handler
4379 @cindex @code{interrupt} function attribute, m68k
4380 @cindex @code{interrupt_handler} function attribute, m68k
4381 Use this attribute to
4382 indicate that the specified function is an interrupt handler. The compiler
4383 generates function entry and exit sequences suitable for use in an
4384 interrupt handler when this attribute is present. Either name may be used.
4386 @item interrupt_thread
4387 @cindex @code{interrupt_thread} function attribute, fido
4388 Use this attribute on fido, a subarchitecture of the m68k, to indicate
4389 that the specified function is an interrupt handler that is designed
4390 to run as a thread. The compiler omits generate prologue/epilogue
4391 sequences and replaces the return instruction with a @code{sleep}
4392 instruction. This attribute is available only on fido.
4395 @node MCORE Function Attributes
4396 @subsection MCORE Function Attributes
4398 These function attributes are supported by the MCORE back end:
4402 @cindex @code{naked} function attribute, MCORE
4403 This attribute allows the compiler to construct the
4404 requisite function declaration, while allowing the body of the
4405 function to be assembly code. The specified function will not have
4406 prologue/epilogue sequences generated by the compiler. Only basic
4407 @code{asm} statements can safely be included in naked functions
4408 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
4409 basic @code{asm} and C code may appear to work, they cannot be
4410 depended upon to work reliably and are not supported.
4413 @node MeP Function Attributes
4414 @subsection MeP Function Attributes
4416 These function attributes are supported by the MeP back end:
4420 @cindex @code{disinterrupt} function attribute, MeP
4421 On MeP targets, this attribute causes the compiler to emit
4422 instructions to disable interrupts for the duration of the given
4426 @cindex @code{interrupt} function attribute, MeP
4427 Use this attribute to indicate
4428 that the specified function is an interrupt handler. The compiler generates
4429 function entry and exit sequences suitable for use in an interrupt handler
4430 when this attribute is present.
4433 @cindex @code{near} function attribute, MeP
4434 This attribute causes the compiler to assume the called
4435 function is close enough to use the normal calling convention,
4436 overriding the @option{-mtf} command-line option.
4439 @cindex @code{far} function attribute, MeP
4440 On MeP targets this causes the compiler to use a calling convention
4441 that assumes the called function is too far away for the built-in
4445 @cindex @code{vliw} function attribute, MeP
4446 The @code{vliw} attribute tells the compiler to emit
4447 instructions in VLIW mode instead of core mode. Note that this
4448 attribute is not allowed unless a VLIW coprocessor has been configured
4449 and enabled through command-line options.
4452 @node MicroBlaze Function Attributes
4453 @subsection MicroBlaze Function Attributes
4455 These function attributes are supported on MicroBlaze targets:
4458 @item save_volatiles
4459 @cindex @code{save_volatiles} function attribute, MicroBlaze
4460 Use this attribute to indicate that the function is
4461 an interrupt handler. All volatile registers (in addition to non-volatile
4462 registers) are saved in the function prologue. If the function is a leaf
4463 function, only volatiles used by the function are saved. A normal function
4464 return is generated instead of a return from interrupt.
4467 @cindex @code{break_handler} function attribute, MicroBlaze
4468 @cindex break handler functions
4469 Use this attribute to indicate that
4470 the specified function is a break handler. The compiler generates function
4471 entry and exit sequences suitable for use in an break handler when this
4472 attribute is present. The return from @code{break_handler} is done through
4473 the @code{rtbd} instead of @code{rtsd}.
4476 void f () __attribute__ ((break_handler));
4479 @item interrupt_handler
4480 @itemx fast_interrupt
4481 @cindex @code{interrupt_handler} function attribute, MicroBlaze
4482 @cindex @code{fast_interrupt} function attribute, MicroBlaze
4483 These attributes indicate that the specified function is an interrupt
4484 handler. Use the @code{fast_interrupt} attribute to indicate handlers
4485 used in low-latency interrupt mode, and @code{interrupt_handler} for
4486 interrupts that do not use low-latency handlers. In both cases, GCC
4487 emits appropriate prologue code and generates a return from the handler
4488 using @code{rtid} instead of @code{rtsd}.
4491 @node Microsoft Windows Function Attributes
4492 @subsection Microsoft Windows Function Attributes
4494 The following attributes are available on Microsoft Windows and Symbian OS
4499 @cindex @code{dllexport} function attribute
4500 @cindex @code{__declspec(dllexport)}
4501 On Microsoft Windows targets and Symbian OS targets the
4502 @code{dllexport} attribute causes the compiler to provide a global
4503 pointer to a pointer in a DLL, so that it can be referenced with the
4504 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
4505 name is formed by combining @code{_imp__} and the function or variable
4508 You can use @code{__declspec(dllexport)} as a synonym for
4509 @code{__attribute__ ((dllexport))} for compatibility with other
4512 On systems that support the @code{visibility} attribute, this
4513 attribute also implies ``default'' visibility. It is an error to
4514 explicitly specify any other visibility.
4516 GCC's default behavior is to emit all inline functions with the
4517 @code{dllexport} attribute. Since this can cause object file-size bloat,
4518 you can use @option{-fno-keep-inline-dllexport}, which tells GCC to
4519 ignore the attribute for inlined functions unless the
4520 @option{-fkeep-inline-functions} flag is used instead.
4522 The attribute is ignored for undefined symbols.
4524 When applied to C++ classes, the attribute marks defined non-inlined
4525 member functions and static data members as exports. Static consts
4526 initialized in-class are not marked unless they are also defined
4529 For Microsoft Windows targets there are alternative methods for
4530 including the symbol in the DLL's export table such as using a
4531 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
4532 the @option{--export-all} linker flag.
4535 @cindex @code{dllimport} function attribute
4536 @cindex @code{__declspec(dllimport)}
4537 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
4538 attribute causes the compiler to reference a function or variable via
4539 a global pointer to a pointer that is set up by the DLL exporting the
4540 symbol. The attribute implies @code{extern}. On Microsoft Windows
4541 targets, the pointer name is formed by combining @code{_imp__} and the
4542 function or variable name.
4544 You can use @code{__declspec(dllimport)} as a synonym for
4545 @code{__attribute__ ((dllimport))} for compatibility with other
4548 On systems that support the @code{visibility} attribute, this
4549 attribute also implies ``default'' visibility. It is an error to
4550 explicitly specify any other visibility.
4552 Currently, the attribute is ignored for inlined functions. If the
4553 attribute is applied to a symbol @emph{definition}, an error is reported.
4554 If a symbol previously declared @code{dllimport} is later defined, the
4555 attribute is ignored in subsequent references, and a warning is emitted.
4556 The attribute is also overridden by a subsequent declaration as
4559 When applied to C++ classes, the attribute marks non-inlined
4560 member functions and static data members as imports. However, the
4561 attribute is ignored for virtual methods to allow creation of vtables
4564 On the SH Symbian OS target the @code{dllimport} attribute also has
4565 another affect---it can cause the vtable and run-time type information
4566 for a class to be exported. This happens when the class has a
4567 dllimported constructor or a non-inline, non-pure virtual function
4568 and, for either of those two conditions, the class also has an inline
4569 constructor or destructor and has a key function that is defined in
4570 the current translation unit.
4572 For Microsoft Windows targets the use of the @code{dllimport}
4573 attribute on functions is not necessary, but provides a small
4574 performance benefit by eliminating a thunk in the DLL@. The use of the
4575 @code{dllimport} attribute on imported variables can be avoided by passing the
4576 @option{--enable-auto-import} switch to the GNU linker. As with
4577 functions, using the attribute for a variable eliminates a thunk in
4580 One drawback to using this attribute is that a pointer to a
4581 @emph{variable} marked as @code{dllimport} cannot be used as a constant
4582 address. However, a pointer to a @emph{function} with the
4583 @code{dllimport} attribute can be used as a constant initializer; in
4584 this case, the address of a stub function in the import lib is
4585 referenced. On Microsoft Windows targets, the attribute can be disabled
4586 for functions by setting the @option{-mnop-fun-dllimport} flag.
4589 @node MIPS Function Attributes
4590 @subsection MIPS Function Attributes
4592 These function attributes are supported by the MIPS back end:
4596 @cindex @code{interrupt} function attribute, MIPS
4597 Use this attribute to indicate that the specified function is an interrupt
4598 handler. The compiler generates function entry and exit sequences suitable
4599 for use in an interrupt handler when this attribute is present.
4600 An optional argument is supported for the interrupt attribute which allows
4601 the interrupt mode to be described. By default GCC assumes the external
4602 interrupt controller (EIC) mode is in use, this can be explicitly set using
4603 @code{eic}. When interrupts are non-masked then the requested Interrupt
4604 Priority Level (IPL) is copied to the current IPL which has the effect of only
4605 enabling higher priority interrupts. To use vectored interrupt mode use
4606 the argument @code{vector=[sw0|sw1|hw0|hw1|hw2|hw3|hw4|hw5]}, this will change
4607 the behavior of the non-masked interrupt support and GCC will arrange to mask
4608 all interrupts from sw0 up to and including the specified interrupt vector.
4610 You can use the following attributes to modify the behavior
4611 of an interrupt handler:
4613 @item use_shadow_register_set
4614 @cindex @code{use_shadow_register_set} function attribute, MIPS
4615 Assume that the handler uses a shadow register set, instead of
4616 the main general-purpose registers. An optional argument @code{intstack} is
4617 supported to indicate that the shadow register set contains a valid stack
4620 @item keep_interrupts_masked
4621 @cindex @code{keep_interrupts_masked} function attribute, MIPS
4622 Keep interrupts masked for the whole function. Without this attribute,
4623 GCC tries to reenable interrupts for as much of the function as it can.
4625 @item use_debug_exception_return
4626 @cindex @code{use_debug_exception_return} function attribute, MIPS
4627 Return using the @code{deret} instruction. Interrupt handlers that don't
4628 have this attribute return using @code{eret} instead.
4631 You can use any combination of these attributes, as shown below:
4633 void __attribute__ ((interrupt)) v0 ();
4634 void __attribute__ ((interrupt, use_shadow_register_set)) v1 ();
4635 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 ();
4636 void __attribute__ ((interrupt, use_debug_exception_return)) v3 ();
4637 void __attribute__ ((interrupt, use_shadow_register_set,
4638 keep_interrupts_masked)) v4 ();
4639 void __attribute__ ((interrupt, use_shadow_register_set,
4640 use_debug_exception_return)) v5 ();
4641 void __attribute__ ((interrupt, keep_interrupts_masked,
4642 use_debug_exception_return)) v6 ();
4643 void __attribute__ ((interrupt, use_shadow_register_set,
4644 keep_interrupts_masked,
4645 use_debug_exception_return)) v7 ();
4646 void __attribute__ ((interrupt("eic"))) v8 ();
4647 void __attribute__ ((interrupt("vector=hw3"))) v9 ();
4654 @cindex indirect calls, MIPS
4655 @cindex @code{long_call} function attribute, MIPS
4656 @cindex @code{short_call} function attribute, MIPS
4657 @cindex @code{near} function attribute, MIPS
4658 @cindex @code{far} function attribute, MIPS
4659 These attributes specify how a particular function is called on MIPS@.
4660 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
4661 command-line switch. The @code{long_call} and @code{far} attributes are
4662 synonyms, and cause the compiler to always call
4663 the function by first loading its address into a register, and then using
4664 the contents of that register. The @code{short_call} and @code{near}
4665 attributes are synonyms, and have the opposite
4666 effect; they specify that non-PIC calls should be made using the more
4667 efficient @code{jal} instruction.
4671 @cindex @code{mips16} function attribute, MIPS
4672 @cindex @code{nomips16} function attribute, MIPS
4674 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
4675 function attributes to locally select or turn off MIPS16 code generation.
4676 A function with the @code{mips16} attribute is emitted as MIPS16 code,
4677 while MIPS16 code generation is disabled for functions with the
4678 @code{nomips16} attribute. These attributes override the
4679 @option{-mips16} and @option{-mno-mips16} options on the command line
4680 (@pxref{MIPS Options}).
4682 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
4683 preprocessor symbol @code{__mips16} reflects the setting on the command line,
4684 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
4685 may interact badly with some GCC extensions such as @code{__builtin_apply}
4686 (@pxref{Constructing Calls}).
4688 @item micromips, MIPS
4689 @itemx nomicromips, MIPS
4690 @cindex @code{micromips} function attribute
4691 @cindex @code{nomicromips} function attribute
4693 On MIPS targets, you can use the @code{micromips} and @code{nomicromips}
4694 function attributes to locally select or turn off microMIPS code generation.
4695 A function with the @code{micromips} attribute is emitted as microMIPS code,
4696 while microMIPS code generation is disabled for functions with the
4697 @code{nomicromips} attribute. These attributes override the
4698 @option{-mmicromips} and @option{-mno-micromips} options on the command line
4699 (@pxref{MIPS Options}).
4701 When compiling files containing mixed microMIPS and non-microMIPS code, the
4702 preprocessor symbol @code{__mips_micromips} reflects the setting on the
4704 not that within individual functions. Mixed microMIPS and non-microMIPS code
4705 may interact badly with some GCC extensions such as @code{__builtin_apply}
4706 (@pxref{Constructing Calls}).
4709 @cindex @code{nocompression} function attribute, MIPS
4710 On MIPS targets, you can use the @code{nocompression} function attribute
4711 to locally turn off MIPS16 and microMIPS code generation. This attribute
4712 overrides the @option{-mips16} and @option{-mmicromips} options on the
4713 command line (@pxref{MIPS Options}).
4716 @node MSP430 Function Attributes
4717 @subsection MSP430 Function Attributes
4719 These function attributes are supported by the MSP430 back end:
4723 @cindex @code{critical} function attribute, MSP430
4724 Critical functions disable interrupts upon entry and restore the
4725 previous interrupt state upon exit. Critical functions cannot also
4726 have the @code{naked} or @code{reentrant} attributes. They can have
4727 the @code{interrupt} attribute.
4730 @cindex @code{interrupt} function attribute, MSP430
4731 Use this attribute to indicate
4732 that the specified function is an interrupt handler. The compiler generates
4733 function entry and exit sequences suitable for use in an interrupt handler
4734 when this attribute is present.
4736 You can provide an argument to the interrupt
4737 attribute which specifies a name or number. If the argument is a
4738 number it indicates the slot in the interrupt vector table (0 - 31) to
4739 which this handler should be assigned. If the argument is a name it
4740 is treated as a symbolic name for the vector slot. These names should
4741 match up with appropriate entries in the linker script. By default
4742 the names @code{watchdog} for vector 26, @code{nmi} for vector 30 and
4743 @code{reset} for vector 31 are recognized.
4746 @cindex @code{naked} function attribute, MSP430
4747 This attribute allows the compiler to construct the
4748 requisite function declaration, while allowing the body of the
4749 function to be assembly code. The specified function will not have
4750 prologue/epilogue sequences generated by the compiler. Only basic
4751 @code{asm} statements can safely be included in naked functions
4752 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
4753 basic @code{asm} and C code may appear to work, they cannot be
4754 depended upon to work reliably and are not supported.
4757 @cindex @code{reentrant} function attribute, MSP430
4758 Reentrant functions disable interrupts upon entry and enable them
4759 upon exit. Reentrant functions cannot also have the @code{naked}
4760 or @code{critical} attributes. They can have the @code{interrupt}
4764 @cindex @code{wakeup} function attribute, MSP430
4765 This attribute only applies to interrupt functions. It is silently
4766 ignored if applied to a non-interrupt function. A wakeup interrupt
4767 function will rouse the processor from any low-power state that it
4768 might be in when the function exits.
4773 @cindex @code{lower} function attribute, MSP430
4774 @cindex @code{upper} function attribute, MSP430
4775 @cindex @code{either} function attribute, MSP430
4776 On the MSP430 target these attributes can be used to specify whether
4777 the function or variable should be placed into low memory, high
4778 memory, or the placement should be left to the linker to decide. The
4779 attributes are only significant if compiling for the MSP430X
4782 The attributes work in conjunction with a linker script that has been
4783 augmented to specify where to place sections with a @code{.lower} and
4784 a @code{.upper} prefix. So, for example, as well as placing the
4785 @code{.data} section, the script also specifies the placement of a
4786 @code{.lower.data} and a @code{.upper.data} section. The intention
4787 is that @code{lower} sections are placed into a small but easier to
4788 access memory region and the upper sections are placed into a larger, but
4789 slower to access, region.
4791 The @code{either} attribute is special. It tells the linker to place
4792 the object into the corresponding @code{lower} section if there is
4793 room for it. If there is insufficient room then the object is placed
4794 into the corresponding @code{upper} section instead. Note that the
4795 placement algorithm is not very sophisticated. It does not attempt to
4796 find an optimal packing of the @code{lower} sections. It just makes
4797 one pass over the objects and does the best that it can. Using the
4798 @option{-ffunction-sections} and @option{-fdata-sections} command-line
4799 options can help the packing, however, since they produce smaller,
4800 easier to pack regions.
4803 @node NDS32 Function Attributes
4804 @subsection NDS32 Function Attributes
4806 These function attributes are supported by the NDS32 back end:
4810 @cindex @code{exception} function attribute
4811 @cindex exception handler functions, NDS32
4812 Use this attribute on the NDS32 target to indicate that the specified function
4813 is an exception handler. The compiler will generate corresponding sections
4814 for use in an exception handler.
4817 @cindex @code{interrupt} function attribute, NDS32
4818 On NDS32 target, this attribute indicates that the specified function
4819 is an interrupt handler. The compiler generates corresponding sections
4820 for use in an interrupt handler. You can use the following attributes
4821 to modify the behavior:
4824 @cindex @code{nested} function attribute, NDS32
4825 This interrupt service routine is interruptible.
4827 @cindex @code{not_nested} function attribute, NDS32
4828 This interrupt service routine is not interruptible.
4830 @cindex @code{nested_ready} function attribute, NDS32
4831 This interrupt service routine is interruptible after @code{PSW.GIE}
4832 (global interrupt enable) is set. This allows interrupt service routine to
4833 finish some short critical code before enabling interrupts.
4835 @cindex @code{save_all} function attribute, NDS32
4836 The system will help save all registers into stack before entering
4839 @cindex @code{partial_save} function attribute, NDS32
4840 The system will help save caller registers into stack before entering
4845 @cindex @code{naked} function attribute, NDS32
4846 This attribute allows the compiler to construct the
4847 requisite function declaration, while allowing the body of the
4848 function to be assembly code. The specified function will not have
4849 prologue/epilogue sequences generated by the compiler. Only basic
4850 @code{asm} statements can safely be included in naked functions
4851 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
4852 basic @code{asm} and C code may appear to work, they cannot be
4853 depended upon to work reliably and are not supported.
4856 @cindex @code{reset} function attribute, NDS32
4857 @cindex reset handler functions
4858 Use this attribute on the NDS32 target to indicate that the specified function
4859 is a reset handler. The compiler will generate corresponding sections
4860 for use in a reset handler. You can use the following attributes
4861 to provide extra exception handling:
4864 @cindex @code{nmi} function attribute, NDS32
4865 Provide a user-defined function to handle NMI exception.
4867 @cindex @code{warm} function attribute, NDS32
4868 Provide a user-defined function to handle warm reset exception.
4872 @node Nios II Function Attributes
4873 @subsection Nios II Function Attributes
4875 These function attributes are supported by the Nios II back end:
4878 @item target (@var{options})
4879 @cindex @code{target} function attribute
4880 As discussed in @ref{Common Function Attributes}, this attribute
4881 allows specification of target-specific compilation options.
4883 When compiling for Nios II, the following options are allowed:
4886 @item custom-@var{insn}=@var{N}
4887 @itemx no-custom-@var{insn}
4888 @cindex @code{target("custom-@var{insn}=@var{N}")} function attribute, Nios II
4889 @cindex @code{target("no-custom-@var{insn}")} function attribute, Nios II
4890 Each @samp{custom-@var{insn}=@var{N}} attribute locally enables use of a
4891 custom instruction with encoding @var{N} when generating code that uses
4892 @var{insn}. Similarly, @samp{no-custom-@var{insn}} locally inhibits use of
4893 the custom instruction @var{insn}.
4894 These target attributes correspond to the
4895 @option{-mcustom-@var{insn}=@var{N}} and @option{-mno-custom-@var{insn}}
4896 command-line options, and support the same set of @var{insn} keywords.
4897 @xref{Nios II Options}, for more information.
4899 @item custom-fpu-cfg=@var{name}
4900 @cindex @code{target("custom-fpu-cfg=@var{name}")} function attribute, Nios II
4901 This attribute corresponds to the @option{-mcustom-fpu-cfg=@var{name}}
4902 command-line option, to select a predefined set of custom instructions
4904 @xref{Nios II Options}, for more information.
4908 @node Nvidia PTX Function Attributes
4909 @subsection Nvidia PTX Function Attributes
4911 These function attributes are supported by the Nvidia PTX back end:
4915 @cindex @code{kernel} attribute, Nvidia PTX
4916 This attribute indicates that the corresponding function should be compiled
4917 as a kernel function, which can be invoked from the host via the CUDA RT
4919 By default functions are only callable only from other PTX functions.
4921 Kernel functions must have @code{void} return type.
4924 @node PowerPC Function Attributes
4925 @subsection PowerPC Function Attributes
4927 These function attributes are supported by the PowerPC back end:
4932 @cindex indirect calls, PowerPC
4933 @cindex @code{longcall} function attribute, PowerPC
4934 @cindex @code{shortcall} function attribute, PowerPC
4935 The @code{longcall} attribute
4936 indicates that the function might be far away from the call site and
4937 require a different (more expensive) calling sequence. The
4938 @code{shortcall} attribute indicates that the function is always close
4939 enough for the shorter calling sequence to be used. These attributes
4940 override both the @option{-mlongcall} switch and
4941 the @code{#pragma longcall} setting.
4943 @xref{RS/6000 and PowerPC Options}, for more information on whether long
4944 calls are necessary.
4946 @item target (@var{options})
4947 @cindex @code{target} function attribute
4948 As discussed in @ref{Common Function Attributes}, this attribute
4949 allows specification of target-specific compilation options.
4951 On the PowerPC, the following options are allowed:
4956 @cindex @code{target("altivec")} function attribute, PowerPC
4957 Generate code that uses (does not use) AltiVec instructions. In
4958 32-bit code, you cannot enable AltiVec instructions unless
4959 @option{-mabi=altivec} is used on the command line.
4963 @cindex @code{target("cmpb")} function attribute, PowerPC
4964 Generate code that uses (does not use) the compare bytes instruction
4965 implemented on the POWER6 processor and other processors that support
4966 the PowerPC V2.05 architecture.
4970 @cindex @code{target("dlmzb")} function attribute, PowerPC
4971 Generate code that uses (does not use) the string-search @samp{dlmzb}
4972 instruction on the IBM 405, 440, 464 and 476 processors. This instruction is
4973 generated by default when targeting those processors.
4977 @cindex @code{target("fprnd")} function attribute, PowerPC
4978 Generate code that uses (does not use) the FP round to integer
4979 instructions implemented on the POWER5+ processor and other processors
4980 that support the PowerPC V2.03 architecture.
4984 @cindex @code{target("hard-dfp")} function attribute, PowerPC
4985 Generate code that uses (does not use) the decimal floating-point
4986 instructions implemented on some POWER processors.
4990 @cindex @code{target("isel")} function attribute, PowerPC
4991 Generate code that uses (does not use) ISEL instruction.
4995 @cindex @code{target("mfcrf")} function attribute, PowerPC
4996 Generate code that uses (does not use) the move from condition
4997 register field instruction implemented on the POWER4 processor and
4998 other processors that support the PowerPC V2.01 architecture.
5002 @cindex @code{target("mfpgpr")} function attribute, PowerPC
5003 Generate code that uses (does not use) the FP move to/from general
5004 purpose register instructions implemented on the POWER6X processor and
5005 other processors that support the extended PowerPC V2.05 architecture.
5009 @cindex @code{target("mulhw")} function attribute, PowerPC
5010 Generate code that uses (does not use) the half-word multiply and
5011 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors.
5012 These instructions are generated by default when targeting those
5017 @cindex @code{target("multiple")} function attribute, PowerPC
5018 Generate code that uses (does not use) the load multiple word
5019 instructions and the store multiple word instructions.
5023 @cindex @code{target("update")} function attribute, PowerPC
5024 Generate code that uses (does not use) the load or store instructions
5025 that update the base register to the address of the calculated memory
5030 @cindex @code{target("popcntb")} function attribute, PowerPC
5031 Generate code that uses (does not use) the popcount and double-precision
5032 FP reciprocal estimate instruction implemented on the POWER5
5033 processor and other processors that support the PowerPC V2.02
5038 @cindex @code{target("popcntd")} function attribute, PowerPC
5039 Generate code that uses (does not use) the popcount instruction
5040 implemented on the POWER7 processor and other processors that support
5041 the PowerPC V2.06 architecture.
5043 @item powerpc-gfxopt
5044 @itemx no-powerpc-gfxopt
5045 @cindex @code{target("powerpc-gfxopt")} function attribute, PowerPC
5046 Generate code that uses (does not use) the optional PowerPC
5047 architecture instructions in the Graphics group, including
5048 floating-point select.
5051 @itemx no-powerpc-gpopt
5052 @cindex @code{target("powerpc-gpopt")} function attribute, PowerPC
5053 Generate code that uses (does not use) the optional PowerPC
5054 architecture instructions in the General Purpose group, including
5055 floating-point square root.
5057 @item recip-precision
5058 @itemx no-recip-precision
5059 @cindex @code{target("recip-precision")} function attribute, PowerPC
5060 Assume (do not assume) that the reciprocal estimate instructions
5061 provide higher-precision estimates than is mandated by the PowerPC
5066 @cindex @code{target("string")} function attribute, PowerPC
5067 Generate code that uses (does not use) the load string instructions
5068 and the store string word instructions to save multiple registers and
5069 do small block moves.
5073 @cindex @code{target("vsx")} function attribute, PowerPC
5074 Generate code that uses (does not use) vector/scalar (VSX)
5075 instructions, and also enable the use of built-in functions that allow
5076 more direct access to the VSX instruction set. In 32-bit code, you
5077 cannot enable VSX or AltiVec instructions unless
5078 @option{-mabi=altivec} is used on the command line.
5082 @cindex @code{target("friz")} function attribute, PowerPC
5083 Generate (do not generate) the @code{friz} instruction when the
5084 @option{-funsafe-math-optimizations} option is used to optimize
5085 rounding a floating-point value to 64-bit integer and back to floating
5086 point. The @code{friz} instruction does not return the same value if
5087 the floating-point number is too large to fit in an integer.
5089 @item avoid-indexed-addresses
5090 @itemx no-avoid-indexed-addresses
5091 @cindex @code{target("avoid-indexed-addresses")} function attribute, PowerPC
5092 Generate code that tries to avoid (not avoid) the use of indexed load
5093 or store instructions.
5097 @cindex @code{target("paired")} function attribute, PowerPC
5098 Generate code that uses (does not use) the generation of PAIRED simd
5103 @cindex @code{target("longcall")} function attribute, PowerPC
5104 Generate code that assumes (does not assume) that all calls are far
5105 away so that a longer more expensive calling sequence is required.
5108 @cindex @code{target("cpu=@var{CPU}")} function attribute, PowerPC
5109 Specify the architecture to generate code for when compiling the
5110 function. If you select the @code{target("cpu=power7")} attribute when
5111 generating 32-bit code, VSX and AltiVec instructions are not generated
5112 unless you use the @option{-mabi=altivec} option on the command line.
5114 @item tune=@var{TUNE}
5115 @cindex @code{target("tune=@var{TUNE}")} function attribute, PowerPC
5116 Specify the architecture to tune for when compiling the function. If
5117 you do not specify the @code{target("tune=@var{TUNE}")} attribute and
5118 you do specify the @code{target("cpu=@var{CPU}")} attribute,
5119 compilation tunes for the @var{CPU} architecture, and not the
5120 default tuning specified on the command line.
5123 On the PowerPC, the inliner does not inline a
5124 function that has different target options than the caller, unless the
5125 callee has a subset of the target options of the caller.
5128 @node RISC-V Function Attributes
5129 @subsection RISC-V Function Attributes
5131 These function attributes are supported by the RISC-V back end:
5135 @cindex @code{naked} function attribute, RISC-V
5136 This attribute allows the compiler to construct the
5137 requisite function declaration, while allowing the body of the
5138 function to be assembly code. The specified function will not have
5139 prologue/epilogue sequences generated by the compiler. Only basic
5140 @code{asm} statements can safely be included in naked functions
5141 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
5142 basic @code{asm} and C code may appear to work, they cannot be
5143 depended upon to work reliably and are not supported.
5146 @cindex @code{interrupt} function attribute, RISC-V
5147 Use this attribute to indicate that the specified function is an interrupt
5148 handler. The compiler generates function entry and exit sequences suitable
5149 for use in an interrupt handler when this attribute is present.
5151 You can specify the kind of interrupt to be handled by adding an optional
5152 parameter to the interrupt attribute like this:
5155 void f (void) __attribute__ ((interrupt ("user")));
5158 Permissible values for this parameter are @code{user}, @code{supervisor},
5159 and @code{machine}. If there is no parameter, then it defaults to
5163 @node RL78 Function Attributes
5164 @subsection RL78 Function Attributes
5166 These function attributes are supported by the RL78 back end:
5170 @itemx brk_interrupt
5171 @cindex @code{interrupt} function attribute, RL78
5172 @cindex @code{brk_interrupt} function attribute, RL78
5173 These attributes indicate
5174 that the specified function is an interrupt handler. The compiler generates
5175 function entry and exit sequences suitable for use in an interrupt handler
5176 when this attribute is present.
5178 Use @code{brk_interrupt} instead of @code{interrupt} for
5179 handlers intended to be used with the @code{BRK} opcode (i.e.@: those
5180 that must end with @code{RETB} instead of @code{RETI}).
5183 @cindex @code{naked} function attribute, RL78
5184 This attribute allows the compiler to construct the
5185 requisite function declaration, while allowing the body of the
5186 function to be assembly code. The specified function will not have
5187 prologue/epilogue sequences generated by the compiler. Only basic
5188 @code{asm} statements can safely be included in naked functions
5189 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
5190 basic @code{asm} and C code may appear to work, they cannot be
5191 depended upon to work reliably and are not supported.
5194 @node RX Function Attributes
5195 @subsection RX Function Attributes
5197 These function attributes are supported by the RX back end:
5200 @item fast_interrupt
5201 @cindex @code{fast_interrupt} function attribute, RX
5202 Use this attribute on the RX port to indicate that the specified
5203 function is a fast interrupt handler. This is just like the
5204 @code{interrupt} attribute, except that @code{freit} is used to return
5205 instead of @code{reit}.
5208 @cindex @code{interrupt} function attribute, RX
5209 Use this attribute to indicate
5210 that the specified function is an interrupt handler. The compiler generates
5211 function entry and exit sequences suitable for use in an interrupt handler
5212 when this attribute is present.
5214 On RX and RL78 targets, you may specify one or more vector numbers as arguments
5215 to the attribute, as well as naming an alternate table name.
5216 Parameters are handled sequentially, so one handler can be assigned to
5217 multiple entries in multiple tables. One may also pass the magic
5218 string @code{"$default"} which causes the function to be used for any
5219 unfilled slots in the current table.
5221 This example shows a simple assignment of a function to one vector in
5222 the default table (note that preprocessor macros may be used for
5223 chip-specific symbolic vector names):
5225 void __attribute__ ((interrupt (5))) txd1_handler ();
5228 This example assigns a function to two slots in the default table
5229 (using preprocessor macros defined elsewhere) and makes it the default
5230 for the @code{dct} table:
5232 void __attribute__ ((interrupt (RXD1_VECT,RXD2_VECT,"dct","$default")))
5237 @cindex @code{naked} function attribute, RX
5238 This attribute allows the compiler to construct the
5239 requisite function declaration, while allowing the body of the
5240 function to be assembly code. The specified function will not have
5241 prologue/epilogue sequences generated by the compiler. Only basic
5242 @code{asm} statements can safely be included in naked functions
5243 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
5244 basic @code{asm} and C code may appear to work, they cannot be
5245 depended upon to work reliably and are not supported.
5248 @cindex @code{vector} function attribute, RX
5249 This RX attribute is similar to the @code{interrupt} attribute, including its
5250 parameters, but does not make the function an interrupt-handler type
5251 function (i.e. it retains the normal C function calling ABI). See the
5252 @code{interrupt} attribute for a description of its arguments.
5255 @node S/390 Function Attributes
5256 @subsection S/390 Function Attributes
5258 These function attributes are supported on the S/390:
5261 @item hotpatch (@var{halfwords-before-function-label},@var{halfwords-after-function-label})
5262 @cindex @code{hotpatch} function attribute, S/390
5264 On S/390 System z targets, you can use this function attribute to
5265 make GCC generate a ``hot-patching'' function prologue. If the
5266 @option{-mhotpatch=} command-line option is used at the same time,
5267 the @code{hotpatch} attribute takes precedence. The first of the
5268 two arguments specifies the number of halfwords to be added before
5269 the function label. A second argument can be used to specify the
5270 number of halfwords to be added after the function label. For
5271 both arguments the maximum allowed value is 1000000.
5273 If both arguments are zero, hotpatching is disabled.
5275 @item target (@var{options})
5276 @cindex @code{target} function attribute
5277 As discussed in @ref{Common Function Attributes}, this attribute
5278 allows specification of target-specific compilation options.
5280 On S/390, the following options are supported:
5288 @item warn-framesize=
5300 @itemx no-packed-stack
5302 @itemx no-small-exec
5305 @item warn-dynamicstack
5306 @itemx no-warn-dynamicstack
5309 The options work exactly like the S/390 specific command line
5310 options (without the prefix @option{-m}) except that they do not
5311 change any feature macros. For example,
5314 @code{target("no-vx")}
5317 does not undefine the @code{__VEC__} macro.
5320 @node SH Function Attributes
5321 @subsection SH Function Attributes
5323 These function attributes are supported on the SH family of processors:
5326 @item function_vector
5327 @cindex @code{function_vector} function attribute, SH
5328 @cindex calling functions through the function vector on SH2A
5329 On SH2A targets, this attribute declares a function to be called using the
5330 TBR relative addressing mode. The argument to this attribute is the entry
5331 number of the same function in a vector table containing all the TBR
5332 relative addressable functions. For correct operation the TBR must be setup
5333 accordingly to point to the start of the vector table before any functions with
5334 this attribute are invoked. Usually a good place to do the initialization is
5335 the startup routine. The TBR relative vector table can have at max 256 function
5336 entries. The jumps to these functions are generated using a SH2A specific,
5337 non delayed branch instruction JSR/N @@(disp8,TBR). You must use GAS and GLD
5338 from GNU binutils version 2.7 or later for this attribute to work correctly.
5340 In an application, for a function being called once, this attribute
5341 saves at least 8 bytes of code; and if other successive calls are being
5342 made to the same function, it saves 2 bytes of code per each of these
5345 @item interrupt_handler
5346 @cindex @code{interrupt_handler} function attribute, SH
5347 Use this attribute to
5348 indicate that the specified function is an interrupt handler. The compiler
5349 generates function entry and exit sequences suitable for use in an
5350 interrupt handler when this attribute is present.
5352 @item nosave_low_regs
5353 @cindex @code{nosave_low_regs} function attribute, SH
5354 Use this attribute on SH targets to indicate that an @code{interrupt_handler}
5355 function should not save and restore registers R0..R7. This can be used on SH3*
5356 and SH4* targets that have a second R0..R7 register bank for non-reentrant
5360 @cindex @code{renesas} function attribute, SH
5361 On SH targets this attribute specifies that the function or struct follows the
5365 @cindex @code{resbank} function attribute, SH
5366 On the SH2A target, this attribute enables the high-speed register
5367 saving and restoration using a register bank for @code{interrupt_handler}
5368 routines. Saving to the bank is performed automatically after the CPU
5369 accepts an interrupt that uses a register bank.
5371 The nineteen 32-bit registers comprising general register R0 to R14,
5372 control register GBR, and system registers MACH, MACL, and PR and the
5373 vector table address offset are saved into a register bank. Register
5374 banks are stacked in first-in last-out (FILO) sequence. Restoration
5375 from the bank is executed by issuing a RESBANK instruction.
5378 @cindex @code{sp_switch} function attribute, SH
5379 Use this attribute on the SH to indicate an @code{interrupt_handler}
5380 function should switch to an alternate stack. It expects a string
5381 argument that names a global variable holding the address of the
5386 void f () __attribute__ ((interrupt_handler,
5387 sp_switch ("alt_stack")));
5391 @cindex @code{trap_exit} function attribute, SH
5392 Use this attribute on the SH for an @code{interrupt_handler} to return using
5393 @code{trapa} instead of @code{rte}. This attribute expects an integer
5394 argument specifying the trap number to be used.
5397 @cindex @code{trapa_handler} function attribute, SH
5398 On SH targets this function attribute is similar to @code{interrupt_handler}
5399 but it does not save and restore all registers.
5402 @node SPU Function Attributes
5403 @subsection SPU Function Attributes
5405 These function attributes are supported by the SPU back end:
5409 @cindex @code{naked} function attribute, SPU
5410 This attribute allows the compiler to construct the
5411 requisite function declaration, while allowing the body of the
5412 function to be assembly code. The specified function will not have
5413 prologue/epilogue sequences generated by the compiler. Only basic
5414 @code{asm} statements can safely be included in naked functions
5415 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
5416 basic @code{asm} and C code may appear to work, they cannot be
5417 depended upon to work reliably and are not supported.
5420 @node Symbian OS Function Attributes
5421 @subsection Symbian OS Function Attributes
5423 @xref{Microsoft Windows Function Attributes}, for discussion of the
5424 @code{dllexport} and @code{dllimport} attributes.
5426 @node V850 Function Attributes
5427 @subsection V850 Function Attributes
5429 The V850 back end supports these function attributes:
5433 @itemx interrupt_handler
5434 @cindex @code{interrupt} function attribute, V850
5435 @cindex @code{interrupt_handler} function attribute, V850
5436 Use these attributes to indicate
5437 that the specified function is an interrupt handler. The compiler generates
5438 function entry and exit sequences suitable for use in an interrupt handler
5439 when either attribute is present.
5442 @node Visium Function Attributes
5443 @subsection Visium Function Attributes
5445 These function attributes are supported by the Visium back end:
5449 @cindex @code{interrupt} function attribute, Visium
5450 Use this attribute to indicate
5451 that the specified function is an interrupt handler. The compiler generates
5452 function entry and exit sequences suitable for use in an interrupt handler
5453 when this attribute is present.
5456 @node x86 Function Attributes
5457 @subsection x86 Function Attributes
5459 These function attributes are supported by the x86 back end:
5463 @cindex @code{cdecl} function attribute, x86-32
5464 @cindex functions that pop the argument stack on x86-32
5466 On the x86-32 targets, the @code{cdecl} attribute causes the compiler to
5467 assume that the calling function pops off the stack space used to
5468 pass arguments. This is
5469 useful to override the effects of the @option{-mrtd} switch.
5472 @cindex @code{fastcall} function attribute, x86-32
5473 @cindex functions that pop the argument stack on x86-32
5474 On x86-32 targets, the @code{fastcall} attribute causes the compiler to
5475 pass the first argument (if of integral type) in the register ECX and
5476 the second argument (if of integral type) in the register EDX@. Subsequent
5477 and other typed arguments are passed on the stack. The called function
5478 pops the arguments off the stack. If the number of arguments is variable all
5479 arguments are pushed on the stack.
5482 @cindex @code{thiscall} function attribute, x86-32
5483 @cindex functions that pop the argument stack on x86-32
5484 On x86-32 targets, the @code{thiscall} attribute causes the compiler to
5485 pass the first argument (if of integral type) in the register ECX.
5486 Subsequent and other typed arguments are passed on the stack. The called
5487 function pops the arguments off the stack.
5488 If the number of arguments is variable all arguments are pushed on the
5490 The @code{thiscall} attribute is intended for C++ non-static member functions.
5491 As a GCC extension, this calling convention can be used for C functions
5492 and for static member methods.
5496 @cindex @code{ms_abi} function attribute, x86
5497 @cindex @code{sysv_abi} function attribute, x86
5499 On 32-bit and 64-bit x86 targets, you can use an ABI attribute
5500 to indicate which calling convention should be used for a function. The
5501 @code{ms_abi} attribute tells the compiler to use the Microsoft ABI,
5502 while the @code{sysv_abi} attribute tells the compiler to use the ABI
5503 used on GNU/Linux and other systems. The default is to use the Microsoft ABI
5504 when targeting Windows. On all other systems, the default is the x86/AMD ABI.
5506 Note, the @code{ms_abi} attribute for Microsoft Windows 64-bit targets currently
5507 requires the @option{-maccumulate-outgoing-args} option.
5509 @item callee_pop_aggregate_return (@var{number})
5510 @cindex @code{callee_pop_aggregate_return} function attribute, x86
5512 On x86-32 targets, you can use this attribute to control how
5513 aggregates are returned in memory. If the caller is responsible for
5514 popping the hidden pointer together with the rest of the arguments, specify
5515 @var{number} equal to zero. If callee is responsible for popping the
5516 hidden pointer, specify @var{number} equal to one.
5518 The default x86-32 ABI assumes that the callee pops the
5519 stack for hidden pointer. However, on x86-32 Microsoft Windows targets,
5520 the compiler assumes that the
5521 caller pops the stack for hidden pointer.
5523 @item ms_hook_prologue
5524 @cindex @code{ms_hook_prologue} function attribute, x86
5526 On 32-bit and 64-bit x86 targets, you can use
5527 this function attribute to make GCC generate the ``hot-patching'' function
5528 prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2
5532 @cindex @code{naked} function attribute, x86
5533 This attribute allows the compiler to construct the
5534 requisite function declaration, while allowing the body of the
5535 function to be assembly code. The specified function will not have
5536 prologue/epilogue sequences generated by the compiler. Only basic
5537 @code{asm} statements can safely be included in naked functions
5538 (@pxref{Basic Asm}). While using extended @code{asm} or a mixture of
5539 basic @code{asm} and C code may appear to work, they cannot be
5540 depended upon to work reliably and are not supported.
5542 @item regparm (@var{number})
5543 @cindex @code{regparm} function attribute, x86
5544 @cindex functions that are passed arguments in registers on x86-32
5545 On x86-32 targets, the @code{regparm} attribute causes the compiler to
5546 pass arguments number one to @var{number} if they are of integral type
5547 in registers EAX, EDX, and ECX instead of on the stack. Functions that
5548 take a variable number of arguments continue to be passed all of their
5549 arguments on the stack.
5551 Beware that on some ELF systems this attribute is unsuitable for
5552 global functions in shared libraries with lazy binding (which is the
5553 default). Lazy binding sends the first call via resolving code in
5554 the loader, which might assume EAX, EDX and ECX can be clobbered, as
5555 per the standard calling conventions. Solaris 8 is affected by this.
5556 Systems with the GNU C Library version 2.1 or higher
5557 and FreeBSD are believed to be
5558 safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be
5559 disabled with the linker or the loader if desired, to avoid the
5563 @cindex @code{sseregparm} function attribute, x86
5564 On x86-32 targets with SSE support, the @code{sseregparm} attribute
5565 causes the compiler to pass up to 3 floating-point arguments in
5566 SSE registers instead of on the stack. Functions that take a
5567 variable number of arguments continue to pass all of their
5568 floating-point arguments on the stack.
5570 @item force_align_arg_pointer
5571 @cindex @code{force_align_arg_pointer} function attribute, x86
5572 On x86 targets, the @code{force_align_arg_pointer} attribute may be
5573 applied to individual function definitions, generating an alternate
5574 prologue and epilogue that realigns the run-time stack if necessary.
5575 This supports mixing legacy codes that run with a 4-byte aligned stack
5576 with modern codes that keep a 16-byte stack for SSE compatibility.
5579 @cindex @code{stdcall} function attribute, x86-32
5580 @cindex functions that pop the argument stack on x86-32
5581 On x86-32 targets, the @code{stdcall} attribute causes the compiler to
5582 assume that the called function pops off the stack space used to
5583 pass arguments, unless it takes a variable number of arguments.
5585 @item no_caller_saved_registers
5586 @cindex @code{no_caller_saved_registers} function attribute, x86
5587 Use this attribute to indicate that the specified function has no
5588 caller-saved registers. That is, all registers are callee-saved. For
5589 example, this attribute can be used for a function called from an
5590 interrupt handler. The compiler generates proper function entry and
5591 exit sequences to save and restore any modified registers, except for
5592 the EFLAGS register. Since GCC doesn't preserve MPX, SSE, MMX nor x87
5593 states, the GCC option @option{-mgeneral-regs-only} should be used to
5594 compile functions with @code{no_caller_saved_registers} attribute.
5597 @cindex @code{interrupt} function attribute, x86
5598 Use this attribute to indicate that the specified function is an
5599 interrupt handler or an exception handler (depending on parameters passed
5600 to the function, explained further). The compiler generates function
5601 entry and exit sequences suitable for use in an interrupt handler when
5602 this attribute is present. The @code{IRET} instruction, instead of the
5603 @code{RET} instruction, is used to return from interrupt handlers. All
5604 registers, except for the EFLAGS register which is restored by the
5605 @code{IRET} instruction, are preserved by the compiler. Since GCC
5606 doesn't preserve MPX, SSE, MMX nor x87 states, the GCC option
5607 @option{-mgeneral-regs-only} should be used to compile interrupt and
5610 Any interruptible-without-stack-switch code must be compiled with
5611 @option{-mno-red-zone} since interrupt handlers can and will, because
5612 of the hardware design, touch the red zone.
5614 An interrupt handler must be declared with a mandatory pointer
5618 struct interrupt_frame;
5620 __attribute__ ((interrupt))
5622 f (struct interrupt_frame *frame)
5628 and you must define @code{struct interrupt_frame} as described in the
5631 Exception handlers differ from interrupt handlers because the system
5632 pushes an error code on the stack. An exception handler declaration is
5633 similar to that for an interrupt handler, but with a different mandatory
5634 function signature. The compiler arranges to pop the error code off the
5635 stack before the @code{IRET} instruction.
5639 typedef unsigned long long int uword_t;
5641 typedef unsigned int uword_t;
5644 struct interrupt_frame;
5646 __attribute__ ((interrupt))
5648 f (struct interrupt_frame *frame, uword_t error_code)
5654 Exception handlers should only be used for exceptions that push an error
5655 code; you should use an interrupt handler in other cases. The system
5656 will crash if the wrong kind of handler is used.
5658 @item target (@var{options})
5659 @cindex @code{target} function attribute
5660 As discussed in @ref{Common Function Attributes}, this attribute
5661 allows specification of target-specific compilation options.
5663 On the x86, the following options are allowed:
5667 @cindex @code{target("abm")} function attribute, x86
5668 Enable/disable the generation of the advanced bit instructions.
5672 @cindex @code{target("aes")} function attribute, x86
5673 Enable/disable the generation of the AES instructions.
5676 @cindex @code{target("default")} function attribute, x86
5677 @xref{Function Multiversioning}, where it is used to specify the
5678 default function version.
5682 @cindex @code{target("mmx")} function attribute, x86
5683 Enable/disable the generation of the MMX instructions.
5687 @cindex @code{target("pclmul")} function attribute, x86
5688 Enable/disable the generation of the PCLMUL instructions.
5692 @cindex @code{target("popcnt")} function attribute, x86
5693 Enable/disable the generation of the POPCNT instruction.
5697 @cindex @code{target("sse")} function attribute, x86
5698 Enable/disable the generation of the SSE instructions.
5702 @cindex @code{target("sse2")} function attribute, x86
5703 Enable/disable the generation of the SSE2 instructions.
5707 @cindex @code{target("sse3")} function attribute, x86
5708 Enable/disable the generation of the SSE3 instructions.
5712 @cindex @code{target("sse4")} function attribute, x86
5713 Enable/disable the generation of the SSE4 instructions (both SSE4.1
5718 @cindex @code{target("sse4.1")} function attribute, x86
5719 Enable/disable the generation of the sse4.1 instructions.
5723 @cindex @code{target("sse4.2")} function attribute, x86
5724 Enable/disable the generation of the sse4.2 instructions.
5728 @cindex @code{target("sse4a")} function attribute, x86
5729 Enable/disable the generation of the SSE4A instructions.
5733 @cindex @code{target("fma4")} function attribute, x86
5734 Enable/disable the generation of the FMA4 instructions.
5738 @cindex @code{target("xop")} function attribute, x86
5739 Enable/disable the generation of the XOP instructions.
5743 @cindex @code{target("lwp")} function attribute, x86
5744 Enable/disable the generation of the LWP instructions.
5748 @cindex @code{target("ssse3")} function attribute, x86
5749 Enable/disable the generation of the SSSE3 instructions.
5753 @cindex @code{target("cld")} function attribute, x86
5754 Enable/disable the generation of the CLD before string moves.
5756 @item fancy-math-387
5757 @itemx no-fancy-math-387
5758 @cindex @code{target("fancy-math-387")} function attribute, x86
5759 Enable/disable the generation of the @code{sin}, @code{cos}, and
5760 @code{sqrt} instructions on the 387 floating-point unit.
5764 @cindex @code{target("ieee-fp")} function attribute, x86
5765 Enable/disable the generation of floating point that depends on IEEE arithmetic.
5767 @item inline-all-stringops
5768 @itemx no-inline-all-stringops
5769 @cindex @code{target("inline-all-stringops")} function attribute, x86
5770 Enable/disable inlining of string operations.
5772 @item inline-stringops-dynamically
5773 @itemx no-inline-stringops-dynamically
5774 @cindex @code{target("inline-stringops-dynamically")} function attribute, x86
5775 Enable/disable the generation of the inline code to do small string
5776 operations and calling the library routines for large operations.
5778 @item align-stringops
5779 @itemx no-align-stringops
5780 @cindex @code{target("align-stringops")} function attribute, x86
5781 Do/do not align destination of inlined string operations.
5785 @cindex @code{target("recip")} function attribute, x86
5786 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
5787 instructions followed an additional Newton-Raphson step instead of
5788 doing a floating-point division.
5790 @item arch=@var{ARCH}
5791 @cindex @code{target("arch=@var{ARCH}")} function attribute, x86
5792 Specify the architecture to generate code for in compiling the function.
5794 @item tune=@var{TUNE}
5795 @cindex @code{target("tune=@var{TUNE}")} function attribute, x86
5796 Specify the architecture to tune for in compiling the function.
5798 @item fpmath=@var{FPMATH}
5799 @cindex @code{target("fpmath=@var{FPMATH}")} function attribute, x86
5800 Specify which floating-point unit to use. You must specify the
5801 @code{target("fpmath=sse,387")} option as
5802 @code{target("fpmath=sse+387")} because the comma would separate
5805 @item indirect_branch("@var{choice}")
5806 @cindex @code{indirect_branch} function attribute, x86
5807 On x86 targets, the @code{indirect_branch} attribute causes the compiler
5808 to convert indirect call and jump with @var{choice}. @samp{keep}
5809 keeps indirect call and jump unmodified. @samp{thunk} converts indirect
5810 call and jump to call and return thunk. @samp{thunk-inline} converts
5811 indirect call and jump to inlined call and return thunk.
5812 @samp{thunk-extern} converts indirect call and jump to external call
5813 and return thunk provided in a separate object file.
5815 @item function_return("@var{choice}")
5816 @cindex @code{function_return} function attribute, x86
5817 On x86 targets, the @code{function_return} attribute causes the compiler
5818 to convert function return with @var{choice}. @samp{keep} keeps function
5819 return unmodified. @samp{thunk} converts function return to call and
5820 return thunk. @samp{thunk-inline} converts function return to inlined
5821 call and return thunk. @samp{thunk-extern} converts function return to
5822 external call and return thunk provided in a separate object file.
5825 @cindex @code{nocf_check} function attribute
5826 The @code{nocf_check} attribute on a function is used to inform the
5827 compiler that the function's prologue should not be instrumented when
5828 compiled with the @option{-fcf-protection=branch} option. The
5829 compiler assumes that the function's address is a valid target for a
5830 control-flow transfer.
5832 The @code{nocf_check} attribute on a type of pointer to function is
5833 used to inform the compiler that a call through the pointer should
5834 not be instrumented when compiled with the
5835 @option{-fcf-protection=branch} option. The compiler assumes
5836 that the function's address from the pointer is a valid target for
5837 a control-flow transfer. A direct function call through a function
5838 name is assumed to be a safe call thus direct calls are not
5839 instrumented by the compiler.
5841 The @code{nocf_check} attribute is applied to an object's type.
5842 In case of assignment of a function address or a function pointer to
5843 another pointer, the attribute is not carried over from the right-hand
5844 object's type; the type of left-hand object stays unchanged. The
5845 compiler checks for @code{nocf_check} attribute mismatch and reports
5846 a warning in case of mismatch.
5850 int foo (void) __attribute__(nocf_check);
5851 void (*foo1)(void) __attribute__(nocf_check);
5854 /* foo's address is assumed to be valid. */
5858 /* This call site is not checked for control-flow
5862 /* A warning is issued about attribute mismatch. */
5865 /* This call site is still not checked. */
5868 /* This call site is checked. */
5871 /* A warning is issued about attribute mismatch. */
5874 /* This call site is still checked. */
5883 On the x86, the inliner does not inline a
5884 function that has different target options than the caller, unless the
5885 callee has a subset of the target options of the caller. For example
5886 a function declared with @code{target("sse3")} can inline a function
5887 with @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}.
5890 @node Xstormy16 Function Attributes
5891 @subsection Xstormy16 Function Attributes
5893 These function attributes are supported by the Xstormy16 back end:
5897 @cindex @code{interrupt} function attribute, Xstormy16
5898 Use this attribute to indicate
5899 that the specified function is an interrupt handler. The compiler generates
5900 function entry and exit sequences suitable for use in an interrupt handler
5901 when this attribute is present.
5904 @node Variable Attributes
5905 @section Specifying Attributes of Variables
5906 @cindex attribute of variables
5907 @cindex variable attributes
5909 The keyword @code{__attribute__} allows you to specify special
5910 attributes of variables or structure fields. This keyword is followed
5911 by an attribute specification inside double parentheses. Some
5912 attributes are currently defined generically for variables.
5913 Other attributes are defined for variables on particular target
5914 systems. Other attributes are available for functions
5915 (@pxref{Function Attributes}), labels (@pxref{Label Attributes}),
5916 enumerators (@pxref{Enumerator Attributes}), statements
5917 (@pxref{Statement Attributes}), and for types (@pxref{Type Attributes}).
5918 Other front ends might define more attributes
5919 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
5921 @xref{Attribute Syntax}, for details of the exact syntax for using
5925 * Common Variable Attributes::
5926 * ARC Variable Attributes::
5927 * AVR Variable Attributes::
5928 * Blackfin Variable Attributes::
5929 * H8/300 Variable Attributes::
5930 * IA-64 Variable Attributes::
5931 * M32R/D Variable Attributes::
5932 * MeP Variable Attributes::
5933 * Microsoft Windows Variable Attributes::
5934 * MSP430 Variable Attributes::
5935 * Nvidia PTX Variable Attributes::
5936 * PowerPC Variable Attributes::
5937 * RL78 Variable Attributes::
5938 * SPU Variable Attributes::
5939 * V850 Variable Attributes::
5940 * x86 Variable Attributes::
5941 * Xstormy16 Variable Attributes::
5944 @node Common Variable Attributes
5945 @subsection Common Variable Attributes
5947 The following attributes are supported on most targets.
5950 @cindex @code{aligned} variable attribute
5951 @item aligned (@var{alignment})
5952 This attribute specifies a minimum alignment for the variable or
5953 structure field, measured in bytes. For example, the declaration:
5956 int x __attribute__ ((aligned (16))) = 0;
5960 causes the compiler to allocate the global variable @code{x} on a
5961 16-byte boundary. On a 68040, this could be used in conjunction with
5962 an @code{asm} expression to access the @code{move16} instruction which
5963 requires 16-byte aligned operands.
5965 You can also specify the alignment of structure fields. For example, to
5966 create a double-word aligned @code{int} pair, you could write:
5969 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
5973 This is an alternative to creating a union with a @code{double} member,
5974 which forces the union to be double-word aligned.
5976 As in the preceding examples, you can explicitly specify the alignment
5977 (in bytes) that you wish the compiler to use for a given variable or
5978 structure field. Alternatively, you can leave out the alignment factor
5979 and just ask the compiler to align a variable or field to the
5980 default alignment for the target architecture you are compiling for.
5981 The default alignment is sufficient for all scalar types, but may not be
5982 enough for all vector types on a target that supports vector operations.
5983 The default alignment is fixed for a particular target ABI.
5985 GCC also provides a target specific macro @code{__BIGGEST_ALIGNMENT__},
5986 which is the largest alignment ever used for any data type on the
5987 target machine you are compiling for. For example, you could write:
5990 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__)));
5993 The compiler automatically sets the alignment for the declared
5994 variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can
5995 often make copy operations more efficient, because the compiler can
5996 use whatever instructions copy the biggest chunks of memory when
5997 performing copies to or from the variables or fields that you have
5998 aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__}
5999 may change depending on command-line options.
6001 When used on a struct, or struct member, the @code{aligned} attribute can
6002 only increase the alignment; in order to decrease it, the @code{packed}
6003 attribute must be specified as well. When used as part of a typedef, the
6004 @code{aligned} attribute can both increase and decrease alignment, and
6005 specifying the @code{packed} attribute generates a warning.
6007 Note that the effectiveness of @code{aligned} attributes may be limited
6008 by inherent limitations in your linker. On many systems, the linker is
6009 only able to arrange for variables to be aligned up to a certain maximum
6010 alignment. (For some linkers, the maximum supported alignment may
6011 be very very small.) If your linker is only able to align variables
6012 up to a maximum of 8-byte alignment, then specifying @code{aligned(16)}
6013 in an @code{__attribute__} still only provides you with 8-byte
6014 alignment. See your linker documentation for further information.
6016 The @code{aligned} attribute can also be used for functions
6017 (@pxref{Common Function Attributes}.)
6019 @cindex @code{warn_if_not_aligned} variable attribute
6020 @item warn_if_not_aligned (@var{alignment})
6021 This attribute specifies a threshold for the structure field, measured
6022 in bytes. If the structure field is aligned below the threshold, a
6023 warning will be issued. For example, the declaration:
6030 unsigned long long x __attribute__((warn_if_not_aligned(16)));
6035 causes the compiler to issue an warning on @code{struct foo}, like
6036 @samp{warning: alignment 8 of 'struct foo' is less than 16}.
6037 The compiler also issues a warning, like @samp{warning: 'x' offset
6038 8 in 'struct foo' isn't aligned to 16}, when the structure field has
6039 the misaligned offset:
6046 unsigned long long x __attribute__((warn_if_not_aligned(16)));
6047 @} __attribute__((aligned(16)));
6050 This warning can be disabled by @option{-Wno-if-not-aligned}.
6051 The @code{warn_if_not_aligned} attribute can also be used for types
6052 (@pxref{Common Type Attributes}.)
6054 @item cleanup (@var{cleanup_function})
6055 @cindex @code{cleanup} variable attribute
6056 The @code{cleanup} attribute runs a function when the variable goes
6057 out of scope. This attribute can only be applied to auto function
6058 scope variables; it may not be applied to parameters or variables
6059 with static storage duration. The function must take one parameter,
6060 a pointer to a type compatible with the variable. The return value
6061 of the function (if any) is ignored.
6063 If @option{-fexceptions} is enabled, then @var{cleanup_function}
6064 is run during the stack unwinding that happens during the
6065 processing of the exception. Note that the @code{cleanup} attribute
6066 does not allow the exception to be caught, only to perform an action.
6067 It is undefined what happens if @var{cleanup_function} does not
6072 @cindex @code{common} variable attribute
6073 @cindex @code{nocommon} variable attribute
6076 The @code{common} attribute requests GCC to place a variable in
6077 ``common'' storage. The @code{nocommon} attribute requests the
6078 opposite---to allocate space for it directly.
6080 These attributes override the default chosen by the
6081 @option{-fno-common} and @option{-fcommon} flags respectively.
6084 @itemx deprecated (@var{msg})
6085 @cindex @code{deprecated} variable attribute
6086 The @code{deprecated} attribute results in a warning if the variable
6087 is used anywhere in the source file. This is useful when identifying
6088 variables that are expected to be removed in a future version of a
6089 program. The warning also includes the location of the declaration
6090 of the deprecated variable, to enable users to easily find further
6091 information about why the variable is deprecated, or what they should
6092 do instead. Note that the warning only occurs for uses:
6095 extern int old_var __attribute__ ((deprecated));
6097 int new_fn () @{ return old_var; @}
6101 results in a warning on line 3 but not line 2. The optional @var{msg}
6102 argument, which must be a string, is printed in the warning if
6105 The @code{deprecated} attribute can also be used for functions and
6106 types (@pxref{Common Function Attributes},
6107 @pxref{Common Type Attributes}).
6110 @cindex @code{nonstring} variable attribute
6111 The @code{nonstring} variable attribute specifies that an object or member
6112 declaration with type array of @code{char}, @code{signed char}, or
6113 @code{unsigned char}, or pointer to such a type is intended to store
6114 character arrays that do not necessarily contain a terminating @code{NUL}.
6115 This is useful in detecting uses of such arrays or pointers with functions
6116 that expect @code{NUL}-terminated strings, and to avoid warnings when such
6117 an array or pointer is used as an argument to a bounded string manipulation
6118 function such as @code{strncpy}. For example, without the attribute, GCC
6119 will issue a warning for the @code{strncpy} call below because it may
6120 truncate the copy without appending the terminating @code{NUL} character.
6121 Using the attribute makes it possible to suppress the warning. However,
6122 when the array is declared with the attribute the call to @code{strlen} is
6123 diagnosed because when the array doesn't contain a @code{NUL}-terminated
6124 string the call is undefined. To copy, compare, of search non-string
6125 character arrays use the @code{memcpy}, @code{memcmp}, @code{memchr},
6126 and other functions that operate on arrays of bytes. In addition,
6127 calling @code{strnlen} and @code{strndup} with such arrays is safe
6128 provided a suitable bound is specified, and not diagnosed.
6133 char name [32] __attribute__ ((nonstring));
6136 int f (struct Data *pd, const char *s)
6138 strncpy (pd->name, s, sizeof pd->name);
6140 return strlen (pd->name); // unsafe, gets a warning
6144 @item mode (@var{mode})
6145 @cindex @code{mode} variable attribute
6146 This attribute specifies the data type for the declaration---whichever
6147 type corresponds to the mode @var{mode}. This in effect lets you
6148 request an integer or floating-point type according to its width.
6150 @xref{Machine Modes,,, gccint, GNU Compiler Collection (GCC) Internals},
6151 for a list of the possible keywords for @var{mode}.
6152 You may also specify a mode of @code{byte} or @code{__byte__} to
6153 indicate the mode corresponding to a one-byte integer, @code{word} or
6154 @code{__word__} for the mode of a one-word integer, and @code{pointer}
6155 or @code{__pointer__} for the mode used to represent pointers.
6158 @cindex @code{packed} variable attribute
6159 The @code{packed} attribute specifies that a variable or structure field
6160 should have the smallest possible alignment---one byte for a variable,
6161 and one bit for a field, unless you specify a larger value with the
6162 @code{aligned} attribute.
6164 Here is a structure in which the field @code{x} is packed, so that it
6165 immediately follows @code{a}:
6171 int x[2] __attribute__ ((packed));
6175 @emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the
6176 @code{packed} attribute on bit-fields of type @code{char}. This has
6177 been fixed in GCC 4.4 but the change can lead to differences in the
6178 structure layout. See the documentation of
6179 @option{-Wpacked-bitfield-compat} for more information.
6181 @item section ("@var{section-name}")
6182 @cindex @code{section} variable attribute
6183 Normally, the compiler places the objects it generates in sections like
6184 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
6185 or you need certain particular variables to appear in special sections,
6186 for example to map to special hardware. The @code{section}
6187 attribute specifies that a variable (or function) lives in a particular
6188 section. For example, this small program uses several specific section names:
6191 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
6192 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
6193 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
6194 int init_data __attribute__ ((section ("INITDATA")));
6198 /* @r{Initialize stack pointer} */
6199 init_sp (stack + sizeof (stack));
6201 /* @r{Initialize initialized data} */
6202 memcpy (&init_data, &data, &edata - &data);
6204 /* @r{Turn on the serial ports} */
6211 Use the @code{section} attribute with
6212 @emph{global} variables and not @emph{local} variables,
6213 as shown in the example.
6215 You may use the @code{section} attribute with initialized or
6216 uninitialized global variables but the linker requires
6217 each object be defined once, with the exception that uninitialized
6218 variables tentatively go in the @code{common} (or @code{bss}) section
6219 and can be multiply ``defined''. Using the @code{section} attribute
6220 changes what section the variable goes into and may cause the
6221 linker to issue an error if an uninitialized variable has multiple
6222 definitions. You can force a variable to be initialized with the
6223 @option{-fno-common} flag or the @code{nocommon} attribute.
6225 Some file formats do not support arbitrary sections so the @code{section}
6226 attribute is not available on all platforms.
6227 If you need to map the entire contents of a module to a particular
6228 section, consider using the facilities of the linker instead.
6230 @item tls_model ("@var{tls_model}")
6231 @cindex @code{tls_model} variable attribute
6232 The @code{tls_model} attribute sets thread-local storage model
6233 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
6234 overriding @option{-ftls-model=} command-line switch on a per-variable
6236 The @var{tls_model} argument should be one of @code{global-dynamic},
6237 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
6239 Not all targets support this attribute.
6242 @cindex @code{unused} variable attribute
6243 This attribute, attached to a variable, means that the variable is meant
6244 to be possibly unused. GCC does not produce a warning for this
6248 @cindex @code{used} variable attribute
6249 This attribute, attached to a variable with static storage, means that
6250 the variable must be emitted even if it appears that the variable is not
6253 When applied to a static data member of a C++ class template, the
6254 attribute also means that the member is instantiated if the
6255 class itself is instantiated.
6257 @item vector_size (@var{bytes})
6258 @cindex @code{vector_size} variable attribute
6259 This attribute specifies the vector size for the variable, measured in
6260 bytes. For example, the declaration:
6263 int foo __attribute__ ((vector_size (16)));
6267 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
6268 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
6269 4 units of 4 bytes), the corresponding mode of @code{foo} is V4SI@.
6271 This attribute is only applicable to integral and float scalars,
6272 although arrays, pointers, and function return values are allowed in
6273 conjunction with this construct.
6275 Aggregates with this attribute are invalid, even if they are of the same
6276 size as a corresponding scalar. For example, the declaration:
6279 struct S @{ int a; @};
6280 struct S __attribute__ ((vector_size (16))) foo;
6284 is invalid even if the size of the structure is the same as the size of
6287 @item visibility ("@var{visibility_type}")
6288 @cindex @code{visibility} variable attribute
6289 This attribute affects the linkage of the declaration to which it is attached.
6290 The @code{visibility} attribute is described in
6291 @ref{Common Function Attributes}.
6294 @cindex @code{weak} variable attribute
6295 The @code{weak} attribute is described in
6296 @ref{Common Function Attributes}.
6300 @node ARC Variable Attributes
6301 @subsection ARC Variable Attributes
6305 @cindex @code{aux} variable attribute, ARC
6306 The @code{aux} attribute is used to directly access the ARC's
6307 auxiliary register space from C. The auxilirary register number is
6308 given via attribute argument.
6312 @node AVR Variable Attributes
6313 @subsection AVR Variable Attributes
6317 @cindex @code{progmem} variable attribute, AVR
6318 The @code{progmem} attribute is used on the AVR to place read-only
6319 data in the non-volatile program memory (flash). The @code{progmem}
6320 attribute accomplishes this by putting respective variables into a
6321 section whose name starts with @code{.progmem}.
6323 This attribute works similar to the @code{section} attribute
6324 but adds additional checking.
6327 @item @bullet{}@tie{} Ordinary AVR cores with 32 general purpose registers:
6328 @code{progmem} affects the location
6329 of the data but not how this data is accessed.
6330 In order to read data located with the @code{progmem} attribute
6331 (inline) assembler must be used.
6333 /* Use custom macros from @w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC}} */
6334 #include <avr/pgmspace.h>
6336 /* Locate var in flash memory */
6337 const int var[2] PROGMEM = @{ 1, 2 @};
6339 int read_var (int i)
6341 /* Access var[] by accessor macro from avr/pgmspace.h */
6342 return (int) pgm_read_word (& var[i]);
6346 AVR is a Harvard architecture processor and data and read-only data
6347 normally resides in the data memory (RAM).
6349 See also the @ref{AVR Named Address Spaces} section for
6350 an alternate way to locate and access data in flash memory.
6352 @item @bullet{}@tie{} AVR cores with flash memory visible in the RAM address range:
6353 On such devices, there is no need for attribute @code{progmem} or
6354 @ref{AVR Named Address Spaces,,@code{__flash}} qualifier at all.
6355 Just use standard C / C++. The compiler will generate @code{LD*}
6356 instructions. As flash memory is visible in the RAM address range,
6357 and the default linker script does @emph{not} locate @code{.rodata} in
6358 RAM, no special features are needed in order not to waste RAM for
6359 read-only data or to read from flash. You might even get slightly better
6361 avoiding @code{progmem} and @code{__flash}. This applies to devices from
6362 families @code{avrtiny} and @code{avrxmega3}, see @ref{AVR Options} for
6365 @item @bullet{}@tie{}Reduced AVR Tiny cores like ATtiny40:
6366 The compiler adds @code{0x4000}
6367 to the addresses of objects and declarations in @code{progmem} and locates
6368 the objects in flash memory, namely in section @code{.progmem.data}.
6369 The offset is needed because the flash memory is visible in the RAM
6370 address space starting at address @code{0x4000}.
6372 Data in @code{progmem} can be accessed by means of ordinary C@tie{}code,
6373 no special functions or macros are needed.
6376 /* var is located in flash memory */
6377 extern const int var[2] __attribute__((progmem));
6379 int read_var (int i)
6385 Please notice that on these devices, there is no need for @code{progmem}
6391 @itemx io (@var{addr})
6392 @cindex @code{io} variable attribute, AVR
6393 Variables with the @code{io} attribute are used to address
6394 memory-mapped peripherals in the io address range.
6395 If an address is specified, the variable
6396 is assigned that address, and the value is interpreted as an
6397 address in the data address space.
6401 volatile int porta __attribute__((io (0x22)));
6404 The address specified in the address in the data address range.
6406 Otherwise, the variable it is not assigned an address, but the
6407 compiler will still use in/out instructions where applicable,
6408 assuming some other module assigns an address in the io address range.
6412 extern volatile int porta __attribute__((io));
6416 @itemx io_low (@var{addr})
6417 @cindex @code{io_low} variable attribute, AVR
6418 This is like the @code{io} attribute, but additionally it informs the
6419 compiler that the object lies in the lower half of the I/O area,
6420 allowing the use of @code{cbi}, @code{sbi}, @code{sbic} and @code{sbis}
6424 @itemx address (@var{addr})
6425 @cindex @code{address} variable attribute, AVR
6426 Variables with the @code{address} attribute are used to address
6427 memory-mapped peripherals that may lie outside the io address range.
6430 volatile int porta __attribute__((address (0x600)));
6434 @cindex @code{absdata} variable attribute, AVR
6435 Variables in static storage and with the @code{absdata} attribute can
6436 be accessed by the @code{LDS} and @code{STS} instructions which take
6441 This attribute is only supported for the reduced AVR Tiny core
6445 You must make sure that respective data is located in the
6446 address range @code{0x40}@dots{}@code{0xbf} accessible by
6447 @code{LDS} and @code{STS}. One way to achieve this as an
6448 appropriate linker description file.
6451 If the location does not fit the address range of @code{LDS}
6452 and @code{STS}, there is currently (Binutils 2.26) just an unspecific
6455 @code{module.c:(.text+0x1c): warning: internal error: out of range error}
6460 See also the @option{-mabsdata} @ref{AVR Options,command-line option}.
6464 @node Blackfin Variable Attributes
6465 @subsection Blackfin Variable Attributes
6467 Three attributes are currently defined for the Blackfin.
6473 @cindex @code{l1_data} variable attribute, Blackfin
6474 @cindex @code{l1_data_A} variable attribute, Blackfin
6475 @cindex @code{l1_data_B} variable attribute, Blackfin
6476 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
6477 Variables with @code{l1_data} attribute are put into the specific section
6478 named @code{.l1.data}. Those with @code{l1_data_A} attribute are put into
6479 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
6480 attribute are put into the specific section named @code{.l1.data.B}.
6483 @cindex @code{l2} variable attribute, Blackfin
6484 Use this attribute on the Blackfin to place the variable into L2 SRAM.
6485 Variables with @code{l2} attribute are put into the specific section
6486 named @code{.l2.data}.
6489 @node H8/300 Variable Attributes
6490 @subsection H8/300 Variable Attributes
6492 These variable attributes are available for H8/300 targets:
6496 @cindex @code{eightbit_data} variable attribute, H8/300
6497 @cindex eight-bit data on the H8/300, H8/300H, and H8S
6498 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
6499 variable should be placed into the eight-bit data section.
6500 The compiler generates more efficient code for certain operations
6501 on data in the eight-bit data area. Note the eight-bit data area is limited to
6504 You must use GAS and GLD from GNU binutils version 2.7 or later for
6505 this attribute to work correctly.
6508 @cindex @code{tiny_data} variable attribute, H8/300
6509 @cindex tiny data section on the H8/300H and H8S
6510 Use this attribute on the H8/300H and H8S to indicate that the specified
6511 variable should be placed into the tiny data section.
6512 The compiler generates more efficient code for loads and stores
6513 on data in the tiny data section. Note the tiny data area is limited to
6514 slightly under 32KB of data.
6518 @node IA-64 Variable Attributes
6519 @subsection IA-64 Variable Attributes
6521 The IA-64 back end supports the following variable attribute:
6524 @item model (@var{model-name})
6525 @cindex @code{model} variable attribute, IA-64
6527 On IA-64, use this attribute to set the addressability of an object.
6528 At present, the only supported identifier for @var{model-name} is
6529 @code{small}, indicating addressability via ``small'' (22-bit)
6530 addresses (so that their addresses can be loaded with the @code{addl}
6531 instruction). Caveat: such addressing is by definition not position
6532 independent and hence this attribute must not be used for objects
6533 defined by shared libraries.
6537 @node M32R/D Variable Attributes
6538 @subsection M32R/D Variable Attributes
6540 One attribute is currently defined for the M32R/D@.
6543 @item model (@var{model-name})
6544 @cindex @code{model-name} variable attribute, M32R/D
6545 @cindex variable addressability on the M32R/D
6546 Use this attribute on the M32R/D to set the addressability of an object.
6547 The identifier @var{model-name} is one of @code{small}, @code{medium},
6548 or @code{large}, representing each of the code models.
6550 Small model objects live in the lower 16MB of memory (so that their
6551 addresses can be loaded with the @code{ld24} instruction).
6553 Medium and large model objects may live anywhere in the 32-bit address space
6554 (the compiler generates @code{seth/add3} instructions to load their
6558 @node MeP Variable Attributes
6559 @subsection MeP Variable Attributes
6561 The MeP target has a number of addressing modes and busses. The
6562 @code{near} space spans the standard memory space's first 16 megabytes
6563 (24 bits). The @code{far} space spans the entire 32-bit memory space.
6564 The @code{based} space is a 128-byte region in the memory space that
6565 is addressed relative to the @code{$tp} register. The @code{tiny}
6566 space is a 65536-byte region relative to the @code{$gp} register. In
6567 addition to these memory regions, the MeP target has a separate 16-bit
6568 control bus which is specified with @code{cb} attributes.
6573 @cindex @code{based} variable attribute, MeP
6574 Any variable with the @code{based} attribute is assigned to the
6575 @code{.based} section, and is accessed with relative to the
6576 @code{$tp} register.
6579 @cindex @code{tiny} variable attribute, MeP
6580 Likewise, the @code{tiny} attribute assigned variables to the
6581 @code{.tiny} section, relative to the @code{$gp} register.
6584 @cindex @code{near} variable attribute, MeP
6585 Variables with the @code{near} attribute are assumed to have addresses
6586 that fit in a 24-bit addressing mode. This is the default for large
6587 variables (@code{-mtiny=4} is the default) but this attribute can
6588 override @code{-mtiny=} for small variables, or override @code{-ml}.
6591 @cindex @code{far} variable attribute, MeP
6592 Variables with the @code{far} attribute are addressed using a full
6593 32-bit address. Since this covers the entire memory space, this
6594 allows modules to make no assumptions about where variables might be
6598 @cindex @code{io} variable attribute, MeP
6599 @itemx io (@var{addr})
6600 Variables with the @code{io} attribute are used to address
6601 memory-mapped peripherals. If an address is specified, the variable
6602 is assigned that address, else it is not assigned an address (it is
6603 assumed some other module assigns an address). Example:
6606 int timer_count __attribute__((io(0x123)));
6610 @itemx cb (@var{addr})
6611 @cindex @code{cb} variable attribute, MeP
6612 Variables with the @code{cb} attribute are used to access the control
6613 bus, using special instructions. @code{addr} indicates the control bus
6617 int cpu_clock __attribute__((cb(0x123)));
6622 @node Microsoft Windows Variable Attributes
6623 @subsection Microsoft Windows Variable Attributes
6625 You can use these attributes on Microsoft Windows targets.
6626 @ref{x86 Variable Attributes} for additional Windows compatibility
6627 attributes available on all x86 targets.
6632 @cindex @code{dllimport} variable attribute
6633 @cindex @code{dllexport} variable attribute
6634 The @code{dllimport} and @code{dllexport} attributes are described in
6635 @ref{Microsoft Windows Function Attributes}.
6638 @cindex @code{selectany} variable attribute
6639 The @code{selectany} attribute causes an initialized global variable to
6640 have link-once semantics. When multiple definitions of the variable are
6641 encountered by the linker, the first is selected and the remainder are
6642 discarded. Following usage by the Microsoft compiler, the linker is told
6643 @emph{not} to warn about size or content differences of the multiple
6646 Although the primary usage of this attribute is for POD types, the
6647 attribute can also be applied to global C++ objects that are initialized
6648 by a constructor. In this case, the static initialization and destruction
6649 code for the object is emitted in each translation defining the object,
6650 but the calls to the constructor and destructor are protected by a
6651 link-once guard variable.
6653 The @code{selectany} attribute is only available on Microsoft Windows
6654 targets. You can use @code{__declspec (selectany)} as a synonym for
6655 @code{__attribute__ ((selectany))} for compatibility with other
6659 @cindex @code{shared} variable attribute
6660 On Microsoft Windows, in addition to putting variable definitions in a named
6661 section, the section can also be shared among all running copies of an
6662 executable or DLL@. For example, this small program defines shared data
6663 by putting it in a named section @code{shared} and marking the section
6667 int foo __attribute__((section ("shared"), shared)) = 0;
6672 /* @r{Read and write foo. All running
6673 copies see the same value.} */
6679 You may only use the @code{shared} attribute along with @code{section}
6680 attribute with a fully-initialized global definition because of the way
6681 linkers work. See @code{section} attribute for more information.
6683 The @code{shared} attribute is only available on Microsoft Windows@.
6687 @node MSP430 Variable Attributes
6688 @subsection MSP430 Variable Attributes
6692 @cindex @code{noinit} variable attribute, MSP430
6693 Any data with the @code{noinit} attribute will not be initialised by
6694 the C runtime startup code, or the program loader. Not initialising
6695 data in this way can reduce program startup times.
6698 @cindex @code{persistent} variable attribute, MSP430
6699 Any variable with the @code{persistent} attribute will not be
6700 initialised by the C runtime startup code. Instead its value will be
6701 set once, when the application is loaded, and then never initialised
6702 again, even if the processor is reset or the program restarts.
6703 Persistent data is intended to be placed into FLASH RAM, where its
6704 value will be retained across resets. The linker script being used to
6705 create the application should ensure that persistent data is correctly
6711 @cindex @code{lower} variable attribute, MSP430
6712 @cindex @code{upper} variable attribute, MSP430
6713 @cindex @code{either} variable attribute, MSP430
6714 These attributes are the same as the MSP430 function attributes of the
6715 same name (@pxref{MSP430 Function Attributes}).
6716 These attributes can be applied to both functions and variables.
6719 @node Nvidia PTX Variable Attributes
6720 @subsection Nvidia PTX Variable Attributes
6722 These variable attributes are supported by the Nvidia PTX back end:
6726 @cindex @code{shared} attribute, Nvidia PTX
6727 Use this attribute to place a variable in the @code{.shared} memory space.
6728 This memory space is private to each cooperative thread array; only threads
6729 within one thread block refer to the same instance of the variable.
6730 The runtime does not initialize variables in this memory space.
6733 @node PowerPC Variable Attributes
6734 @subsection PowerPC Variable Attributes
6736 Three attributes currently are defined for PowerPC configurations:
6737 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
6739 @cindex @code{ms_struct} variable attribute, PowerPC
6740 @cindex @code{gcc_struct} variable attribute, PowerPC
6741 For full documentation of the struct attributes please see the
6742 documentation in @ref{x86 Variable Attributes}.
6744 @cindex @code{altivec} variable attribute, PowerPC
6745 For documentation of @code{altivec} attribute please see the
6746 documentation in @ref{PowerPC Type Attributes}.
6748 @node RL78 Variable Attributes
6749 @subsection RL78 Variable Attributes
6751 @cindex @code{saddr} variable attribute, RL78
6752 The RL78 back end supports the @code{saddr} variable attribute. This
6753 specifies placement of the corresponding variable in the SADDR area,
6754 which can be accessed more efficiently than the default memory region.
6756 @node SPU Variable Attributes
6757 @subsection SPU Variable Attributes
6759 @cindex @code{spu_vector} variable attribute, SPU
6760 The SPU supports the @code{spu_vector} attribute for variables. For
6761 documentation of this attribute please see the documentation in
6762 @ref{SPU Type Attributes}.
6764 @node V850 Variable Attributes
6765 @subsection V850 Variable Attributes
6767 These variable attributes are supported by the V850 back end:
6772 @cindex @code{sda} variable attribute, V850
6773 Use this attribute to explicitly place a variable in the small data area,
6774 which can hold up to 64 kilobytes.
6777 @cindex @code{tda} variable attribute, V850
6778 Use this attribute to explicitly place a variable in the tiny data area,
6779 which can hold up to 256 bytes in total.
6782 @cindex @code{zda} variable attribute, V850
6783 Use this attribute to explicitly place a variable in the first 32 kilobytes
6787 @node x86 Variable Attributes
6788 @subsection x86 Variable Attributes
6790 Two attributes are currently defined for x86 configurations:
6791 @code{ms_struct} and @code{gcc_struct}.
6796 @cindex @code{ms_struct} variable attribute, x86
6797 @cindex @code{gcc_struct} variable attribute, x86
6799 If @code{packed} is used on a structure, or if bit-fields are used,
6800 it may be that the Microsoft ABI lays out the structure differently
6801 than the way GCC normally does. Particularly when moving packed
6802 data between functions compiled with GCC and the native Microsoft compiler
6803 (either via function call or as data in a file), it may be necessary to access
6806 The @code{ms_struct} and @code{gcc_struct} attributes correspond
6807 to the @option{-mms-bitfields} and @option{-mno-ms-bitfields}
6808 command-line options, respectively;
6809 see @ref{x86 Options}, for details of how structure layout is affected.
6810 @xref{x86 Type Attributes}, for information about the corresponding
6811 attributes on types.
6815 @node Xstormy16 Variable Attributes
6816 @subsection Xstormy16 Variable Attributes
6818 One attribute is currently defined for xstormy16 configurations:
6823 @cindex @code{below100} variable attribute, Xstormy16
6825 If a variable has the @code{below100} attribute (@code{BELOW100} is
6826 allowed also), GCC places the variable in the first 0x100 bytes of
6827 memory and use special opcodes to access it. Such variables are
6828 placed in either the @code{.bss_below100} section or the
6829 @code{.data_below100} section.
6833 @node Type Attributes
6834 @section Specifying Attributes of Types
6835 @cindex attribute of types
6836 @cindex type attributes
6838 The keyword @code{__attribute__} allows you to specify special
6839 attributes of types. Some type attributes apply only to @code{struct}
6840 and @code{union} types, while others can apply to any type defined
6841 via a @code{typedef} declaration. Other attributes are defined for
6842 functions (@pxref{Function Attributes}), labels (@pxref{Label
6843 Attributes}), enumerators (@pxref{Enumerator Attributes}),
6844 statements (@pxref{Statement Attributes}), and for
6845 variables (@pxref{Variable Attributes}).
6847 The @code{__attribute__} keyword is followed by an attribute specification
6848 inside double parentheses.
6850 You may specify type attributes in an enum, struct or union type
6851 declaration or definition by placing them immediately after the
6852 @code{struct}, @code{union} or @code{enum} keyword. A less preferred
6853 syntax is to place them just past the closing curly brace of the
6856 You can also include type attributes in a @code{typedef} declaration.
6857 @xref{Attribute Syntax}, for details of the exact syntax for using
6861 * Common Type Attributes::
6862 * ARC Type Attributes::
6863 * ARM Type Attributes::
6864 * MeP Type Attributes::
6865 * PowerPC Type Attributes::
6866 * SPU Type Attributes::
6867 * x86 Type Attributes::
6870 @node Common Type Attributes
6871 @subsection Common Type Attributes
6873 The following type attributes are supported on most targets.
6876 @cindex @code{aligned} type attribute
6877 @item aligned (@var{alignment})
6878 This attribute specifies a minimum alignment (in bytes) for variables
6879 of the specified type. For example, the declarations:
6882 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
6883 typedef int more_aligned_int __attribute__ ((aligned (8)));
6887 force the compiler to ensure (as far as it can) that each variable whose
6888 type is @code{struct S} or @code{more_aligned_int} is allocated and
6889 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
6890 variables of type @code{struct S} aligned to 8-byte boundaries allows
6891 the compiler to use the @code{ldd} and @code{std} (doubleword load and
6892 store) instructions when copying one variable of type @code{struct S} to
6893 another, thus improving run-time efficiency.
6895 Note that the alignment of any given @code{struct} or @code{union} type
6896 is required by the ISO C standard to be at least a perfect multiple of
6897 the lowest common multiple of the alignments of all of the members of
6898 the @code{struct} or @code{union} in question. This means that you @emph{can}
6899 effectively adjust the alignment of a @code{struct} or @code{union}
6900 type by attaching an @code{aligned} attribute to any one of the members
6901 of such a type, but the notation illustrated in the example above is a
6902 more obvious, intuitive, and readable way to request the compiler to
6903 adjust the alignment of an entire @code{struct} or @code{union} type.
6905 As in the preceding example, you can explicitly specify the alignment
6906 (in bytes) that you wish the compiler to use for a given @code{struct}
6907 or @code{union} type. Alternatively, you can leave out the alignment factor
6908 and just ask the compiler to align a type to the maximum
6909 useful alignment for the target machine you are compiling for. For
6910 example, you could write:
6913 struct S @{ short f[3]; @} __attribute__ ((aligned));
6916 Whenever you leave out the alignment factor in an @code{aligned}
6917 attribute specification, the compiler automatically sets the alignment
6918 for the type to the largest alignment that is ever used for any data
6919 type on the target machine you are compiling for. Doing this can often
6920 make copy operations more efficient, because the compiler can use
6921 whatever instructions copy the biggest chunks of memory when performing
6922 copies to or from the variables that have types that you have aligned
6925 In the example above, if the size of each @code{short} is 2 bytes, then
6926 the size of the entire @code{struct S} type is 6 bytes. The smallest
6927 power of two that is greater than or equal to that is 8, so the
6928 compiler sets the alignment for the entire @code{struct S} type to 8
6931 Note that although you can ask the compiler to select a time-efficient
6932 alignment for a given type and then declare only individual stand-alone
6933 objects of that type, the compiler's ability to select a time-efficient
6934 alignment is primarily useful only when you plan to create arrays of
6935 variables having the relevant (efficiently aligned) type. If you
6936 declare or use arrays of variables of an efficiently-aligned type, then
6937 it is likely that your program also does pointer arithmetic (or
6938 subscripting, which amounts to the same thing) on pointers to the
6939 relevant type, and the code that the compiler generates for these
6940 pointer arithmetic operations is often more efficient for
6941 efficiently-aligned types than for other types.
6943 Note that the effectiveness of @code{aligned} attributes may be limited
6944 by inherent limitations in your linker. On many systems, the linker is
6945 only able to arrange for variables to be aligned up to a certain maximum
6946 alignment. (For some linkers, the maximum supported alignment may
6947 be very very small.) If your linker is only able to align variables
6948 up to a maximum of 8-byte alignment, then specifying @code{aligned(16)}
6949 in an @code{__attribute__} still only provides you with 8-byte
6950 alignment. See your linker documentation for further information.
6952 The @code{aligned} attribute can only increase alignment. Alignment
6953 can be decreased by specifying the @code{packed} attribute. See below.
6955 @cindex @code{warn_if_not_aligned} type attribute
6956 @item warn_if_not_aligned (@var{alignment})
6957 This attribute specifies a threshold for the structure field, measured
6958 in bytes. If the structure field is aligned below the threshold, a
6959 warning will be issued. For example, the declaration:
6962 typedef unsigned long long __u64
6963 __attribute__((aligned(4),warn_if_not_aligned(8)));
6974 causes the compiler to issue an warning on @code{struct foo}, like
6975 @samp{warning: alignment 4 of 'struct foo' is less than 8}.
6976 It is used to define @code{struct foo} in such a way that
6977 @code{struct foo} has the same layout and the structure field @code{x}
6978 has the same alignment when @code{__u64} is aligned at either 4 or
6979 8 bytes. Align @code{struct foo} to 8 bytes:
6987 @} __attribute__((aligned(8)));
6991 silences the warning. The compiler also issues a warning, like
6992 @samp{warning: 'x' offset 12 in 'struct foo' isn't aligned to 8},
6993 when the structure field has the misaligned offset:
7002 @} __attribute__((aligned(8)));
7005 This warning can be disabled by @option{-Wno-if-not-aligned}.
7007 @item bnd_variable_size
7008 @cindex @code{bnd_variable_size} type attribute
7009 @cindex Pointer Bounds Checker attributes
7010 When applied to a structure field, this attribute tells Pointer
7011 Bounds Checker that the size of this field should not be computed
7012 using static type information. It may be used to mark variably-sized
7013 static array fields placed at the end of a structure.
7021 S *p = (S *)malloc (sizeof(S) + 100);
7022 p->data[10] = 0; //Bounds violation
7026 By using an attribute for the field we may avoid unwanted bound
7033 char data[1] __attribute__((bnd_variable_size));
7035 S *p = (S *)malloc (sizeof(S) + 100);
7036 p->data[10] = 0; //OK
7040 @itemx deprecated (@var{msg})
7041 @cindex @code{deprecated} type attribute
7042 The @code{deprecated} attribute results in a warning if the type
7043 is used anywhere in the source file. This is useful when identifying
7044 types that are expected to be removed in a future version of a program.
7045 If possible, the warning also includes the location of the declaration
7046 of the deprecated type, to enable users to easily find further
7047 information about why the type is deprecated, or what they should do
7048 instead. Note that the warnings only occur for uses and then only
7049 if the type is being applied to an identifier that itself is not being
7050 declared as deprecated.
7053 typedef int T1 __attribute__ ((deprecated));
7057 typedef T1 T3 __attribute__ ((deprecated));
7058 T3 z __attribute__ ((deprecated));
7062 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
7063 warning is issued for line 4 because T2 is not explicitly
7064 deprecated. Line 5 has no warning because T3 is explicitly
7065 deprecated. Similarly for line 6. The optional @var{msg}
7066 argument, which must be a string, is printed in the warning if
7069 The @code{deprecated} attribute can also be used for functions and
7070 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
7072 @item designated_init
7073 @cindex @code{designated_init} type attribute
7074 This attribute may only be applied to structure types. It indicates
7075 that any initialization of an object of this type must use designated
7076 initializers rather than positional initializers. The intent of this
7077 attribute is to allow the programmer to indicate that a structure's
7078 layout may change, and that therefore relying on positional
7079 initialization will result in future breakage.
7081 GCC emits warnings based on this attribute by default; use
7082 @option{-Wno-designated-init} to suppress them.
7085 @cindex @code{may_alias} type attribute
7086 Accesses through pointers to types with this attribute are not subject
7087 to type-based alias analysis, but are instead assumed to be able to alias
7088 any other type of objects.
7089 In the context of section 6.5 paragraph 7 of the C99 standard,
7090 an lvalue expression
7091 dereferencing such a pointer is treated like having a character type.
7092 See @option{-fstrict-aliasing} for more information on aliasing issues.
7093 This extension exists to support some vector APIs, in which pointers to
7094 one vector type are permitted to alias pointers to a different vector type.
7096 Note that an object of a type with this attribute does not have any
7102 typedef short __attribute__((__may_alias__)) short_a;
7108 short_a *b = (short_a *) &a;
7112 if (a == 0x12345678)
7120 If you replaced @code{short_a} with @code{short} in the variable
7121 declaration, the above program would abort when compiled with
7122 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
7126 @cindex @code{packed} type attribute
7127 This attribute, attached to @code{struct} or @code{union} type
7128 definition, specifies that each member (other than zero-width bit-fields)
7129 of the structure or union is placed to minimize the memory required. When
7130 attached to an @code{enum} definition, it indicates that the smallest
7131 integral type should be used.
7133 @opindex fshort-enums
7134 Specifying the @code{packed} attribute for @code{struct} and @code{union}
7135 types is equivalent to specifying the @code{packed} attribute on each
7136 of the structure or union members. Specifying the @option{-fshort-enums}
7137 flag on the command line is equivalent to specifying the @code{packed}
7138 attribute on all @code{enum} definitions.
7140 In the following example @code{struct my_packed_struct}'s members are
7141 packed closely together, but the internal layout of its @code{s} member
7142 is not packed---to do that, @code{struct my_unpacked_struct} needs to
7146 struct my_unpacked_struct
7152 struct __attribute__ ((__packed__)) my_packed_struct
7156 struct my_unpacked_struct s;
7160 You may only specify the @code{packed} attribute attribute on the definition
7161 of an @code{enum}, @code{struct} or @code{union}, not on a @code{typedef}
7162 that does not also define the enumerated type, structure or union.
7164 @item scalar_storage_order ("@var{endianness}")
7165 @cindex @code{scalar_storage_order} type attribute
7166 When attached to a @code{union} or a @code{struct}, this attribute sets
7167 the storage order, aka endianness, of the scalar fields of the type, as
7168 well as the array fields whose component is scalar. The supported
7169 endiannesses are @code{big-endian} and @code{little-endian}. The attribute
7170 has no effects on fields which are themselves a @code{union}, a @code{struct}
7171 or an array whose component is a @code{union} or a @code{struct}, and it is
7172 possible for these fields to have a different scalar storage order than the
7175 This attribute is supported only for targets that use a uniform default
7176 scalar storage order (fortunately, most of them), i.e. targets that store
7177 the scalars either all in big-endian or all in little-endian.
7179 Additional restrictions are enforced for types with the reverse scalar
7180 storage order with regard to the scalar storage order of the target:
7183 @item Taking the address of a scalar field of a @code{union} or a
7184 @code{struct} with reverse scalar storage order is not permitted and yields
7186 @item Taking the address of an array field, whose component is scalar, of
7187 a @code{union} or a @code{struct} with reverse scalar storage order is
7188 permitted but yields a warning, unless @option{-Wno-scalar-storage-order}
7190 @item Taking the address of a @code{union} or a @code{struct} with reverse
7191 scalar storage order is permitted.
7194 These restrictions exist because the storage order attribute is lost when
7195 the address of a scalar or the address of an array with scalar component is
7196 taken, so storing indirectly through this address generally does not work.
7197 The second case is nevertheless allowed to be able to perform a block copy
7198 from or to the array.
7200 Moreover, the use of type punning or aliasing to toggle the storage order
7201 is not supported; that is to say, a given scalar object cannot be accessed
7202 through distinct types that assign a different storage order to it.
7204 @item transparent_union
7205 @cindex @code{transparent_union} type attribute
7207 This attribute, attached to a @code{union} type definition, indicates
7208 that any function parameter having that union type causes calls to that
7209 function to be treated in a special way.
7211 First, the argument corresponding to a transparent union type can be of
7212 any type in the union; no cast is required. Also, if the union contains
7213 a pointer type, the corresponding argument can be a null pointer
7214 constant or a void pointer expression; and if the union contains a void
7215 pointer type, the corresponding argument can be any pointer expression.
7216 If the union member type is a pointer, qualifiers like @code{const} on
7217 the referenced type must be respected, just as with normal pointer
7220 Second, the argument is passed to the function using the calling
7221 conventions of the first member of the transparent union, not the calling
7222 conventions of the union itself. All members of the union must have the
7223 same machine representation; this is necessary for this argument passing
7226 Transparent unions are designed for library functions that have multiple
7227 interfaces for compatibility reasons. For example, suppose the
7228 @code{wait} function must accept either a value of type @code{int *} to
7229 comply with POSIX, or a value of type @code{union wait *} to comply with
7230 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
7231 @code{wait} would accept both kinds of arguments, but it would also
7232 accept any other pointer type and this would make argument type checking
7233 less useful. Instead, @code{<sys/wait.h>} might define the interface
7237 typedef union __attribute__ ((__transparent_union__))
7241 @} wait_status_ptr_t;
7243 pid_t wait (wait_status_ptr_t);
7247 This interface allows either @code{int *} or @code{union wait *}
7248 arguments to be passed, using the @code{int *} calling convention.
7249 The program can call @code{wait} with arguments of either type:
7252 int w1 () @{ int w; return wait (&w); @}
7253 int w2 () @{ union wait w; return wait (&w); @}
7257 With this interface, @code{wait}'s implementation might look like this:
7260 pid_t wait (wait_status_ptr_t p)
7262 return waitpid (-1, p.__ip, 0);
7267 @cindex @code{unused} type attribute
7268 When attached to a type (including a @code{union} or a @code{struct}),
7269 this attribute means that variables of that type are meant to appear
7270 possibly unused. GCC does not produce a warning for any variables of
7271 that type, even if the variable appears to do nothing. This is often
7272 the case with lock or thread classes, which are usually defined and then
7273 not referenced, but contain constructors and destructors that have
7274 nontrivial bookkeeping functions.
7277 @cindex @code{visibility} type attribute
7278 In C++, attribute visibility (@pxref{Function Attributes}) can also be
7279 applied to class, struct, union and enum types. Unlike other type
7280 attributes, the attribute must appear between the initial keyword and
7281 the name of the type; it cannot appear after the body of the type.
7283 Note that the type visibility is applied to vague linkage entities
7284 associated with the class (vtable, typeinfo node, etc.). In
7285 particular, if a class is thrown as an exception in one shared object
7286 and caught in another, the class must have default visibility.
7287 Otherwise the two shared objects are unable to use the same
7288 typeinfo node and exception handling will break.
7292 To specify multiple attributes, separate them by commas within the
7293 double parentheses: for example, @samp{__attribute__ ((aligned (16),
7296 @node ARC Type Attributes
7297 @subsection ARC Type Attributes
7299 @cindex @code{uncached} type attribute, ARC
7300 Declaring objects with @code{uncached} allows you to exclude
7301 data-cache participation in load and store operations on those objects
7302 without involving the additional semantic implications of
7303 @code{volatile}. The @code{.di} instruction suffix is used for all
7304 loads and stores of data declared @code{uncached}.
7306 @node ARM Type Attributes
7307 @subsection ARM Type Attributes
7309 @cindex @code{notshared} type attribute, ARM
7310 On those ARM targets that support @code{dllimport} (such as Symbian
7311 OS), you can use the @code{notshared} attribute to indicate that the
7312 virtual table and other similar data for a class should not be
7313 exported from a DLL@. For example:
7316 class __declspec(notshared) C @{
7318 __declspec(dllimport) C();
7322 __declspec(dllexport)
7327 In this code, @code{C::C} is exported from the current DLL, but the
7328 virtual table for @code{C} is not exported. (You can use
7329 @code{__attribute__} instead of @code{__declspec} if you prefer, but
7330 most Symbian OS code uses @code{__declspec}.)
7332 @node MeP Type Attributes
7333 @subsection MeP Type Attributes
7335 @cindex @code{based} type attribute, MeP
7336 @cindex @code{tiny} type attribute, MeP
7337 @cindex @code{near} type attribute, MeP
7338 @cindex @code{far} type attribute, MeP
7339 Many of the MeP variable attributes may be applied to types as well.
7340 Specifically, the @code{based}, @code{tiny}, @code{near}, and
7341 @code{far} attributes may be applied to either. The @code{io} and
7342 @code{cb} attributes may not be applied to types.
7344 @node PowerPC Type Attributes
7345 @subsection PowerPC Type Attributes
7347 Three attributes currently are defined for PowerPC configurations:
7348 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
7350 @cindex @code{ms_struct} type attribute, PowerPC
7351 @cindex @code{gcc_struct} type attribute, PowerPC
7352 For full documentation of the @code{ms_struct} and @code{gcc_struct}
7353 attributes please see the documentation in @ref{x86 Type Attributes}.
7355 @cindex @code{altivec} type attribute, PowerPC
7356 The @code{altivec} attribute allows one to declare AltiVec vector data
7357 types supported by the AltiVec Programming Interface Manual. The
7358 attribute requires an argument to specify one of three vector types:
7359 @code{vector__}, @code{pixel__} (always followed by unsigned short),
7360 and @code{bool__} (always followed by unsigned).
7363 __attribute__((altivec(vector__)))
7364 __attribute__((altivec(pixel__))) unsigned short
7365 __attribute__((altivec(bool__))) unsigned
7368 These attributes mainly are intended to support the @code{__vector},
7369 @code{__pixel}, and @code{__bool} AltiVec keywords.
7371 @node SPU Type Attributes
7372 @subsection SPU Type Attributes
7374 @cindex @code{spu_vector} type attribute, SPU
7375 The SPU supports the @code{spu_vector} attribute for types. This attribute
7376 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
7377 Language Extensions Specification. It is intended to support the
7378 @code{__vector} keyword.
7380 @node x86 Type Attributes
7381 @subsection x86 Type Attributes
7383 Two attributes are currently defined for x86 configurations:
7384 @code{ms_struct} and @code{gcc_struct}.
7390 @cindex @code{ms_struct} type attribute, x86
7391 @cindex @code{gcc_struct} type attribute, x86
7393 If @code{packed} is used on a structure, or if bit-fields are used
7394 it may be that the Microsoft ABI packs them differently
7395 than GCC normally packs them. Particularly when moving packed
7396 data between functions compiled with GCC and the native Microsoft compiler
7397 (either via function call or as data in a file), it may be necessary to access
7400 The @code{ms_struct} and @code{gcc_struct} attributes correspond
7401 to the @option{-mms-bitfields} and @option{-mno-ms-bitfields}
7402 command-line options, respectively;
7403 see @ref{x86 Options}, for details of how structure layout is affected.
7404 @xref{x86 Variable Attributes}, for information about the corresponding
7405 attributes on variables.
7409 @node Label Attributes
7410 @section Label Attributes
7411 @cindex Label Attributes
7413 GCC allows attributes to be set on C labels. @xref{Attribute Syntax}, for
7414 details of the exact syntax for using attributes. Other attributes are
7415 available for functions (@pxref{Function Attributes}), variables
7416 (@pxref{Variable Attributes}), enumerators (@pxref{Enumerator Attributes}),
7417 statements (@pxref{Statement Attributes}), and for types
7418 (@pxref{Type Attributes}).
7420 This example uses the @code{cold} label attribute to indicate the
7421 @code{ErrorHandling} branch is unlikely to be taken and that the
7422 @code{ErrorHandling} label is unused:
7426 asm goto ("some asm" : : : : NoError);
7428 /* This branch (the fall-through from the asm) is less commonly used */
7430 __attribute__((cold, unused)); /* Semi-colon is required here */
7435 printf("no error\n");
7441 @cindex @code{unused} label attribute
7442 This feature is intended for program-generated code that may contain
7443 unused labels, but which is compiled with @option{-Wall}. It is
7444 not normally appropriate to use in it human-written code, though it
7445 could be useful in cases where the code that jumps to the label is
7446 contained within an @code{#ifdef} conditional.
7449 @cindex @code{hot} label attribute
7450 The @code{hot} attribute on a label is used to inform the compiler that
7451 the path following the label is more likely than paths that are not so
7452 annotated. This attribute is used in cases where @code{__builtin_expect}
7453 cannot be used, for instance with computed goto or @code{asm goto}.
7456 @cindex @code{cold} label attribute
7457 The @code{cold} attribute on labels is used to inform the compiler that
7458 the path following the label is unlikely to be executed. This attribute
7459 is used in cases where @code{__builtin_expect} cannot be used, for instance
7460 with computed goto or @code{asm goto}.
7464 @node Enumerator Attributes
7465 @section Enumerator Attributes
7466 @cindex Enumerator Attributes
7468 GCC allows attributes to be set on enumerators. @xref{Attribute Syntax}, for
7469 details of the exact syntax for using attributes. Other attributes are
7470 available for functions (@pxref{Function Attributes}), variables
7471 (@pxref{Variable Attributes}), labels (@pxref{Label Attributes}), statements
7472 (@pxref{Statement Attributes}), and for types (@pxref{Type Attributes}).
7474 This example uses the @code{deprecated} enumerator attribute to indicate the
7475 @code{oldval} enumerator is deprecated:
7479 oldval __attribute__((deprecated)),
7492 @cindex @code{deprecated} enumerator attribute
7493 The @code{deprecated} attribute results in a warning if the enumerator
7494 is used anywhere in the source file. This is useful when identifying
7495 enumerators that are expected to be removed in a future version of a
7496 program. The warning also includes the location of the declaration
7497 of the deprecated enumerator, to enable users to easily find further
7498 information about why the enumerator is deprecated, or what they should
7499 do instead. Note that the warnings only occurs for uses.
7503 @node Statement Attributes
7504 @section Statement Attributes
7505 @cindex Statement Attributes
7507 GCC allows attributes to be set on null statements. @xref{Attribute Syntax},
7508 for details of the exact syntax for using attributes. Other attributes are
7509 available for functions (@pxref{Function Attributes}), variables
7510 (@pxref{Variable Attributes}), labels (@pxref{Label Attributes}), enumerators
7511 (@pxref{Enumerator Attributes}), and for types (@pxref{Type Attributes}).
7513 This example uses the @code{fallthrough} statement attribute to indicate that
7514 the @option{-Wimplicit-fallthrough} warning should not be emitted:
7521 __attribute__((fallthrough));
7529 @cindex @code{fallthrough} statement attribute
7530 The @code{fallthrough} attribute with a null statement serves as a
7531 fallthrough statement. It hints to the compiler that a statement
7532 that falls through to another case label, or user-defined label
7533 in a switch statement is intentional and thus the
7534 @option{-Wimplicit-fallthrough} warning must not trigger. The
7535 fallthrough attribute may appear at most once in each attribute
7536 list, and may not be mixed with other attributes. It can only
7537 be used in a switch statement (the compiler will issue an error
7538 otherwise), after a preceding statement and before a logically
7539 succeeding case label, or user-defined label.
7543 @node Attribute Syntax
7544 @section Attribute Syntax
7545 @cindex attribute syntax
7547 This section describes the syntax with which @code{__attribute__} may be
7548 used, and the constructs to which attribute specifiers bind, for the C
7549 language. Some details may vary for C++ and Objective-C@. Because of
7550 infelicities in the grammar for attributes, some forms described here
7551 may not be successfully parsed in all cases.
7553 There are some problems with the semantics of attributes in C++. For
7554 example, there are no manglings for attributes, although they may affect
7555 code generation, so problems may arise when attributed types are used in
7556 conjunction with templates or overloading. Similarly, @code{typeid}
7557 does not distinguish between types with different attributes. Support
7558 for attributes in C++ may be restricted in future to attributes on
7559 declarations only, but not on nested declarators.
7561 @xref{Function Attributes}, for details of the semantics of attributes
7562 applying to functions. @xref{Variable Attributes}, for details of the
7563 semantics of attributes applying to variables. @xref{Type Attributes},
7564 for details of the semantics of attributes applying to structure, union
7565 and enumerated types.
7566 @xref{Label Attributes}, for details of the semantics of attributes
7568 @xref{Enumerator Attributes}, for details of the semantics of attributes
7569 applying to enumerators.
7570 @xref{Statement Attributes}, for details of the semantics of attributes
7571 applying to statements.
7573 An @dfn{attribute specifier} is of the form
7574 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
7575 is a possibly empty comma-separated sequence of @dfn{attributes}, where
7576 each attribute is one of the following:
7580 Empty. Empty attributes are ignored.
7584 (which may be an identifier such as @code{unused}, or a reserved
7585 word such as @code{const}).
7588 An attribute name followed by a parenthesized list of
7589 parameters for the attribute.
7590 These parameters take one of the following forms:
7594 An identifier. For example, @code{mode} attributes use this form.
7597 An identifier followed by a comma and a non-empty comma-separated list
7598 of expressions. For example, @code{format} attributes use this form.
7601 A possibly empty comma-separated list of expressions. For example,
7602 @code{format_arg} attributes use this form with the list being a single
7603 integer constant expression, and @code{alias} attributes use this form
7604 with the list being a single string constant.
7608 An @dfn{attribute specifier list} is a sequence of one or more attribute
7609 specifiers, not separated by any other tokens.
7611 You may optionally specify attribute names with @samp{__}
7612 preceding and following the name.
7613 This allows you to use them in header files without
7614 being concerned about a possible macro of the same name. For example,
7615 you may use the attribute name @code{__noreturn__} instead of @code{noreturn}.
7618 @subsubheading Label Attributes
7620 In GNU C, an attribute specifier list may appear after the colon following a
7621 label, other than a @code{case} or @code{default} label. GNU C++ only permits
7622 attributes on labels if the attribute specifier is immediately
7623 followed by a semicolon (i.e., the label applies to an empty
7624 statement). If the semicolon is missing, C++ label attributes are
7625 ambiguous, as it is permissible for a declaration, which could begin
7626 with an attribute list, to be labelled in C++. Declarations cannot be
7627 labelled in C90 or C99, so the ambiguity does not arise there.
7629 @subsubheading Enumerator Attributes
7631 In GNU C, an attribute specifier list may appear as part of an enumerator.
7632 The attribute goes after the enumeration constant, before @code{=}, if
7633 present. The optional attribute in the enumerator appertains to the
7634 enumeration constant. It is not possible to place the attribute after
7635 the constant expression, if present.
7637 @subsubheading Statement Attributes
7638 In GNU C, an attribute specifier list may appear as part of a null
7639 statement. The attribute goes before the semicolon.
7641 @subsubheading Type Attributes
7643 An attribute specifier list may appear as part of a @code{struct},
7644 @code{union} or @code{enum} specifier. It may go either immediately
7645 after the @code{struct}, @code{union} or @code{enum} keyword, or after
7646 the closing brace. The former syntax is preferred.
7647 Where attribute specifiers follow the closing brace, they are considered
7648 to relate to the structure, union or enumerated type defined, not to any
7649 enclosing declaration the type specifier appears in, and the type
7650 defined is not complete until after the attribute specifiers.
7651 @c Otherwise, there would be the following problems: a shift/reduce
7652 @c conflict between attributes binding the struct/union/enum and
7653 @c binding to the list of specifiers/qualifiers; and "aligned"
7654 @c attributes could use sizeof for the structure, but the size could be
7655 @c changed later by "packed" attributes.
7658 @subsubheading All other attributes
7660 Otherwise, an attribute specifier appears as part of a declaration,
7661 counting declarations of unnamed parameters and type names, and relates
7662 to that declaration (which may be nested in another declaration, for
7663 example in the case of a parameter declaration), or to a particular declarator
7664 within a declaration. Where an
7665 attribute specifier is applied to a parameter declared as a function or
7666 an array, it should apply to the function or array rather than the
7667 pointer to which the parameter is implicitly converted, but this is not
7668 yet correctly implemented.
7670 Any list of specifiers and qualifiers at the start of a declaration may
7671 contain attribute specifiers, whether or not such a list may in that
7672 context contain storage class specifiers. (Some attributes, however,
7673 are essentially in the nature of storage class specifiers, and only make
7674 sense where storage class specifiers may be used; for example,
7675 @code{section}.) There is one necessary limitation to this syntax: the
7676 first old-style parameter declaration in a function definition cannot
7677 begin with an attribute specifier, because such an attribute applies to
7678 the function instead by syntax described below (which, however, is not
7679 yet implemented in this case). In some other cases, attribute
7680 specifiers are permitted by this grammar but not yet supported by the
7681 compiler. All attribute specifiers in this place relate to the
7682 declaration as a whole. In the obsolescent usage where a type of
7683 @code{int} is implied by the absence of type specifiers, such a list of
7684 specifiers and qualifiers may be an attribute specifier list with no
7685 other specifiers or qualifiers.
7687 At present, the first parameter in a function prototype must have some
7688 type specifier that is not an attribute specifier; this resolves an
7689 ambiguity in the interpretation of @code{void f(int
7690 (__attribute__((foo)) x))}, but is subject to change. At present, if
7691 the parentheses of a function declarator contain only attributes then
7692 those attributes are ignored, rather than yielding an error or warning
7693 or implying a single parameter of type int, but this is subject to
7696 An attribute specifier list may appear immediately before a declarator
7697 (other than the first) in a comma-separated list of declarators in a
7698 declaration of more than one identifier using a single list of
7699 specifiers and qualifiers. Such attribute specifiers apply
7700 only to the identifier before whose declarator they appear. For
7704 __attribute__((noreturn)) void d0 (void),
7705 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
7710 the @code{noreturn} attribute applies to all the functions
7711 declared; the @code{format} attribute only applies to @code{d1}.
7713 An attribute specifier list may appear immediately before the comma,
7714 @code{=} or semicolon terminating the declaration of an identifier other
7715 than a function definition. Such attribute specifiers apply
7716 to the declared object or function. Where an
7717 assembler name for an object or function is specified (@pxref{Asm
7718 Labels}), the attribute must follow the @code{asm}
7721 An attribute specifier list may, in future, be permitted to appear after
7722 the declarator in a function definition (before any old-style parameter
7723 declarations or the function body).
7725 Attribute specifiers may be mixed with type qualifiers appearing inside
7726 the @code{[]} of a parameter array declarator, in the C99 construct by
7727 which such qualifiers are applied to the pointer to which the array is
7728 implicitly converted. Such attribute specifiers apply to the pointer,
7729 not to the array, but at present this is not implemented and they are
7732 An attribute specifier list may appear at the start of a nested
7733 declarator. At present, there are some limitations in this usage: the
7734 attributes correctly apply to the declarator, but for most individual
7735 attributes the semantics this implies are not implemented.
7736 When attribute specifiers follow the @code{*} of a pointer
7737 declarator, they may be mixed with any type qualifiers present.
7738 The following describes the formal semantics of this syntax. It makes the
7739 most sense if you are familiar with the formal specification of
7740 declarators in the ISO C standard.
7742 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
7743 D1}, where @code{T} contains declaration specifiers that specify a type
7744 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
7745 contains an identifier @var{ident}. The type specified for @var{ident}
7746 for derived declarators whose type does not include an attribute
7747 specifier is as in the ISO C standard.
7749 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
7750 and the declaration @code{T D} specifies the type
7751 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
7752 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
7753 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
7755 If @code{D1} has the form @code{*
7756 @var{type-qualifier-and-attribute-specifier-list} D}, and the
7757 declaration @code{T D} specifies the type
7758 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
7759 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
7760 @var{type-qualifier-and-attribute-specifier-list} pointer to @var{Type}'' for
7766 void (__attribute__((noreturn)) ****f) (void);
7770 specifies the type ``pointer to pointer to pointer to pointer to
7771 non-returning function returning @code{void}''. As another example,
7774 char *__attribute__((aligned(8))) *f;
7778 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
7779 Note again that this does not work with most attributes; for example,
7780 the usage of @samp{aligned} and @samp{noreturn} attributes given above
7781 is not yet supported.
7783 For compatibility with existing code written for compiler versions that
7784 did not implement attributes on nested declarators, some laxity is
7785 allowed in the placing of attributes. If an attribute that only applies
7786 to types is applied to a declaration, it is treated as applying to
7787 the type of that declaration. If an attribute that only applies to
7788 declarations is applied to the type of a declaration, it is treated
7789 as applying to that declaration; and, for compatibility with code
7790 placing the attributes immediately before the identifier declared, such
7791 an attribute applied to a function return type is treated as
7792 applying to the function type, and such an attribute applied to an array
7793 element type is treated as applying to the array type. If an
7794 attribute that only applies to function types is applied to a
7795 pointer-to-function type, it is treated as applying to the pointer
7796 target type; if such an attribute is applied to a function return type
7797 that is not a pointer-to-function type, it is treated as applying
7798 to the function type.
7800 @node Function Prototypes
7801 @section Prototypes and Old-Style Function Definitions
7802 @cindex function prototype declarations
7803 @cindex old-style function definitions
7804 @cindex promotion of formal parameters
7806 GNU C extends ISO C to allow a function prototype to override a later
7807 old-style non-prototype definition. Consider the following example:
7810 /* @r{Use prototypes unless the compiler is old-fashioned.} */
7817 /* @r{Prototype function declaration.} */
7818 int isroot P((uid_t));
7820 /* @r{Old-style function definition.} */
7822 isroot (x) /* @r{??? lossage here ???} */
7829 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
7830 not allow this example, because subword arguments in old-style
7831 non-prototype definitions are promoted. Therefore in this example the
7832 function definition's argument is really an @code{int}, which does not
7833 match the prototype argument type of @code{short}.
7835 This restriction of ISO C makes it hard to write code that is portable
7836 to traditional C compilers, because the programmer does not know
7837 whether the @code{uid_t} type is @code{short}, @code{int}, or
7838 @code{long}. Therefore, in cases like these GNU C allows a prototype
7839 to override a later old-style definition. More precisely, in GNU C, a
7840 function prototype argument type overrides the argument type specified
7841 by a later old-style definition if the former type is the same as the
7842 latter type before promotion. Thus in GNU C the above example is
7843 equivalent to the following:
7856 GNU C++ does not support old-style function definitions, so this
7857 extension is irrelevant.
7860 @section C++ Style Comments
7862 @cindex C++ comments
7863 @cindex comments, C++ style
7865 In GNU C, you may use C++ style comments, which start with @samp{//} and
7866 continue until the end of the line. Many other C implementations allow
7867 such comments, and they are included in the 1999 C standard. However,
7868 C++ style comments are not recognized if you specify an @option{-std}
7869 option specifying a version of ISO C before C99, or @option{-ansi}
7870 (equivalent to @option{-std=c90}).
7873 @section Dollar Signs in Identifier Names
7875 @cindex dollar signs in identifier names
7876 @cindex identifier names, dollar signs in
7878 In GNU C, you may normally use dollar signs in identifier names.
7879 This is because many traditional C implementations allow such identifiers.
7880 However, dollar signs in identifiers are not supported on a few target
7881 machines, typically because the target assembler does not allow them.
7883 @node Character Escapes
7884 @section The Character @key{ESC} in Constants
7886 You can use the sequence @samp{\e} in a string or character constant to
7887 stand for the ASCII character @key{ESC}.
7890 @section Inquiring on Alignment of Types or Variables
7892 @cindex type alignment
7893 @cindex variable alignment
7895 The keyword @code{__alignof__} allows you to inquire about how an object
7896 is aligned, or the minimum alignment usually required by a type. Its
7897 syntax is just like @code{sizeof}.
7899 For example, if the target machine requires a @code{double} value to be
7900 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
7901 This is true on many RISC machines. On more traditional machine
7902 designs, @code{__alignof__ (double)} is 4 or even 2.
7904 Some machines never actually require alignment; they allow reference to any
7905 data type even at an odd address. For these machines, @code{__alignof__}
7906 reports the smallest alignment that GCC gives the data type, usually as
7907 mandated by the target ABI.
7909 If the operand of @code{__alignof__} is an lvalue rather than a type,
7910 its value is the required alignment for its type, taking into account
7911 any minimum alignment specified with GCC's @code{__attribute__}
7912 extension (@pxref{Variable Attributes}). For example, after this
7916 struct foo @{ int x; char y; @} foo1;
7920 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
7921 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
7923 It is an error to ask for the alignment of an incomplete type.
7927 @section An Inline Function is As Fast As a Macro
7928 @cindex inline functions
7929 @cindex integrating function code
7931 @cindex macros, inline alternative
7933 By declaring a function inline, you can direct GCC to make
7934 calls to that function faster. One way GCC can achieve this is to
7935 integrate that function's code into the code for its callers. This
7936 makes execution faster by eliminating the function-call overhead; in
7937 addition, if any of the actual argument values are constant, their
7938 known values may permit simplifications at compile time so that not
7939 all of the inline function's code needs to be included. The effect on
7940 code size is less predictable; object code may be larger or smaller
7941 with function inlining, depending on the particular case. You can
7942 also direct GCC to try to integrate all ``simple enough'' functions
7943 into their callers with the option @option{-finline-functions}.
7945 GCC implements three different semantics of declaring a function
7946 inline. One is available with @option{-std=gnu89} or
7947 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
7948 on all inline declarations, another when
7950 @option{-std=gnu99} or an option for a later C version is used
7951 (without @option{-fgnu89-inline}), and the third
7952 is used when compiling C++.
7954 To declare a function inline, use the @code{inline} keyword in its
7955 declaration, like this:
7965 If you are writing a header file to be included in ISO C90 programs, write
7966 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
7968 The three types of inlining behave similarly in two important cases:
7969 when the @code{inline} keyword is used on a @code{static} function,
7970 like the example above, and when a function is first declared without
7971 using the @code{inline} keyword and then is defined with
7972 @code{inline}, like this:
7975 extern int inc (int *a);
7983 In both of these common cases, the program behaves the same as if you
7984 had not used the @code{inline} keyword, except for its speed.
7986 @cindex inline functions, omission of
7987 @opindex fkeep-inline-functions
7988 When a function is both inline and @code{static}, if all calls to the
7989 function are integrated into the caller, and the function's address is
7990 never used, then the function's own assembler code is never referenced.
7991 In this case, GCC does not actually output assembler code for the
7992 function, unless you specify the option @option{-fkeep-inline-functions}.
7993 If there is a nonintegrated call, then the function is compiled to
7994 assembler code as usual. The function must also be compiled as usual if
7995 the program refers to its address, because that cannot be inlined.
7998 Note that certain usages in a function definition can make it unsuitable
7999 for inline substitution. Among these usages are: variadic functions,
8000 use of @code{alloca}, use of computed goto (@pxref{Labels as Values}),
8001 use of nonlocal goto, use of nested functions, use of @code{setjmp}, use
8002 of @code{__builtin_longjmp} and use of @code{__builtin_return} or
8003 @code{__builtin_apply_args}. Using @option{-Winline} warns when a
8004 function marked @code{inline} could not be substituted, and gives the
8005 reason for the failure.
8007 @cindex automatic @code{inline} for C++ member fns
8008 @cindex @code{inline} automatic for C++ member fns
8009 @cindex member fns, automatically @code{inline}
8010 @cindex C++ member fns, automatically @code{inline}
8011 @opindex fno-default-inline
8012 As required by ISO C++, GCC considers member functions defined within
8013 the body of a class to be marked inline even if they are
8014 not explicitly declared with the @code{inline} keyword. You can
8015 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
8016 Options,,Options Controlling C++ Dialect}.
8018 GCC does not inline any functions when not optimizing unless you specify
8019 the @samp{always_inline} attribute for the function, like this:
8022 /* @r{Prototype.} */
8023 inline void foo (const char) __attribute__((always_inline));
8026 The remainder of this section is specific to GNU C90 inlining.
8028 @cindex non-static inline function
8029 When an inline function is not @code{static}, then the compiler must assume
8030 that there may be calls from other source files; since a global symbol can
8031 be defined only once in any program, the function must not be defined in
8032 the other source files, so the calls therein cannot be integrated.
8033 Therefore, a non-@code{static} inline function is always compiled on its
8034 own in the usual fashion.
8036 If you specify both @code{inline} and @code{extern} in the function
8037 definition, then the definition is used only for inlining. In no case
8038 is the function compiled on its own, not even if you refer to its
8039 address explicitly. Such an address becomes an external reference, as
8040 if you had only declared the function, and had not defined it.
8042 This combination of @code{inline} and @code{extern} has almost the
8043 effect of a macro. The way to use it is to put a function definition in
8044 a header file with these keywords, and put another copy of the
8045 definition (lacking @code{inline} and @code{extern}) in a library file.
8046 The definition in the header file causes most calls to the function
8047 to be inlined. If any uses of the function remain, they refer to
8048 the single copy in the library.
8051 @section When is a Volatile Object Accessed?
8052 @cindex accessing volatiles
8053 @cindex volatile read
8054 @cindex volatile write
8055 @cindex volatile access
8057 C has the concept of volatile objects. These are normally accessed by
8058 pointers and used for accessing hardware or inter-thread
8059 communication. The standard encourages compilers to refrain from
8060 optimizations concerning accesses to volatile objects, but leaves it
8061 implementation defined as to what constitutes a volatile access. The
8062 minimum requirement is that at a sequence point all previous accesses
8063 to volatile objects have stabilized and no subsequent accesses have
8064 occurred. Thus an implementation is free to reorder and combine
8065 volatile accesses that occur between sequence points, but cannot do
8066 so for accesses across a sequence point. The use of volatile does
8067 not allow you to violate the restriction on updating objects multiple
8068 times between two sequence points.
8070 Accesses to non-volatile objects are not ordered with respect to
8071 volatile accesses. You cannot use a volatile object as a memory
8072 barrier to order a sequence of writes to non-volatile memory. For
8076 int *ptr = @var{something};
8078 *ptr = @var{something};
8083 Unless @var{*ptr} and @var{vobj} can be aliased, it is not guaranteed
8084 that the write to @var{*ptr} occurs by the time the update
8085 of @var{vobj} happens. If you need this guarantee, you must use
8086 a stronger memory barrier such as:
8089 int *ptr = @var{something};
8091 *ptr = @var{something};
8092 asm volatile ("" : : : "memory");
8096 A scalar volatile object is read when it is accessed in a void context:
8099 volatile int *src = @var{somevalue};
8103 Such expressions are rvalues, and GCC implements this as a
8104 read of the volatile object being pointed to.
8106 Assignments are also expressions and have an rvalue. However when
8107 assigning to a scalar volatile, the volatile object is not reread,
8108 regardless of whether the assignment expression's rvalue is used or
8109 not. If the assignment's rvalue is used, the value is that assigned
8110 to the volatile object. For instance, there is no read of @var{vobj}
8111 in all the following cases:
8116 vobj = @var{something};
8117 obj = vobj = @var{something};
8118 obj ? vobj = @var{onething} : vobj = @var{anotherthing};
8119 obj = (@var{something}, vobj = @var{anotherthing});
8122 If you need to read the volatile object after an assignment has
8123 occurred, you must use a separate expression with an intervening
8126 As bit-fields are not individually addressable, volatile bit-fields may
8127 be implicitly read when written to, or when adjacent bit-fields are
8128 accessed. Bit-field operations may be optimized such that adjacent
8129 bit-fields are only partially accessed, if they straddle a storage unit
8130 boundary. For these reasons it is unwise to use volatile bit-fields to
8133 @node Using Assembly Language with C
8134 @section How to Use Inline Assembly Language in C Code
8135 @cindex @code{asm} keyword
8136 @cindex assembly language in C
8137 @cindex inline assembly language
8138 @cindex mixing assembly language and C
8140 The @code{asm} keyword allows you to embed assembler instructions
8141 within C code. GCC provides two forms of inline @code{asm}
8142 statements. A @dfn{basic @code{asm}} statement is one with no
8143 operands (@pxref{Basic Asm}), while an @dfn{extended @code{asm}}
8144 statement (@pxref{Extended Asm}) includes one or more operands.
8145 The extended form is preferred for mixing C and assembly language
8146 within a function, but to include assembly language at
8147 top level you must use basic @code{asm}.
8149 You can also use the @code{asm} keyword to override the assembler name
8150 for a C symbol, or to place a C variable in a specific register.
8153 * Basic Asm:: Inline assembler without operands.
8154 * Extended Asm:: Inline assembler with operands.
8155 * Constraints:: Constraints for @code{asm} operands
8156 * Asm Labels:: Specifying the assembler name to use for a C symbol.
8157 * Explicit Register Variables:: Defining variables residing in specified
8159 * Size of an asm:: How GCC calculates the size of an @code{asm} block.
8163 @subsection Basic Asm --- Assembler Instructions Without Operands
8164 @cindex basic @code{asm}
8165 @cindex assembly language in C, basic
8167 A basic @code{asm} statement has the following syntax:
8170 asm @r{[} volatile @r{]} ( @var{AssemblerInstructions} )
8173 The @code{asm} keyword is a GNU extension.
8174 When writing code that can be compiled with @option{-ansi} and the
8175 various @option{-std} options, use @code{__asm__} instead of
8176 @code{asm} (@pxref{Alternate Keywords}).
8178 @subsubheading Qualifiers
8181 The optional @code{volatile} qualifier has no effect.
8182 All basic @code{asm} blocks are implicitly volatile.
8185 @subsubheading Parameters
8188 @item AssemblerInstructions
8189 This is a literal string that specifies the assembler code. The string can
8190 contain any instructions recognized by the assembler, including directives.
8191 GCC does not parse the assembler instructions themselves and
8192 does not know what they mean or even whether they are valid assembler input.
8194 You may place multiple assembler instructions together in a single @code{asm}
8195 string, separated by the characters normally used in assembly code for the
8196 system. A combination that works in most places is a newline to break the
8197 line, plus a tab character (written as @samp{\n\t}).
8198 Some assemblers allow semicolons as a line separator. However,
8199 note that some assembler dialects use semicolons to start a comment.
8202 @subsubheading Remarks
8203 Using extended @code{asm} (@pxref{Extended Asm}) typically produces
8204 smaller, safer, and more efficient code, and in most cases it is a
8205 better solution than basic @code{asm}. However, there are two
8206 situations where only basic @code{asm} can be used:
8210 Extended @code{asm} statements have to be inside a C
8211 function, so to write inline assembly language at file scope (``top-level''),
8212 outside of C functions, you must use basic @code{asm}.
8213 You can use this technique to emit assembler directives,
8214 define assembly language macros that can be invoked elsewhere in the file,
8215 or write entire functions in assembly language.
8219 with the @code{naked} attribute also require basic @code{asm}
8220 (@pxref{Function Attributes}).
8223 Safely accessing C data and calling functions from basic @code{asm} is more
8224 complex than it may appear. To access C data, it is better to use extended
8227 Do not expect a sequence of @code{asm} statements to remain perfectly
8228 consecutive after compilation. If certain instructions need to remain
8229 consecutive in the output, put them in a single multi-instruction @code{asm}
8230 statement. Note that GCC's optimizers can move @code{asm} statements
8231 relative to other code, including across jumps.
8233 @code{asm} statements may not perform jumps into other @code{asm} statements.
8234 GCC does not know about these jumps, and therefore cannot take
8235 account of them when deciding how to optimize. Jumps from @code{asm} to C
8236 labels are only supported in extended @code{asm}.
8238 Under certain circumstances, GCC may duplicate (or remove duplicates of) your
8239 assembly code when optimizing. This can lead to unexpected duplicate
8240 symbol errors during compilation if your assembly code defines symbols or
8243 @strong{Warning:} The C standards do not specify semantics for @code{asm},
8244 making it a potential source of incompatibilities between compilers. These
8245 incompatibilities may not produce compiler warnings/errors.
8247 GCC does not parse basic @code{asm}'s @var{AssemblerInstructions}, which
8248 means there is no way to communicate to the compiler what is happening
8249 inside them. GCC has no visibility of symbols in the @code{asm} and may
8250 discard them as unreferenced. It also does not know about side effects of
8251 the assembler code, such as modifications to memory or registers. Unlike
8252 some compilers, GCC assumes that no changes to general purpose registers
8253 occur. This assumption may change in a future release.
8255 To avoid complications from future changes to the semantics and the
8256 compatibility issues between compilers, consider replacing basic @code{asm}
8257 with extended @code{asm}. See
8258 @uref{https://gcc.gnu.org/wiki/ConvertBasicAsmToExtended, How to convert
8259 from basic asm to extended asm} for information about how to perform this
8262 The compiler copies the assembler instructions in a basic @code{asm}
8263 verbatim to the assembly language output file, without
8264 processing dialects or any of the @samp{%} operators that are available with
8265 extended @code{asm}. This results in minor differences between basic
8266 @code{asm} strings and extended @code{asm} templates. For example, to refer to
8267 registers you might use @samp{%eax} in basic @code{asm} and
8268 @samp{%%eax} in extended @code{asm}.
8270 On targets such as x86 that support multiple assembler dialects,
8271 all basic @code{asm} blocks use the assembler dialect specified by the
8272 @option{-masm} command-line option (@pxref{x86 Options}).
8273 Basic @code{asm} provides no
8274 mechanism to provide different assembler strings for different dialects.
8276 For basic @code{asm} with non-empty assembler string GCC assumes
8277 the assembler block does not change any general purpose registers,
8278 but it may read or write any globally accessible variable.
8280 Here is an example of basic @code{asm} for i386:
8283 /* Note that this code will not compile with -masm=intel */
8284 #define DebugBreak() asm("int $3")
8288 @subsection Extended Asm - Assembler Instructions with C Expression Operands
8289 @cindex extended @code{asm}
8290 @cindex assembly language in C, extended
8292 With extended @code{asm} you can read and write C variables from
8293 assembler and perform jumps from assembler code to C labels.
8294 Extended @code{asm} syntax uses colons (@samp{:}) to delimit
8295 the operand parameters after the assembler template:
8298 asm @r{[}volatile@r{]} ( @var{AssemblerTemplate}
8299 : @var{OutputOperands}
8300 @r{[} : @var{InputOperands}
8301 @r{[} : @var{Clobbers} @r{]} @r{]})
8303 asm @r{[}volatile@r{]} goto ( @var{AssemblerTemplate}
8305 : @var{InputOperands}
8310 The @code{asm} keyword is a GNU extension.
8311 When writing code that can be compiled with @option{-ansi} and the
8312 various @option{-std} options, use @code{__asm__} instead of
8313 @code{asm} (@pxref{Alternate Keywords}).
8315 @subsubheading Qualifiers
8319 The typical use of extended @code{asm} statements is to manipulate input
8320 values to produce output values. However, your @code{asm} statements may
8321 also produce side effects. If so, you may need to use the @code{volatile}
8322 qualifier to disable certain optimizations. @xref{Volatile}.
8325 This qualifier informs the compiler that the @code{asm} statement may
8326 perform a jump to one of the labels listed in the @var{GotoLabels}.
8330 @subsubheading Parameters
8332 @item AssemblerTemplate
8333 This is a literal string that is the template for the assembler code. It is a
8334 combination of fixed text and tokens that refer to the input, output,
8335 and goto parameters. @xref{AssemblerTemplate}.
8337 @item OutputOperands
8338 A comma-separated list of the C variables modified by the instructions in the
8339 @var{AssemblerTemplate}. An empty list is permitted. @xref{OutputOperands}.
8342 A comma-separated list of C expressions read by the instructions in the
8343 @var{AssemblerTemplate}. An empty list is permitted. @xref{InputOperands}.
8346 A comma-separated list of registers or other values changed by the
8347 @var{AssemblerTemplate}, beyond those listed as outputs.
8348 An empty list is permitted. @xref{Clobbers and Scratch Registers}.
8351 When you are using the @code{goto} form of @code{asm}, this section contains
8352 the list of all C labels to which the code in the
8353 @var{AssemblerTemplate} may jump.
8356 @code{asm} statements may not perform jumps into other @code{asm} statements,
8357 only to the listed @var{GotoLabels}.
8358 GCC's optimizers do not know about other jumps; therefore they cannot take
8359 account of them when deciding how to optimize.
8362 The total number of input + output + goto operands is limited to 30.
8364 @subsubheading Remarks
8365 The @code{asm} statement allows you to include assembly instructions directly
8366 within C code. This may help you to maximize performance in time-sensitive
8367 code or to access assembly instructions that are not readily available to C
8370 Note that extended @code{asm} statements must be inside a function. Only
8371 basic @code{asm} may be outside functions (@pxref{Basic Asm}).
8372 Functions declared with the @code{naked} attribute also require basic
8373 @code{asm} (@pxref{Function Attributes}).
8375 While the uses of @code{asm} are many and varied, it may help to think of an
8376 @code{asm} statement as a series of low-level instructions that convert input
8377 parameters to output parameters. So a simple (if not particularly useful)
8378 example for i386 using @code{asm} might look like this:
8384 asm ("mov %1, %0\n\t"
8389 printf("%d\n", dst);
8392 This code copies @code{src} to @code{dst} and add 1 to @code{dst}.
8395 @subsubsection Volatile
8396 @cindex volatile @code{asm}
8397 @cindex @code{asm} volatile
8399 GCC's optimizers sometimes discard @code{asm} statements if they determine
8400 there is no need for the output variables. Also, the optimizers may move
8401 code out of loops if they believe that the code will always return the same
8402 result (i.e. none of its input values change between calls). Using the
8403 @code{volatile} qualifier disables these optimizations. @code{asm} statements
8404 that have no output operands, including @code{asm goto} statements,
8405 are implicitly volatile.
8407 This i386 code demonstrates a case that does not use (or require) the
8408 @code{volatile} qualifier. If it is performing assertion checking, this code
8409 uses @code{asm} to perform the validation. Otherwise, @code{dwRes} is
8410 unreferenced by any code. As a result, the optimizers can discard the
8411 @code{asm} statement, which in turn removes the need for the entire
8412 @code{DoCheck} routine. By omitting the @code{volatile} qualifier when it
8413 isn't needed you allow the optimizers to produce the most efficient code
8417 void DoCheck(uint32_t dwSomeValue)
8421 // Assumes dwSomeValue is not zero.
8431 The next example shows a case where the optimizers can recognize that the input
8432 (@code{dwSomeValue}) never changes during the execution of the function and can
8433 therefore move the @code{asm} outside the loop to produce more efficient code.
8434 Again, using @code{volatile} disables this type of optimization.
8437 void do_print(uint32_t dwSomeValue)
8441 for (uint32_t x=0; x < 5; x++)
8443 // Assumes dwSomeValue is not zero.
8449 printf("%u: %u %u\n", x, dwSomeValue, dwRes);
8454 The following example demonstrates a case where you need to use the
8455 @code{volatile} qualifier.
8456 It uses the x86 @code{rdtsc} instruction, which reads
8457 the computer's time-stamp counter. Without the @code{volatile} qualifier,
8458 the optimizers might assume that the @code{asm} block will always return the
8459 same value and therefore optimize away the second call.
8464 asm volatile ( "rdtsc\n\t" // Returns the time in EDX:EAX.
8465 "shl $32, %%rdx\n\t" // Shift the upper bits left.
8466 "or %%rdx, %0" // 'Or' in the lower bits.
8471 printf("msr: %llx\n", msr);
8475 // Reprint the timestamp
8476 asm volatile ( "rdtsc\n\t" // Returns the time in EDX:EAX.
8477 "shl $32, %%rdx\n\t" // Shift the upper bits left.
8478 "or %%rdx, %0" // 'Or' in the lower bits.
8483 printf("msr: %llx\n", msr);
8486 GCC's optimizers do not treat this code like the non-volatile code in the
8487 earlier examples. They do not move it out of loops or omit it on the
8488 assumption that the result from a previous call is still valid.
8490 Note that the compiler can move even volatile @code{asm} instructions relative
8491 to other code, including across jump instructions. For example, on many
8492 targets there is a system register that controls the rounding mode of
8493 floating-point operations. Setting it with a volatile @code{asm}, as in the
8494 following PowerPC example, does not work reliably.
8497 asm volatile("mtfsf 255, %0" : : "f" (fpenv));
8501 The compiler may move the addition back before the volatile @code{asm}. To
8502 make it work as expected, add an artificial dependency to the @code{asm} by
8503 referencing a variable in the subsequent code, for example:
8506 asm volatile ("mtfsf 255,%1" : "=X" (sum) : "f" (fpenv));
8510 Under certain circumstances, GCC may duplicate (or remove duplicates of) your
8511 assembly code when optimizing. This can lead to unexpected duplicate symbol
8512 errors during compilation if your asm code defines symbols or labels.
8514 (@pxref{AssemblerTemplate}) may help resolve this problem.
8516 @anchor{AssemblerTemplate}
8517 @subsubsection Assembler Template
8518 @cindex @code{asm} assembler template
8520 An assembler template is a literal string containing assembler instructions.
8521 The compiler replaces tokens in the template that refer
8522 to inputs, outputs, and goto labels,
8523 and then outputs the resulting string to the assembler. The
8524 string can contain any instructions recognized by the assembler, including
8525 directives. GCC does not parse the assembler instructions
8526 themselves and does not know what they mean or even whether they are valid
8527 assembler input. However, it does count the statements
8528 (@pxref{Size of an asm}).
8530 You may place multiple assembler instructions together in a single @code{asm}
8531 string, separated by the characters normally used in assembly code for the
8532 system. A combination that works in most places is a newline to break the
8533 line, plus a tab character to move to the instruction field (written as
8535 Some assemblers allow semicolons as a line separator. However, note
8536 that some assembler dialects use semicolons to start a comment.
8538 Do not expect a sequence of @code{asm} statements to remain perfectly
8539 consecutive after compilation, even when you are using the @code{volatile}
8540 qualifier. If certain instructions need to remain consecutive in the output,
8541 put them in a single multi-instruction asm statement.
8543 Accessing data from C programs without using input/output operands (such as
8544 by using global symbols directly from the assembler template) may not work as
8545 expected. Similarly, calling functions directly from an assembler template
8546 requires a detailed understanding of the target assembler and ABI.
8548 Since GCC does not parse the assembler template,
8549 it has no visibility of any
8550 symbols it references. This may result in GCC discarding those symbols as
8551 unreferenced unless they are also listed as input, output, or goto operands.
8553 @subsubheading Special format strings
8555 In addition to the tokens described by the input, output, and goto operands,
8556 these tokens have special meanings in the assembler template:
8560 Outputs a single @samp{%} into the assembler code.
8563 Outputs a number that is unique to each instance of the @code{asm}
8564 statement in the entire compilation. This option is useful when creating local
8565 labels and referring to them multiple times in a single template that
8566 generates multiple assembler instructions.
8571 Outputs @samp{@{}, @samp{|}, and @samp{@}} characters (respectively)
8572 into the assembler code. When unescaped, these characters have special
8573 meaning to indicate multiple assembler dialects, as described below.
8576 @subsubheading Multiple assembler dialects in @code{asm} templates
8578 On targets such as x86, GCC supports multiple assembler dialects.
8579 The @option{-masm} option controls which dialect GCC uses as its
8580 default for inline assembler. The target-specific documentation for the
8581 @option{-masm} option contains the list of supported dialects, as well as the
8582 default dialect if the option is not specified. This information may be
8583 important to understand, since assembler code that works correctly when
8584 compiled using one dialect will likely fail if compiled using another.
8587 If your code needs to support multiple assembler dialects (for example, if
8588 you are writing public headers that need to support a variety of compilation
8589 options), use constructs of this form:
8592 @{ dialect0 | dialect1 | dialect2... @}
8595 This construct outputs @code{dialect0}
8596 when using dialect #0 to compile the code,
8597 @code{dialect1} for dialect #1, etc. If there are fewer alternatives within the
8598 braces than the number of dialects the compiler supports, the construct
8601 For example, if an x86 compiler supports two dialects
8602 (@samp{att}, @samp{intel}), an
8603 assembler template such as this:
8606 "bt@{l %[Offset],%[Base] | %[Base],%[Offset]@}; jc %l2"
8610 is equivalent to one of
8613 "btl %[Offset],%[Base] ; jc %l2" @r{/* att dialect */}
8614 "bt %[Base],%[Offset]; jc %l2" @r{/* intel dialect */}
8617 Using that same compiler, this code:
8620 "xchg@{l@}\t@{%%@}ebx, %1"
8624 corresponds to either
8627 "xchgl\t%%ebx, %1" @r{/* att dialect */}
8628 "xchg\tebx, %1" @r{/* intel dialect */}
8631 There is no support for nesting dialect alternatives.
8633 @anchor{OutputOperands}
8634 @subsubsection Output Operands
8635 @cindex @code{asm} output operands
8637 An @code{asm} statement has zero or more output operands indicating the names
8638 of C variables modified by the assembler code.
8640 In this i386 example, @code{old} (referred to in the template string as
8641 @code{%0}) and @code{*Base} (as @code{%1}) are outputs and @code{Offset}
8642 (@code{%2}) is an input:
8647 __asm__ ("btsl %2,%1\n\t" // Turn on zero-based bit #Offset in Base.
8648 "sbb %0,%0" // Use the CF to calculate old.
8649 : "=r" (old), "+rm" (*Base)
8656 Operands are separated by commas. Each operand has this format:
8659 @r{[} [@var{asmSymbolicName}] @r{]} @var{constraint} (@var{cvariablename})
8663 @item asmSymbolicName
8664 Specifies a symbolic name for the operand.
8665 Reference the name in the assembler template
8666 by enclosing it in square brackets
8667 (i.e. @samp{%[Value]}). The scope of the name is the @code{asm} statement
8668 that contains the definition. Any valid C variable name is acceptable,
8669 including names already defined in the surrounding code. No two operands
8670 within the same @code{asm} statement can use the same symbolic name.
8672 When not using an @var{asmSymbolicName}, use the (zero-based) position
8674 in the list of operands in the assembler template. For example if there are
8675 three output operands, use @samp{%0} in the template to refer to the first,
8676 @samp{%1} for the second, and @samp{%2} for the third.
8679 A string constant specifying constraints on the placement of the operand;
8680 @xref{Constraints}, for details.
8682 Output constraints must begin with either @samp{=} (a variable overwriting an
8683 existing value) or @samp{+} (when reading and writing). When using
8684 @samp{=}, do not assume the location contains the existing value
8685 on entry to the @code{asm}, except
8686 when the operand is tied to an input; @pxref{InputOperands,,Input Operands}.
8688 After the prefix, there must be one or more additional constraints
8689 (@pxref{Constraints}) that describe where the value resides. Common
8690 constraints include @samp{r} for register and @samp{m} for memory.
8691 When you list more than one possible location (for example, @code{"=rm"}),
8692 the compiler chooses the most efficient one based on the current context.
8693 If you list as many alternates as the @code{asm} statement allows, you permit
8694 the optimizers to produce the best possible code.
8695 If you must use a specific register, but your Machine Constraints do not
8696 provide sufficient control to select the specific register you want,
8697 local register variables may provide a solution (@pxref{Local Register
8701 Specifies a C lvalue expression to hold the output, typically a variable name.
8702 The enclosing parentheses are a required part of the syntax.
8706 When the compiler selects the registers to use to
8707 represent the output operands, it does not use any of the clobbered registers
8708 (@pxref{Clobbers and Scratch Registers}).
8710 Output operand expressions must be lvalues. The compiler cannot check whether
8711 the operands have data types that are reasonable for the instruction being
8712 executed. For output expressions that are not directly addressable (for
8713 example a bit-field), the constraint must allow a register. In that case, GCC
8714 uses the register as the output of the @code{asm}, and then stores that
8715 register into the output.
8717 Operands using the @samp{+} constraint modifier count as two operands
8718 (that is, both as input and output) towards the total maximum of 30 operands
8719 per @code{asm} statement.
8721 Use the @samp{&} constraint modifier (@pxref{Modifiers}) on all output
8722 operands that must not overlap an input. Otherwise,
8723 GCC may allocate the output operand in the same register as an unrelated
8724 input operand, on the assumption that the assembler code consumes its
8725 inputs before producing outputs. This assumption may be false if the assembler
8726 code actually consists of more than one instruction.
8728 The same problem can occur if one output parameter (@var{a}) allows a register
8729 constraint and another output parameter (@var{b}) allows a memory constraint.
8730 The code generated by GCC to access the memory address in @var{b} can contain
8731 registers which @emph{might} be shared by @var{a}, and GCC considers those
8732 registers to be inputs to the asm. As above, GCC assumes that such input
8733 registers are consumed before any outputs are written. This assumption may
8734 result in incorrect behavior if the asm writes to @var{a} before using
8735 @var{b}. Combining the @samp{&} modifier with the register constraint on @var{a}
8736 ensures that modifying @var{a} does not affect the address referenced by
8737 @var{b}. Otherwise, the location of @var{b}
8738 is undefined if @var{a} is modified before using @var{b}.
8740 @code{asm} supports operand modifiers on operands (for example @samp{%k2}
8741 instead of simply @samp{%2}). Typically these qualifiers are hardware
8742 dependent. The list of supported modifiers for x86 is found at
8743 @ref{x86Operandmodifiers,x86 Operand modifiers}.
8745 If the C code that follows the @code{asm} makes no use of any of the output
8746 operands, use @code{volatile} for the @code{asm} statement to prevent the
8747 optimizers from discarding the @code{asm} statement as unneeded
8748 (see @ref{Volatile}).
8750 This code makes no use of the optional @var{asmSymbolicName}. Therefore it
8751 references the first output operand as @code{%0} (were there a second, it
8752 would be @code{%1}, etc). The number of the first input operand is one greater
8753 than that of the last output operand. In this i386 example, that makes
8754 @code{Mask} referenced as @code{%1}:
8757 uint32_t Mask = 1234;
8766 That code overwrites the variable @code{Index} (@samp{=}),
8767 placing the value in a register (@samp{r}).
8768 Using the generic @samp{r} constraint instead of a constraint for a specific
8769 register allows the compiler to pick the register to use, which can result
8770 in more efficient code. This may not be possible if an assembler instruction
8771 requires a specific register.
8773 The following i386 example uses the @var{asmSymbolicName} syntax.
8775 same result as the code above, but some may consider it more readable or more
8776 maintainable since reordering index numbers is not necessary when adding or
8777 removing operands. The names @code{aIndex} and @code{aMask}
8778 are only used in this example to emphasize which
8779 names get used where.
8780 It is acceptable to reuse the names @code{Index} and @code{Mask}.
8783 uint32_t Mask = 1234;
8786 asm ("bsfl %[aMask], %[aIndex]"
8787 : [aIndex] "=r" (Index)
8788 : [aMask] "r" (Mask)
8792 Here are some more examples of output operands.
8799 asm ("mov %[e], %[d]"
8804 Here, @code{d} may either be in a register or in memory. Since the compiler
8805 might already have the current value of the @code{uint32_t} location
8806 pointed to by @code{e}
8807 in a register, you can enable it to choose the best location
8808 for @code{d} by specifying both constraints.
8810 @anchor{FlagOutputOperands}
8811 @subsubsection Flag Output Operands
8812 @cindex @code{asm} flag output operands
8814 Some targets have a special register that holds the ``flags'' for the
8815 result of an operation or comparison. Normally, the contents of that
8816 register are either unmodifed by the asm, or the asm is considered to
8817 clobber the contents.
8819 On some targets, a special form of output operand exists by which
8820 conditions in the flags register may be outputs of the asm. The set of
8821 conditions supported are target specific, but the general rule is that
8822 the output variable must be a scalar integer, and the value is boolean.
8823 When supported, the target defines the preprocessor symbol
8824 @code{__GCC_ASM_FLAG_OUTPUTS__}.
8826 Because of the special nature of the flag output operands, the constraint
8827 may not include alternatives.
8829 Most often, the target has only one flags register, and thus is an implied
8830 operand of many instructions. In this case, the operand should not be
8831 referenced within the assembler template via @code{%0} etc, as there's
8832 no corresponding text in the assembly language.
8836 The flag output constraints for the x86 family are of the form
8837 @samp{=@@cc@var{cond}} where @var{cond} is one of the standard
8838 conditions defined in the ISA manual for @code{j@var{cc}} or
8843 ``above'' or unsigned greater than
8845 ``above or equal'' or unsigned greater than or equal
8847 ``below'' or unsigned less than
8849 ``below or equal'' or unsigned less than or equal
8854 ``equal'' or zero flag set
8858 signed greater than or equal
8862 signed less than or equal
8883 ``not'' @var{flag}, or inverted versions of those above
8888 @anchor{InputOperands}
8889 @subsubsection Input Operands
8890 @cindex @code{asm} input operands
8891 @cindex @code{asm} expressions
8893 Input operands make values from C variables and expressions available to the
8896 Operands are separated by commas. Each operand has this format:
8899 @r{[} [@var{asmSymbolicName}] @r{]} @var{constraint} (@var{cexpression})
8903 @item asmSymbolicName
8904 Specifies a symbolic name for the operand.
8905 Reference the name in the assembler template
8906 by enclosing it in square brackets
8907 (i.e. @samp{%[Value]}). The scope of the name is the @code{asm} statement
8908 that contains the definition. Any valid C variable name is acceptable,
8909 including names already defined in the surrounding code. No two operands
8910 within the same @code{asm} statement can use the same symbolic name.
8912 When not using an @var{asmSymbolicName}, use the (zero-based) position
8914 in the list of operands in the assembler template. For example if there are
8915 two output operands and three inputs,
8916 use @samp{%2} in the template to refer to the first input operand,
8917 @samp{%3} for the second, and @samp{%4} for the third.
8920 A string constant specifying constraints on the placement of the operand;
8921 @xref{Constraints}, for details.
8923 Input constraint strings may not begin with either @samp{=} or @samp{+}.
8924 When you list more than one possible location (for example, @samp{"irm"}),
8925 the compiler chooses the most efficient one based on the current context.
8926 If you must use a specific register, but your Machine Constraints do not
8927 provide sufficient control to select the specific register you want,
8928 local register variables may provide a solution (@pxref{Local Register
8931 Input constraints can also be digits (for example, @code{"0"}). This indicates
8932 that the specified input must be in the same place as the output constraint
8933 at the (zero-based) index in the output constraint list.
8934 When using @var{asmSymbolicName} syntax for the output operands,
8935 you may use these names (enclosed in brackets @samp{[]}) instead of digits.
8938 This is the C variable or expression being passed to the @code{asm} statement
8939 as input. The enclosing parentheses are a required part of the syntax.
8943 When the compiler selects the registers to use to represent the input
8944 operands, it does not use any of the clobbered registers
8945 (@pxref{Clobbers and Scratch Registers}).
8947 If there are no output operands but there are input operands, place two
8948 consecutive colons where the output operands would go:
8951 __asm__ ("some instructions"
8953 : "r" (Offset / 8));
8956 @strong{Warning:} Do @emph{not} modify the contents of input-only operands
8957 (except for inputs tied to outputs). The compiler assumes that on exit from
8958 the @code{asm} statement these operands contain the same values as they
8959 had before executing the statement.
8960 It is @emph{not} possible to use clobbers
8961 to inform the compiler that the values in these inputs are changing. One
8962 common work-around is to tie the changing input variable to an output variable
8963 that never gets used. Note, however, that if the code that follows the
8964 @code{asm} statement makes no use of any of the output operands, the GCC
8965 optimizers may discard the @code{asm} statement as unneeded
8966 (see @ref{Volatile}).
8968 @code{asm} supports operand modifiers on operands (for example @samp{%k2}
8969 instead of simply @samp{%2}). Typically these qualifiers are hardware
8970 dependent. The list of supported modifiers for x86 is found at
8971 @ref{x86Operandmodifiers,x86 Operand modifiers}.
8973 In this example using the fictitious @code{combine} instruction, the
8974 constraint @code{"0"} for input operand 1 says that it must occupy the same
8975 location as output operand 0. Only input operands may use numbers in
8976 constraints, and they must each refer to an output operand. Only a number (or
8977 the symbolic assembler name) in the constraint can guarantee that one operand
8978 is in the same place as another. The mere fact that @code{foo} is the value of
8979 both operands is not enough to guarantee that they are in the same place in
8980 the generated assembler code.
8983 asm ("combine %2, %0"
8985 : "0" (foo), "g" (bar));
8988 Here is an example using symbolic names.
8991 asm ("cmoveq %1, %2, %[result]"
8992 : [result] "=r"(result)
8993 : "r" (test), "r" (new), "[result]" (old));
8996 @anchor{Clobbers and Scratch Registers}
8997 @subsubsection Clobbers and Scratch Registers
8998 @cindex @code{asm} clobbers
8999 @cindex @code{asm} scratch registers
9001 While the compiler is aware of changes to entries listed in the output
9002 operands, the inline @code{asm} code may modify more than just the outputs. For
9003 example, calculations may require additional registers, or the processor may
9004 overwrite a register as a side effect of a particular assembler instruction.
9005 In order to inform the compiler of these changes, list them in the clobber
9006 list. Clobber list items are either register names or the special clobbers
9007 (listed below). Each clobber list item is a string constant
9008 enclosed in double quotes and separated by commas.
9010 Clobber descriptions may not in any way overlap with an input or output
9011 operand. For example, you may not have an operand describing a register class
9012 with one member when listing that register in the clobber list. Variables
9013 declared to live in specific registers (@pxref{Explicit Register
9014 Variables}) and used
9015 as @code{asm} input or output operands must have no part mentioned in the
9016 clobber description. In particular, there is no way to specify that input
9017 operands get modified without also specifying them as output operands.
9019 When the compiler selects which registers to use to represent input and output
9020 operands, it does not use any of the clobbered registers. As a result,
9021 clobbered registers are available for any use in the assembler code.
9023 Here is a realistic example for the VAX showing the use of clobbered
9027 asm volatile ("movc3 %0, %1, %2"
9029 : "g" (from), "g" (to), "g" (count)
9030 : "r0", "r1", "r2", "r3", "r4", "r5", "memory");
9033 Also, there are two special clobber arguments:
9037 The @code{"cc"} clobber indicates that the assembler code modifies the flags
9038 register. On some machines, GCC represents the condition codes as a specific
9039 hardware register; @code{"cc"} serves to name this register.
9040 On other machines, condition code handling is different,
9041 and specifying @code{"cc"} has no effect. But
9042 it is valid no matter what the target.
9045 The @code{"memory"} clobber tells the compiler that the assembly code
9047 reads or writes to items other than those listed in the input and output
9048 operands (for example, accessing the memory pointed to by one of the input
9049 parameters). To ensure memory contains correct values, GCC may need to flush
9050 specific register values to memory before executing the @code{asm}. Further,
9051 the compiler does not assume that any values read from memory before an
9052 @code{asm} remain unchanged after that @code{asm}; it reloads them as
9054 Using the @code{"memory"} clobber effectively forms a read/write
9055 memory barrier for the compiler.
9057 Note that this clobber does not prevent the @emph{processor} from doing
9058 speculative reads past the @code{asm} statement. To prevent that, you need
9059 processor-specific fence instructions.
9063 Flushing registers to memory has performance implications and may be
9064 an issue for time-sensitive code. You can provide better information
9065 to GCC to avoid this, as shown in the following examples. At a
9066 minimum, aliasing rules allow GCC to know what memory @emph{doesn't}
9069 Here is a fictitious sum of squares instruction, that takes two
9070 pointers to floating point values in memory and produces a floating
9071 point register output.
9072 Notice that @code{x}, and @code{y} both appear twice in the @code{asm}
9073 parameters, once to specify memory accessed, and once to specify a
9074 base register used by the @code{asm}. You won't normally be wasting a
9075 register by doing this as GCC can use the same register for both
9076 purposes. However, it would be foolish to use both @code{%1} and
9077 @code{%3} for @code{x} in this @code{asm} and expect them to be the
9078 same. In fact, @code{%3} may well not be a register. It might be a
9079 symbolic memory reference to the object pointed to by @code{x}.
9082 asm ("sumsq %0, %1, %2"
9084 : "r" (x), "r" (y), "m" (*x), "m" (*y));
9087 Here is a fictitious @code{*z++ = *x++ * *y++} instruction.
9088 Notice that the @code{x}, @code{y} and @code{z} pointer registers
9089 must be specified as input/output because the @code{asm} modifies
9093 asm ("vecmul %0, %1, %2"
9094 : "+r" (z), "+r" (x), "+r" (y), "=m" (*z)
9095 : "m" (*x), "m" (*y));
9098 An x86 example where the string memory argument is of unknown length.
9102 : "=c" (count), "+D" (p)
9103 : "m" (*(const char (*)[]) p), "0" (-1), "a" (0));
9106 If you know the above will only be reading a ten byte array then you
9107 could instead use a memory input like:
9108 @code{"m" (*(const char (*)[10]) p)}.
9110 Here is an example of a PowerPC vector scale implemented in assembly,
9111 complete with vector and condition code clobbers, and some initialized
9112 offset registers that are unchanged by the @code{asm}.
9116 dscal (size_t n, double *x, double alpha)
9118 asm ("/* lots of asm here */"
9119 : "+m" (*(double (*)[n]) x), "+&r" (n), "+b" (x)
9120 : "d" (alpha), "b" (32), "b" (48), "b" (64),
9121 "b" (80), "b" (96), "b" (112)
9123 "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
9124 "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47");
9128 Rather than allocating fixed registers via clobbers to provide scratch
9129 registers for an @code{asm} statement, an alternative is to define a
9130 variable and make it an early-clobber output as with @code{a2} and
9131 @code{a3} in the example below. This gives the compiler register
9132 allocator more freedom. You can also define a variable and make it an
9133 output tied to an input as with @code{a0} and @code{a1}, tied
9134 respectively to @code{ap} and @code{lda}. Of course, with tied
9135 outputs your @code{asm} can't use the input value after modifying the
9136 output register since they are one and the same register. What's
9137 more, if you omit the early-clobber on the output, it is possible that
9138 GCC might allocate the same register to another of the inputs if GCC
9139 could prove they had the same value on entry to the @code{asm}. This
9140 is why @code{a1} has an early-clobber. Its tied input, @code{lda}
9141 might conceivably be known to have the value 16 and without an
9142 early-clobber share the same register as @code{%11}. On the other
9143 hand, @code{ap} can't be the same as any of the other inputs, so an
9144 early-clobber on @code{a0} is not needed. It is also not desirable in
9145 this case. An early-clobber on @code{a0} would cause GCC to allocate
9146 a separate register for the @code{"m" (*(const double (*)[]) ap)}
9147 input. Note that tying an input to an output is the way to set up an
9148 initialized temporary register modified by an @code{asm} statement.
9149 An input not tied to an output is assumed by GCC to be unchanged, for
9150 example @code{"b" (16)} below sets up @code{%11} to 16, and GCC might
9151 use that register in following code if the value 16 happened to be
9152 needed. You can even use a normal @code{asm} output for a scratch if
9153 all inputs that might share the same register are consumed before the
9154 scratch is used. The VSX registers clobbered by the @code{asm}
9155 statement could have used this technique except for GCC's limit on the
9156 number of @code{asm} parameters.
9160 dgemv_kernel_4x4 (long n, const double *ap, long lda,
9161 const double *x, double *y, double alpha)
9170 /* lots of asm here */
9171 "#n=%1 ap=%8=%12 lda=%13 x=%7=%10 y=%0=%2 alpha=%9 o16=%11\n"
9172 "#a0=%3 a1=%4 a2=%5 a3=%6"
9174 "+m" (*(double (*)[n]) y),
9182 "m" (*(const double (*)[n]) x),
9183 "m" (*(const double (*)[]) ap),
9191 "vs32","vs33","vs34","vs35","vs36","vs37",
9192 "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47"
9198 @subsubsection Goto Labels
9199 @cindex @code{asm} goto labels
9201 @code{asm goto} allows assembly code to jump to one or more C labels. The
9202 @var{GotoLabels} section in an @code{asm goto} statement contains
9204 list of all C labels to which the assembler code may jump. GCC assumes that
9205 @code{asm} execution falls through to the next statement (if this is not the
9206 case, consider using the @code{__builtin_unreachable} intrinsic after the
9207 @code{asm} statement). Optimization of @code{asm goto} may be improved by
9208 using the @code{hot} and @code{cold} label attributes (@pxref{Label
9211 An @code{asm goto} statement cannot have outputs.
9212 This is due to an internal restriction of
9213 the compiler: control transfer instructions cannot have outputs.
9214 If the assembler code does modify anything, use the @code{"memory"} clobber
9216 optimizers to flush all register values to memory and reload them if
9217 necessary after the @code{asm} statement.
9219 Also note that an @code{asm goto} statement is always implicitly
9220 considered volatile.
9222 To reference a label in the assembler template,
9223 prefix it with @samp{%l} (lowercase @samp{L}) followed
9224 by its (zero-based) position in @var{GotoLabels} plus the number of input
9225 operands. For example, if the @code{asm} has three inputs and references two
9226 labels, refer to the first label as @samp{%l3} and the second as @samp{%l4}).
9228 Alternately, you can reference labels using the actual C label name enclosed
9229 in brackets. For example, to reference a label named @code{carry}, you can
9230 use @samp{%l[carry]}. The label must still be listed in the @var{GotoLabels}
9231 section when using this approach.
9233 Here is an example of @code{asm goto} for i386:
9240 : "r" (p1), "r" (p2)
9250 The following example shows an @code{asm goto} that uses a memory clobber.
9256 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
9267 @anchor{x86Operandmodifiers}
9268 @subsubsection x86 Operand Modifiers
9270 References to input, output, and goto operands in the assembler template
9271 of extended @code{asm} statements can use
9272 modifiers to affect the way the operands are formatted in
9273 the code output to the assembler. For example, the
9274 following code uses the @samp{h} and @samp{b} modifiers for x86:
9278 asm volatile ("xchg %h0, %b0" : "+a" (num) );
9282 These modifiers generate this assembler code:
9288 The rest of this discussion uses the following code for illustrative purposes.
9297 asm volatile goto ("some assembler instructions here"
9299 : "q" (iInt), "X" (sizeof(unsigned char) + 1), "i" (42)
9300 : /* No clobbers. */
9305 With no modifiers, this is what the output from the operands would be
9306 for the @samp{att} and @samp{intel} dialects of assembler:
9308 @multitable {Operand} {$.L2} {OFFSET FLAT:.L2}
9309 @headitem Operand @tab @samp{att} @tab @samp{intel}
9318 @tab @code{OFFSET FLAT:.L3}
9321 The table below shows the list of supported modifiers and their effects.
9323 @multitable {Modifier} {Print the opcode suffix for the size of th} {Operand} {@samp{att}} {@samp{intel}}
9324 @headitem Modifier @tab Description @tab Operand @tab @samp{att} @tab @samp{intel}
9326 @tab Print an absolute memory reference.
9331 @tab Print the QImode name of the register.
9336 @tab Require a constant operand and print the constant expression with no punctuation.
9341 @tab Print the address in Double Integer (DImode) mode (8 bytes) when the target is 64-bit.
9342 Otherwise mode is unspecified (VOIDmode).
9347 @tab Print the QImode name for a ``high'' register.
9352 @tab Add 8 bytes to an offsettable memory reference. Useful when accessing the
9353 high 8 bytes of SSE values. For a memref in (%rax), it generates
9358 @tab Print the SImode name of the register.
9363 @tab Print the label name with no punctuation.
9368 @tab Print raw symbol name (without syntax-specific prefixes).
9373 @tab If used for a function, print the PLT suffix and generate PIC code.
9374 For example, emit @code{foo@@PLT} instead of 'foo' for the function
9375 foo(). If used for a constant, drop all syntax-specific prefixes and
9376 issue the bare constant. See @code{p} above.
9378 @tab Print the DImode name of the register.
9383 @tab Print the HImode name of the register.
9388 @tab Print the opcode suffix for the size of the current integer operand (one of @code{b}/@code{w}/@code{l}/@code{q}).
9394 @code{V} is a special modifier which prints the name of the full integer
9395 register without @code{%}.
9397 @anchor{x86floatingpointasmoperands}
9398 @subsubsection x86 Floating-Point @code{asm} Operands
9400 On x86 targets, there are several rules on the usage of stack-like registers
9401 in the operands of an @code{asm}. These rules apply only to the operands
9402 that are stack-like registers:
9406 Given a set of input registers that die in an @code{asm}, it is
9407 necessary to know which are implicitly popped by the @code{asm}, and
9408 which must be explicitly popped by GCC@.
9410 An input register that is implicitly popped by the @code{asm} must be
9411 explicitly clobbered, unless it is constrained to match an
9415 For any input register that is implicitly popped by an @code{asm}, it is
9416 necessary to know how to adjust the stack to compensate for the pop.
9417 If any non-popped input is closer to the top of the reg-stack than
9418 the implicitly popped register, it would not be possible to know what the
9419 stack looked like---it's not clear how the rest of the stack ``slides
9422 All implicitly popped input registers must be closer to the top of
9423 the reg-stack than any input that is not implicitly popped.
9425 It is possible that if an input dies in an @code{asm}, the compiler might
9426 use the input register for an output reload. Consider this example:
9429 asm ("foo" : "=t" (a) : "f" (b));
9433 This code says that input @code{b} is not popped by the @code{asm}, and that
9434 the @code{asm} pushes a result onto the reg-stack, i.e., the stack is one
9435 deeper after the @code{asm} than it was before. But, it is possible that
9436 reload may think that it can use the same register for both the input and
9439 To prevent this from happening,
9440 if any input operand uses the @samp{f} constraint, all output register
9441 constraints must use the @samp{&} early-clobber modifier.
9443 The example above is correctly written as:
9446 asm ("foo" : "=&t" (a) : "f" (b));
9450 Some operands need to be in particular places on the stack. All
9451 output operands fall in this category---GCC has no other way to
9452 know which registers the outputs appear in unless you indicate
9453 this in the constraints.
9455 Output operands must specifically indicate which register an output
9456 appears in after an @code{asm}. @samp{=f} is not allowed: the operand
9457 constraints must select a class with a single register.
9460 Output operands may not be ``inserted'' between existing stack registers.
9461 Since no 387 opcode uses a read/write operand, all output operands
9462 are dead before the @code{asm}, and are pushed by the @code{asm}.
9463 It makes no sense to push anywhere but the top of the reg-stack.
9465 Output operands must start at the top of the reg-stack: output
9466 operands may not ``skip'' a register.
9469 Some @code{asm} statements may need extra stack space for internal
9470 calculations. This can be guaranteed by clobbering stack registers
9471 unrelated to the inputs and outputs.
9476 takes one input, which is internally popped, and produces two outputs.
9479 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
9483 This @code{asm} takes two inputs, which are popped by the @code{fyl2xp1} opcode,
9484 and replaces them with one output. The @code{st(1)} clobber is necessary
9485 for the compiler to know that @code{fyl2xp1} pops both inputs.
9488 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
9496 @subsection Controlling Names Used in Assembler Code
9497 @cindex assembler names for identifiers
9498 @cindex names used in assembler code
9499 @cindex identifiers, names in assembler code
9501 You can specify the name to be used in the assembler code for a C
9502 function or variable by writing the @code{asm} (or @code{__asm__})
9503 keyword after the declarator.
9504 It is up to you to make sure that the assembler names you choose do not
9505 conflict with any other assembler symbols, or reference registers.
9507 @subsubheading Assembler names for data:
9509 This sample shows how to specify the assembler name for data:
9512 int foo asm ("myfoo") = 2;
9516 This specifies that the name to be used for the variable @code{foo} in
9517 the assembler code should be @samp{myfoo} rather than the usual
9520 On systems where an underscore is normally prepended to the name of a C
9521 variable, this feature allows you to define names for the
9522 linker that do not start with an underscore.
9524 GCC does not support using this feature with a non-static local variable
9525 since such variables do not have assembler names. If you are
9526 trying to put the variable in a particular register, see
9527 @ref{Explicit Register Variables}.
9529 @subsubheading Assembler names for functions:
9531 To specify the assembler name for functions, write a declaration for the
9532 function before its definition and put @code{asm} there, like this:
9535 int func (int x, int y) asm ("MYFUNC");
9537 int func (int x, int y)
9543 This specifies that the name to be used for the function @code{func} in
9544 the assembler code should be @code{MYFUNC}.
9546 @node Explicit Register Variables
9547 @subsection Variables in Specified Registers
9548 @anchor{Explicit Reg Vars}
9549 @cindex explicit register variables
9550 @cindex variables in specified registers
9551 @cindex specified registers
9553 GNU C allows you to associate specific hardware registers with C
9554 variables. In almost all cases, allowing the compiler to assign
9555 registers produces the best code. However under certain unusual
9556 circumstances, more precise control over the variable storage is
9559 Both global and local variables can be associated with a register. The
9560 consequences of performing this association are very different between
9561 the two, as explained in the sections below.
9564 * Global Register Variables:: Variables declared at global scope.
9565 * Local Register Variables:: Variables declared within a function.
9568 @node Global Register Variables
9569 @subsubsection Defining Global Register Variables
9570 @anchor{Global Reg Vars}
9571 @cindex global register variables
9572 @cindex registers, global variables in
9573 @cindex registers, global allocation
9575 You can define a global register variable and associate it with a specified
9579 register int *foo asm ("r12");
9583 Here @code{r12} is the name of the register that should be used. Note that
9584 this is the same syntax used for defining local register variables, but for
9585 a global variable the declaration appears outside a function. The
9586 @code{register} keyword is required, and cannot be combined with
9587 @code{static}. The register name must be a valid register name for the
9590 Registers are a scarce resource on most systems and allowing the
9591 compiler to manage their usage usually results in the best code. However,
9592 under special circumstances it can make sense to reserve some globally.
9593 For example this may be useful in programs such as programming language
9594 interpreters that have a couple of global variables that are accessed
9597 After defining a global register variable, for the current compilation
9601 @item If the register is a call-saved register, call ABI is affected:
9602 the register will not be restored in function epilogue sequences after
9603 the variable has been assigned. Therefore, functions cannot safely
9604 return to callers that assume standard ABI.
9605 @item Conversely, if the register is a call-clobbered register, making
9606 calls to functions that use standard ABI may lose contents of the variable.
9607 Such calls may be created by the compiler even if none are evident in
9608 the original program, for example when libgcc functions are used to
9609 make up for unavailable instructions.
9610 @item Accesses to the variable may be optimized as usual and the register
9611 remains available for allocation and use in any computations, provided that
9612 observable values of the variable are not affected.
9613 @item If the variable is referenced in inline assembly, the type of access
9614 must be provided to the compiler via constraints (@pxref{Constraints}).
9615 Accesses from basic asms are not supported.
9618 Note that these points @emph{only} apply to code that is compiled with the
9619 definition. The behavior of code that is merely linked in (for example
9620 code from libraries) is not affected.
9622 If you want to recompile source files that do not actually use your global
9623 register variable so they do not use the specified register for any other
9624 purpose, you need not actually add the global register declaration to
9625 their source code. It suffices to specify the compiler option
9626 @option{-ffixed-@var{reg}} (@pxref{Code Gen Options}) to reserve the
9629 @subsubheading Declaring the variable
9631 Global register variables can not have initial values, because an
9632 executable file has no means to supply initial contents for a register.
9634 When selecting a register, choose one that is normally saved and
9635 restored by function calls on your machine. This ensures that code
9636 which is unaware of this reservation (such as library routines) will
9637 restore it before returning.
9639 On machines with register windows, be sure to choose a global
9640 register that is not affected magically by the function call mechanism.
9642 @subsubheading Using the variable
9644 @cindex @code{qsort}, and global register variables
9645 When calling routines that are not aware of the reservation, be
9646 cautious if those routines call back into code which uses them. As an
9647 example, if you call the system library version of @code{qsort}, it may
9648 clobber your registers during execution, but (if you have selected
9649 appropriate registers) it will restore them before returning. However
9650 it will @emph{not} restore them before calling @code{qsort}'s comparison
9651 function. As a result, global values will not reliably be available to
9652 the comparison function unless the @code{qsort} function itself is rebuilt.
9654 Similarly, it is not safe to access the global register variables from signal
9655 handlers or from more than one thread of control. Unless you recompile
9656 them specially for the task at hand, the system library routines may
9657 temporarily use the register for other things. Furthermore, since the register
9658 is not reserved exclusively for the variable, accessing it from handlers of
9659 asynchronous signals may observe unrelated temporary values residing in the
9662 @cindex register variable after @code{longjmp}
9663 @cindex global register after @code{longjmp}
9664 @cindex value after @code{longjmp}
9667 On most machines, @code{longjmp} restores to each global register
9668 variable the value it had at the time of the @code{setjmp}. On some
9669 machines, however, @code{longjmp} does not change the value of global
9670 register variables. To be portable, the function that called @code{setjmp}
9671 should make other arrangements to save the values of the global register
9672 variables, and to restore them in a @code{longjmp}. This way, the same
9673 thing happens regardless of what @code{longjmp} does.
9675 @node Local Register Variables
9676 @subsubsection Specifying Registers for Local Variables
9677 @anchor{Local Reg Vars}
9678 @cindex local variables, specifying registers
9679 @cindex specifying registers for local variables
9680 @cindex registers for local variables
9682 You can define a local register variable and associate it with a specified
9686 register int *foo asm ("r12");
9690 Here @code{r12} is the name of the register that should be used. Note
9691 that this is the same syntax used for defining global register variables,
9692 but for a local variable the declaration appears within a function. The
9693 @code{register} keyword is required, and cannot be combined with
9694 @code{static}. The register name must be a valid register name for the
9697 As with global register variables, it is recommended that you choose
9698 a register that is normally saved and restored by function calls on your
9699 machine, so that calls to library routines will not clobber it.
9701 The only supported use for this feature is to specify registers
9702 for input and output operands when calling Extended @code{asm}
9703 (@pxref{Extended Asm}). This may be necessary if the constraints for a
9704 particular machine don't provide sufficient control to select the desired
9705 register. To force an operand into a register, create a local variable
9706 and specify the register name after the variable's declaration. Then use
9707 the local variable for the @code{asm} operand and specify any constraint
9708 letter that matches the register:
9711 register int *p1 asm ("r0") = @dots{};
9712 register int *p2 asm ("r1") = @dots{};
9713 register int *result asm ("r0");
9714 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
9717 @emph{Warning:} In the above example, be aware that a register (for example
9718 @code{r0}) can be call-clobbered by subsequent code, including function
9719 calls and library calls for arithmetic operators on other variables (for
9720 example the initialization of @code{p2}). In this case, use temporary
9721 variables for expressions between the register assignments:
9725 register int *p1 asm ("r0") = @dots{};
9726 register int *p2 asm ("r1") = t1;
9727 register int *result asm ("r0");
9728 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
9731 Defining a register variable does not reserve the register. Other than
9732 when invoking the Extended @code{asm}, the contents of the specified
9733 register are not guaranteed. For this reason, the following uses
9734 are explicitly @emph{not} supported. If they appear to work, it is only
9735 happenstance, and may stop working as intended due to (seemingly)
9736 unrelated changes in surrounding code, or even minor changes in the
9737 optimization of a future version of gcc:
9740 @item Passing parameters to or from Basic @code{asm}
9741 @item Passing parameters to or from Extended @code{asm} without using input
9743 @item Passing parameters to or from routines written in assembler (or
9744 other languages) using non-standard calling conventions.
9747 Some developers use Local Register Variables in an attempt to improve
9748 gcc's allocation of registers, especially in large functions. In this
9749 case the register name is essentially a hint to the register allocator.
9750 While in some instances this can generate better code, improvements are
9751 subject to the whims of the allocator/optimizers. Since there are no
9752 guarantees that your improvements won't be lost, this usage of Local
9753 Register Variables is discouraged.
9755 On the MIPS platform, there is related use for local register variables
9756 with slightly different characteristics (@pxref{MIPS Coprocessors,,
9757 Defining coprocessor specifics for MIPS targets, gccint,
9758 GNU Compiler Collection (GCC) Internals}).
9760 @node Size of an asm
9761 @subsection Size of an @code{asm}
9763 Some targets require that GCC track the size of each instruction used
9764 in order to generate correct code. Because the final length of the
9765 code produced by an @code{asm} statement is only known by the
9766 assembler, GCC must make an estimate as to how big it will be. It
9767 does this by counting the number of instructions in the pattern of the
9768 @code{asm} and multiplying that by the length of the longest
9769 instruction supported by that processor. (When working out the number
9770 of instructions, it assumes that any occurrence of a newline or of
9771 whatever statement separator character is supported by the assembler --
9772 typically @samp{;} --- indicates the end of an instruction.)
9774 Normally, GCC's estimate is adequate to ensure that correct
9775 code is generated, but it is possible to confuse the compiler if you use
9776 pseudo instructions or assembler macros that expand into multiple real
9777 instructions, or if you use assembler directives that expand to more
9778 space in the object file than is needed for a single instruction.
9779 If this happens then the assembler may produce a diagnostic saying that
9780 a label is unreachable.
9782 @node Alternate Keywords
9783 @section Alternate Keywords
9784 @cindex alternate keywords
9785 @cindex keywords, alternate
9787 @option{-ansi} and the various @option{-std} options disable certain
9788 keywords. This causes trouble when you want to use GNU C extensions, or
9789 a general-purpose header file that should be usable by all programs,
9790 including ISO C programs. The keywords @code{asm}, @code{typeof} and
9791 @code{inline} are not available in programs compiled with
9792 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
9793 program compiled with @option{-std=c99} or @option{-std=c11}). The
9795 @code{restrict} is only available when @option{-std=gnu99} (which will
9796 eventually be the default) or @option{-std=c99} (or the equivalent
9797 @option{-std=iso9899:1999}), or an option for a later standard
9800 The way to solve these problems is to put @samp{__} at the beginning and
9801 end of each problematical keyword. For example, use @code{__asm__}
9802 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
9804 Other C compilers won't accept these alternative keywords; if you want to
9805 compile with another compiler, you can define the alternate keywords as
9806 macros to replace them with the customary keywords. It looks like this:
9814 @findex __extension__
9816 @option{-pedantic} and other options cause warnings for many GNU C extensions.
9818 prevent such warnings within one expression by writing
9819 @code{__extension__} before the expression. @code{__extension__} has no
9820 effect aside from this.
9822 @node Incomplete Enums
9823 @section Incomplete @code{enum} Types
9825 You can define an @code{enum} tag without specifying its possible values.
9826 This results in an incomplete type, much like what you get if you write
9827 @code{struct foo} without describing the elements. A later declaration
9828 that does specify the possible values completes the type.
9830 You cannot allocate variables or storage using the type while it is
9831 incomplete. However, you can work with pointers to that type.
9833 This extension may not be very useful, but it makes the handling of
9834 @code{enum} more consistent with the way @code{struct} and @code{union}
9837 This extension is not supported by GNU C++.
9839 @node Function Names
9840 @section Function Names as Strings
9841 @cindex @code{__func__} identifier
9842 @cindex @code{__FUNCTION__} identifier
9843 @cindex @code{__PRETTY_FUNCTION__} identifier
9845 GCC provides three magic constants that hold the name of the current
9846 function as a string. In C++11 and later modes, all three are treated
9847 as constant expressions and can be used in @code{constexpr} constexts.
9848 The first of these constants is @code{__func__}, which is part of
9851 The identifier @code{__func__} is implicitly declared by the translator
9852 as if, immediately following the opening brace of each function
9853 definition, the declaration
9856 static const char __func__[] = "function-name";
9860 appeared, where function-name is the name of the lexically-enclosing
9861 function. This name is the unadorned name of the function. As an
9862 extension, at file (or, in C++, namespace scope), @code{__func__}
9863 evaluates to the empty string.
9865 @code{__FUNCTION__} is another name for @code{__func__}, provided for
9866 backward compatibility with old versions of GCC.
9868 In C, @code{__PRETTY_FUNCTION__} is yet another name for
9869 @code{__func__}, except that at file (or, in C++, namespace scope),
9870 it evaluates to the string @code{"top level"}. In addition, in C++,
9871 @code{__PRETTY_FUNCTION__} contains the signature of the function as
9872 well as its bare name. For example, this program:
9875 extern "C" int printf (const char *, ...);
9881 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
9882 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
9900 __PRETTY_FUNCTION__ = void a::sub(int)
9903 These identifiers are variables, not preprocessor macros, and may not
9904 be used to initialize @code{char} arrays or be concatenated with string
9907 @node Return Address
9908 @section Getting the Return or Frame Address of a Function
9910 These functions may be used to get information about the callers of a
9913 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
9914 This function returns the return address of the current function, or of
9915 one of its callers. The @var{level} argument is number of frames to
9916 scan up the call stack. A value of @code{0} yields the return address
9917 of the current function, a value of @code{1} yields the return address
9918 of the caller of the current function, and so forth. When inlining
9919 the expected behavior is that the function returns the address of
9920 the function that is returned to. To work around this behavior use
9921 the @code{noinline} function attribute.
9923 The @var{level} argument must be a constant integer.
9925 On some machines it may be impossible to determine the return address of
9926 any function other than the current one; in such cases, or when the top
9927 of the stack has been reached, this function returns @code{0} or a
9928 random value. In addition, @code{__builtin_frame_address} may be used
9929 to determine if the top of the stack has been reached.
9931 Additional post-processing of the returned value may be needed, see
9932 @code{__builtin_extract_return_addr}.
9934 Calling this function with a nonzero argument can have unpredictable
9935 effects, including crashing the calling program. As a result, calls
9936 that are considered unsafe are diagnosed when the @option{-Wframe-address}
9937 option is in effect. Such calls should only be made in debugging
9941 @deftypefn {Built-in Function} {void *} __builtin_extract_return_addr (void *@var{addr})
9942 The address as returned by @code{__builtin_return_address} may have to be fed
9943 through this function to get the actual encoded address. For example, on the
9944 31-bit S/390 platform the highest bit has to be masked out, or on SPARC
9945 platforms an offset has to be added for the true next instruction to be
9948 If no fixup is needed, this function simply passes through @var{addr}.
9951 @deftypefn {Built-in Function} {void *} __builtin_frob_return_address (void *@var{addr})
9952 This function does the reverse of @code{__builtin_extract_return_addr}.
9955 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
9956 This function is similar to @code{__builtin_return_address}, but it
9957 returns the address of the function frame rather than the return address
9958 of the function. Calling @code{__builtin_frame_address} with a value of
9959 @code{0} yields the frame address of the current function, a value of
9960 @code{1} yields the frame address of the caller of the current function,
9963 The frame is the area on the stack that holds local variables and saved
9964 registers. The frame address is normally the address of the first word
9965 pushed on to the stack by the function. However, the exact definition
9966 depends upon the processor and the calling convention. If the processor
9967 has a dedicated frame pointer register, and the function has a frame,
9968 then @code{__builtin_frame_address} returns the value of the frame
9971 On some machines it may be impossible to determine the frame address of
9972 any function other than the current one; in such cases, or when the top
9973 of the stack has been reached, this function returns @code{0} if
9974 the first frame pointer is properly initialized by the startup code.
9976 Calling this function with a nonzero argument can have unpredictable
9977 effects, including crashing the calling program. As a result, calls
9978 that are considered unsafe are diagnosed when the @option{-Wframe-address}
9979 option is in effect. Such calls should only be made in debugging
9983 @node Vector Extensions
9984 @section Using Vector Instructions through Built-in Functions
9986 On some targets, the instruction set contains SIMD vector instructions which
9987 operate on multiple values contained in one large register at the same time.
9988 For example, on the x86 the MMX, 3DNow!@: and SSE extensions can be used
9991 The first step in using these extensions is to provide the necessary data
9992 types. This should be done using an appropriate @code{typedef}:
9995 typedef int v4si __attribute__ ((vector_size (16)));
9999 The @code{int} type specifies the base type, while the attribute specifies
10000 the vector size for the variable, measured in bytes. For example, the
10001 declaration above causes the compiler to set the mode for the @code{v4si}
10002 type to be 16 bytes wide and divided into @code{int} sized units. For
10003 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
10004 corresponding mode of @code{foo} is @acronym{V4SI}.
10006 The @code{vector_size} attribute is only applicable to integral and
10007 float scalars, although arrays, pointers, and function return values
10008 are allowed in conjunction with this construct. Only sizes that are
10009 a power of two are currently allowed.
10011 All the basic integer types can be used as base types, both as signed
10012 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
10013 @code{long long}. In addition, @code{float} and @code{double} can be
10014 used to build floating-point vector types.
10016 Specifying a combination that is not valid for the current architecture
10017 causes GCC to synthesize the instructions using a narrower mode.
10018 For example, if you specify a variable of type @code{V4SI} and your
10019 architecture does not allow for this specific SIMD type, GCC
10020 produces code that uses 4 @code{SIs}.
10022 The types defined in this manner can be used with a subset of normal C
10023 operations. Currently, GCC allows using the following operators
10024 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@.
10026 The operations behave like C++ @code{valarrays}. Addition is defined as
10027 the addition of the corresponding elements of the operands. For
10028 example, in the code below, each of the 4 elements in @var{a} is
10029 added to the corresponding 4 elements in @var{b} and the resulting
10030 vector is stored in @var{c}.
10033 typedef int v4si __attribute__ ((vector_size (16)));
10040 Subtraction, multiplication, division, and the logical operations
10041 operate in a similar manner. Likewise, the result of using the unary
10042 minus or complement operators on a vector type is a vector whose
10043 elements are the negative or complemented values of the corresponding
10044 elements in the operand.
10046 It is possible to use shifting operators @code{<<}, @code{>>} on
10047 integer-type vectors. The operation is defined as following: @code{@{a0,
10048 a1, @dots{}, an@} >> @{b0, b1, @dots{}, bn@} == @{a0 >> b0, a1 >> b1,
10049 @dots{}, an >> bn@}}@. Vector operands must have the same number of
10052 For convenience, it is allowed to use a binary vector operation
10053 where one operand is a scalar. In that case the compiler transforms
10054 the scalar operand into a vector where each element is the scalar from
10055 the operation. The transformation happens only if the scalar could be
10056 safely converted to the vector-element type.
10057 Consider the following code.
10060 typedef int v4si __attribute__ ((vector_size (16)));
10065 a = b + 1; /* a = b + @{1,1,1,1@}; */
10066 a = 2 * b; /* a = @{2,2,2,2@} * b; */
10068 a = l + a; /* Error, cannot convert long to int. */
10071 Vectors can be subscripted as if the vector were an array with
10072 the same number of elements and base type. Out of bound accesses
10073 invoke undefined behavior at run time. Warnings for out of bound
10074 accesses for vector subscription can be enabled with
10075 @option{-Warray-bounds}.
10077 Vector comparison is supported with standard comparison
10078 operators: @code{==, !=, <, <=, >, >=}. Comparison operands can be
10079 vector expressions of integer-type or real-type. Comparison between
10080 integer-type vectors and real-type vectors are not supported. The
10081 result of the comparison is a vector of the same width and number of
10082 elements as the comparison operands with a signed integral element
10085 Vectors are compared element-wise producing 0 when comparison is false
10086 and -1 (constant of the appropriate type where all bits are set)
10087 otherwise. Consider the following example.
10090 typedef int v4si __attribute__ ((vector_size (16)));
10092 v4si a = @{1,2,3,4@};
10093 v4si b = @{3,2,1,4@};
10096 c = a > b; /* The result would be @{0, 0,-1, 0@} */
10097 c = a == b; /* The result would be @{0,-1, 0,-1@} */
10100 In C++, the ternary operator @code{?:} is available. @code{a?b:c}, where
10101 @code{b} and @code{c} are vectors of the same type and @code{a} is an
10102 integer vector with the same number of elements of the same size as @code{b}
10103 and @code{c}, computes all three arguments and creates a vector
10104 @code{@{a[0]?b[0]:c[0], a[1]?b[1]:c[1], @dots{}@}}. Note that unlike in
10105 OpenCL, @code{a} is thus interpreted as @code{a != 0} and not @code{a < 0}.
10106 As in the case of binary operations, this syntax is also accepted when
10107 one of @code{b} or @code{c} is a scalar that is then transformed into a
10108 vector. If both @code{b} and @code{c} are scalars and the type of
10109 @code{true?b:c} has the same size as the element type of @code{a}, then
10110 @code{b} and @code{c} are converted to a vector type whose elements have
10111 this type and with the same number of elements as @code{a}.
10113 In C++, the logic operators @code{!, &&, ||} are available for vectors.
10114 @code{!v} is equivalent to @code{v == 0}, @code{a && b} is equivalent to
10115 @code{a!=0 & b!=0} and @code{a || b} is equivalent to @code{a!=0 | b!=0}.
10116 For mixed operations between a scalar @code{s} and a vector @code{v},
10117 @code{s && v} is equivalent to @code{s?v!=0:0} (the evaluation is
10118 short-circuit) and @code{v && s} is equivalent to @code{v!=0 & (s?-1:0)}.
10120 @findex __builtin_shuffle
10121 Vector shuffling is available using functions
10122 @code{__builtin_shuffle (vec, mask)} and
10123 @code{__builtin_shuffle (vec0, vec1, mask)}.
10124 Both functions construct a permutation of elements from one or two
10125 vectors and return a vector of the same type as the input vector(s).
10126 The @var{mask} is an integral vector with the same width (@var{W})
10127 and element count (@var{N}) as the output vector.
10129 The elements of the input vectors are numbered in memory ordering of
10130 @var{vec0} beginning at 0 and @var{vec1} beginning at @var{N}. The
10131 elements of @var{mask} are considered modulo @var{N} in the single-operand
10132 case and modulo @math{2*@var{N}} in the two-operand case.
10134 Consider the following example,
10137 typedef int v4si __attribute__ ((vector_size (16)));
10139 v4si a = @{1,2,3,4@};
10140 v4si b = @{5,6,7,8@};
10141 v4si mask1 = @{0,1,1,3@};
10142 v4si mask2 = @{0,4,2,5@};
10145 res = __builtin_shuffle (a, mask1); /* res is @{1,2,2,4@} */
10146 res = __builtin_shuffle (a, b, mask2); /* res is @{1,5,3,6@} */
10149 Note that @code{__builtin_shuffle} is intentionally semantically
10150 compatible with the OpenCL @code{shuffle} and @code{shuffle2} functions.
10152 You can declare variables and use them in function calls and returns, as
10153 well as in assignments and some casts. You can specify a vector type as
10154 a return type for a function. Vector types can also be used as function
10155 arguments. It is possible to cast from one vector type to another,
10156 provided they are of the same size (in fact, you can also cast vectors
10157 to and from other datatypes of the same size).
10159 You cannot operate between vectors of different lengths or different
10160 signedness without a cast.
10163 @section Support for @code{offsetof}
10164 @findex __builtin_offsetof
10166 GCC implements for both C and C++ a syntactic extension to implement
10167 the @code{offsetof} macro.
10171 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
10173 offsetof_member_designator:
10175 | offsetof_member_designator "." @code{identifier}
10176 | offsetof_member_designator "[" @code{expr} "]"
10179 This extension is sufficient such that
10182 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
10186 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
10187 may be dependent. In either case, @var{member} may consist of a single
10188 identifier, or a sequence of member accesses and array references.
10190 @node __sync Builtins
10191 @section Legacy @code{__sync} Built-in Functions for Atomic Memory Access
10193 The following built-in functions
10194 are intended to be compatible with those described
10195 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
10196 section 7.4. As such, they depart from normal GCC practice by not using
10197 the @samp{__builtin_} prefix and also by being overloaded so that they
10198 work on multiple types.
10200 The definition given in the Intel documentation allows only for the use of
10201 the types @code{int}, @code{long}, @code{long long} or their unsigned
10202 counterparts. GCC allows any scalar type that is 1, 2, 4 or 8 bytes in
10203 size other than the C type @code{_Bool} or the C++ type @code{bool}.
10204 Operations on pointer arguments are performed as if the operands were
10205 of the @code{uintptr_t} type. That is, they are not scaled by the size
10206 of the type to which the pointer points.
10208 These functions are implemented in terms of the @samp{__atomic}
10209 builtins (@pxref{__atomic Builtins}). They should not be used for new
10210 code which should use the @samp{__atomic} builtins instead.
10212 Not all operations are supported by all target processors. If a particular
10213 operation cannot be implemented on the target processor, a warning is
10214 generated and a call to an external function is generated. The external
10215 function carries the same name as the built-in version,
10216 with an additional suffix
10217 @samp{_@var{n}} where @var{n} is the size of the data type.
10219 @c ??? Should we have a mechanism to suppress this warning? This is almost
10220 @c useful for implementing the operation under the control of an external
10223 In most cases, these built-in functions are considered a @dfn{full barrier}.
10225 no memory operand is moved across the operation, either forward or
10226 backward. Further, instructions are issued as necessary to prevent the
10227 processor from speculating loads across the operation and from queuing stores
10228 after the operation.
10230 All of the routines are described in the Intel documentation to take
10231 ``an optional list of variables protected by the memory barrier''. It's
10232 not clear what is meant by that; it could mean that @emph{only} the
10233 listed variables are protected, or it could mean a list of additional
10234 variables to be protected. The list is ignored by GCC which treats it as
10235 empty. GCC interprets an empty list as meaning that all globally
10236 accessible variables should be protected.
10239 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
10240 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
10241 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
10242 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
10243 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
10244 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
10245 @findex __sync_fetch_and_add
10246 @findex __sync_fetch_and_sub
10247 @findex __sync_fetch_and_or
10248 @findex __sync_fetch_and_and
10249 @findex __sync_fetch_and_xor
10250 @findex __sync_fetch_and_nand
10251 These built-in functions perform the operation suggested by the name, and
10252 returns the value that had previously been in memory. That is, operations
10253 on integer operands have the following semantics. Operations on pointer
10254 arguments are performed as if the operands were of the @code{uintptr_t}
10255 type. That is, they are not scaled by the size of the type to which
10256 the pointer points.
10259 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
10260 @{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand
10263 The object pointed to by the first argument must be of integer or pointer
10264 type. It must not be a boolean type.
10266 @emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand}
10267 as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}.
10269 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
10270 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
10271 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
10272 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
10273 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
10274 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
10275 @findex __sync_add_and_fetch
10276 @findex __sync_sub_and_fetch
10277 @findex __sync_or_and_fetch
10278 @findex __sync_and_and_fetch
10279 @findex __sync_xor_and_fetch
10280 @findex __sync_nand_and_fetch
10281 These built-in functions perform the operation suggested by the name, and
10282 return the new value. That is, operations on integer operands have
10283 the following semantics. Operations on pointer operands are performed as
10284 if the operand's type were @code{uintptr_t}.
10287 @{ *ptr @var{op}= value; return *ptr; @}
10288 @{ *ptr = ~(*ptr & value); return *ptr; @} // nand
10291 The same constraints on arguments apply as for the corresponding
10292 @code{__sync_op_and_fetch} built-in functions.
10294 @emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch}
10295 as @code{*ptr = ~(*ptr & value)} instead of
10296 @code{*ptr = ~*ptr & value}.
10298 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
10299 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
10300 @findex __sync_bool_compare_and_swap
10301 @findex __sync_val_compare_and_swap
10302 These built-in functions perform an atomic compare and swap.
10303 That is, if the current
10304 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
10307 The ``bool'' version returns true if the comparison is successful and
10308 @var{newval} is written. The ``val'' version returns the contents
10309 of @code{*@var{ptr}} before the operation.
10311 @item __sync_synchronize (...)
10312 @findex __sync_synchronize
10313 This built-in function issues a full memory barrier.
10315 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
10316 @findex __sync_lock_test_and_set
10317 This built-in function, as described by Intel, is not a traditional test-and-set
10318 operation, but rather an atomic exchange operation. It writes @var{value}
10319 into @code{*@var{ptr}}, and returns the previous contents of
10322 Many targets have only minimal support for such locks, and do not support
10323 a full exchange operation. In this case, a target may support reduced
10324 functionality here by which the @emph{only} valid value to store is the
10325 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
10326 is implementation defined.
10328 This built-in function is not a full barrier,
10329 but rather an @dfn{acquire barrier}.
10330 This means that references after the operation cannot move to (or be
10331 speculated to) before the operation, but previous memory stores may not
10332 be globally visible yet, and previous memory loads may not yet be
10335 @item void __sync_lock_release (@var{type} *ptr, ...)
10336 @findex __sync_lock_release
10337 This built-in function releases the lock acquired by
10338 @code{__sync_lock_test_and_set}.
10339 Normally this means writing the constant 0 to @code{*@var{ptr}}.
10341 This built-in function is not a full barrier,
10342 but rather a @dfn{release barrier}.
10343 This means that all previous memory stores are globally visible, and all
10344 previous memory loads have been satisfied, but following memory reads
10345 are not prevented from being speculated to before the barrier.
10348 @node __atomic Builtins
10349 @section Built-in Functions for Memory Model Aware Atomic Operations
10351 The following built-in functions approximately match the requirements
10352 for the C++11 memory model. They are all
10353 identified by being prefixed with @samp{__atomic} and most are
10354 overloaded so that they work with multiple types.
10356 These functions are intended to replace the legacy @samp{__sync}
10357 builtins. The main difference is that the memory order that is requested
10358 is a parameter to the functions. New code should always use the
10359 @samp{__atomic} builtins rather than the @samp{__sync} builtins.
10361 Note that the @samp{__atomic} builtins assume that programs will
10362 conform to the C++11 memory model. In particular, they assume
10363 that programs are free of data races. See the C++11 standard for
10364 detailed requirements.
10366 The @samp{__atomic} builtins can be used with any integral scalar or
10367 pointer type that is 1, 2, 4, or 8 bytes in length. 16-byte integral
10368 types are also allowed if @samp{__int128} (@pxref{__int128}) is
10369 supported by the architecture.
10371 The four non-arithmetic functions (load, store, exchange, and
10372 compare_exchange) all have a generic version as well. This generic
10373 version works on any data type. It uses the lock-free built-in function
10374 if the specific data type size makes that possible; otherwise, an
10375 external call is left to be resolved at run time. This external call is
10376 the same format with the addition of a @samp{size_t} parameter inserted
10377 as the first parameter indicating the size of the object being pointed to.
10378 All objects must be the same size.
10380 There are 6 different memory orders that can be specified. These map
10381 to the C++11 memory orders with the same names, see the C++11 standard
10382 or the @uref{http://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync,GCC wiki
10383 on atomic synchronization} for detailed definitions. Individual
10384 targets may also support additional memory orders for use on specific
10385 architectures. Refer to the target documentation for details of
10388 An atomic operation can both constrain code motion and
10389 be mapped to hardware instructions for synchronization between threads
10390 (e.g., a fence). To which extent this happens is controlled by the
10391 memory orders, which are listed here in approximately ascending order of
10392 strength. The description of each memory order is only meant to roughly
10393 illustrate the effects and is not a specification; see the C++11
10394 memory model for precise semantics.
10397 @item __ATOMIC_RELAXED
10398 Implies no inter-thread ordering constraints.
10399 @item __ATOMIC_CONSUME
10400 This is currently implemented using the stronger @code{__ATOMIC_ACQUIRE}
10401 memory order because of a deficiency in C++11's semantics for
10402 @code{memory_order_consume}.
10403 @item __ATOMIC_ACQUIRE
10404 Creates an inter-thread happens-before constraint from the release (or
10405 stronger) semantic store to this acquire load. Can prevent hoisting
10406 of code to before the operation.
10407 @item __ATOMIC_RELEASE
10408 Creates an inter-thread happens-before constraint to acquire (or stronger)
10409 semantic loads that read from this release store. Can prevent sinking
10410 of code to after the operation.
10411 @item __ATOMIC_ACQ_REL
10412 Combines the effects of both @code{__ATOMIC_ACQUIRE} and
10413 @code{__ATOMIC_RELEASE}.
10414 @item __ATOMIC_SEQ_CST
10415 Enforces total ordering with all other @code{__ATOMIC_SEQ_CST} operations.
10418 Note that in the C++11 memory model, @emph{fences} (e.g.,
10419 @samp{__atomic_thread_fence}) take effect in combination with other
10420 atomic operations on specific memory locations (e.g., atomic loads);
10421 operations on specific memory locations do not necessarily affect other
10422 operations in the same way.
10424 Target architectures are encouraged to provide their own patterns for
10425 each of the atomic built-in functions. If no target is provided, the original
10426 non-memory model set of @samp{__sync} atomic built-in functions are
10427 used, along with any required synchronization fences surrounding it in
10428 order to achieve the proper behavior. Execution in this case is subject
10429 to the same restrictions as those built-in functions.
10431 If there is no pattern or mechanism to provide a lock-free instruction
10432 sequence, a call is made to an external routine with the same parameters
10433 to be resolved at run time.
10435 When implementing patterns for these built-in functions, the memory order
10436 parameter can be ignored as long as the pattern implements the most
10437 restrictive @code{__ATOMIC_SEQ_CST} memory order. Any of the other memory
10438 orders execute correctly with this memory order but they may not execute as
10439 efficiently as they could with a more appropriate implementation of the
10440 relaxed requirements.
10442 Note that the C++11 standard allows for the memory order parameter to be
10443 determined at run time rather than at compile time. These built-in
10444 functions map any run-time value to @code{__ATOMIC_SEQ_CST} rather
10445 than invoke a runtime library call or inline a switch statement. This is
10446 standard compliant, safe, and the simplest approach for now.
10448 The memory order parameter is a signed int, but only the lower 16 bits are
10449 reserved for the memory order. The remainder of the signed int is reserved
10450 for target use and should be 0. Use of the predefined atomic values
10451 ensures proper usage.
10453 @deftypefn {Built-in Function} @var{type} __atomic_load_n (@var{type} *ptr, int memorder)
10454 This built-in function implements an atomic load operation. It returns the
10455 contents of @code{*@var{ptr}}.
10457 The valid memory order variants are
10458 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
10459 and @code{__ATOMIC_CONSUME}.
10463 @deftypefn {Built-in Function} void __atomic_load (@var{type} *ptr, @var{type} *ret, int memorder)
10464 This is the generic version of an atomic load. It returns the
10465 contents of @code{*@var{ptr}} in @code{*@var{ret}}.
10469 @deftypefn {Built-in Function} void __atomic_store_n (@var{type} *ptr, @var{type} val, int memorder)
10470 This built-in function implements an atomic store operation. It writes
10471 @code{@var{val}} into @code{*@var{ptr}}.
10473 The valid memory order variants are
10474 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and @code{__ATOMIC_RELEASE}.
10478 @deftypefn {Built-in Function} void __atomic_store (@var{type} *ptr, @var{type} *val, int memorder)
10479 This is the generic version of an atomic store. It stores the value
10480 of @code{*@var{val}} into @code{*@var{ptr}}.
10484 @deftypefn {Built-in Function} @var{type} __atomic_exchange_n (@var{type} *ptr, @var{type} val, int memorder)
10485 This built-in function implements an atomic exchange operation. It writes
10486 @var{val} into @code{*@var{ptr}}, and returns the previous contents of
10489 The valid memory order variants are
10490 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
10491 @code{__ATOMIC_RELEASE}, and @code{__ATOMIC_ACQ_REL}.
10495 @deftypefn {Built-in Function} void __atomic_exchange (@var{type} *ptr, @var{type} *val, @var{type} *ret, int memorder)
10496 This is the generic version of an atomic exchange. It stores the
10497 contents of @code{*@var{val}} into @code{*@var{ptr}}. The original value
10498 of @code{*@var{ptr}} is copied into @code{*@var{ret}}.
10502 @deftypefn {Built-in Function} bool __atomic_compare_exchange_n (@var{type} *ptr, @var{type} *expected, @var{type} desired, bool weak, int success_memorder, int failure_memorder)
10503 This built-in function implements an atomic compare and exchange operation.
10504 This compares the contents of @code{*@var{ptr}} with the contents of
10505 @code{*@var{expected}}. If equal, the operation is a @emph{read-modify-write}
10506 operation that writes @var{desired} into @code{*@var{ptr}}. If they are not
10507 equal, the operation is a @emph{read} and the current contents of
10508 @code{*@var{ptr}} are written into @code{*@var{expected}}. @var{weak} is true
10509 for weak compare_exchange, which may fail spuriously, and false for
10510 the strong variation, which never fails spuriously. Many targets
10511 only offer the strong variation and ignore the parameter. When in doubt, use
10512 the strong variation.
10514 If @var{desired} is written into @code{*@var{ptr}} then true is returned
10515 and memory is affected according to the
10516 memory order specified by @var{success_memorder}. There are no
10517 restrictions on what memory order can be used here.
10519 Otherwise, false is returned and memory is affected according
10520 to @var{failure_memorder}. This memory order cannot be
10521 @code{__ATOMIC_RELEASE} nor @code{__ATOMIC_ACQ_REL}. It also cannot be a
10522 stronger order than that specified by @var{success_memorder}.
10526 @deftypefn {Built-in Function} bool __atomic_compare_exchange (@var{type} *ptr, @var{type} *expected, @var{type} *desired, bool weak, int success_memorder, int failure_memorder)
10527 This built-in function implements the generic version of
10528 @code{__atomic_compare_exchange}. The function is virtually identical to
10529 @code{__atomic_compare_exchange_n}, except the desired value is also a
10534 @deftypefn {Built-in Function} @var{type} __atomic_add_fetch (@var{type} *ptr, @var{type} val, int memorder)
10535 @deftypefnx {Built-in Function} @var{type} __atomic_sub_fetch (@var{type} *ptr, @var{type} val, int memorder)
10536 @deftypefnx {Built-in Function} @var{type} __atomic_and_fetch (@var{type} *ptr, @var{type} val, int memorder)
10537 @deftypefnx {Built-in Function} @var{type} __atomic_xor_fetch (@var{type} *ptr, @var{type} val, int memorder)
10538 @deftypefnx {Built-in Function} @var{type} __atomic_or_fetch (@var{type} *ptr, @var{type} val, int memorder)
10539 @deftypefnx {Built-in Function} @var{type} __atomic_nand_fetch (@var{type} *ptr, @var{type} val, int memorder)
10540 These built-in functions perform the operation suggested by the name, and
10541 return the result of the operation. Operations on pointer arguments are
10542 performed as if the operands were of the @code{uintptr_t} type. That is,
10543 they are not scaled by the size of the type to which the pointer points.
10546 @{ *ptr @var{op}= val; return *ptr; @}
10549 The object pointed to by the first argument must be of integer or pointer
10550 type. It must not be a boolean type. All memory orders are valid.
10554 @deftypefn {Built-in Function} @var{type} __atomic_fetch_add (@var{type} *ptr, @var{type} val, int memorder)
10555 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_sub (@var{type} *ptr, @var{type} val, int memorder)
10556 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_and (@var{type} *ptr, @var{type} val, int memorder)
10557 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_xor (@var{type} *ptr, @var{type} val, int memorder)
10558 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_or (@var{type} *ptr, @var{type} val, int memorder)
10559 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_nand (@var{type} *ptr, @var{type} val, int memorder)
10560 These built-in functions perform the operation suggested by the name, and
10561 return the value that had previously been in @code{*@var{ptr}}. Operations
10562 on pointer arguments are performed as if the operands were of
10563 the @code{uintptr_t} type. That is, they are not scaled by the size of
10564 the type to which the pointer points.
10567 @{ tmp = *ptr; *ptr @var{op}= val; return tmp; @}
10570 The same constraints on arguments apply as for the corresponding
10571 @code{__atomic_op_fetch} built-in functions. All memory orders are valid.
10575 @deftypefn {Built-in Function} bool __atomic_test_and_set (void *ptr, int memorder)
10577 This built-in function performs an atomic test-and-set operation on
10578 the byte at @code{*@var{ptr}}. The byte is set to some implementation
10579 defined nonzero ``set'' value and the return value is @code{true} if and only
10580 if the previous contents were ``set''.
10581 It should be only used for operands of type @code{bool} or @code{char}. For
10582 other types only part of the value may be set.
10584 All memory orders are valid.
10588 @deftypefn {Built-in Function} void __atomic_clear (bool *ptr, int memorder)
10590 This built-in function performs an atomic clear operation on
10591 @code{*@var{ptr}}. After the operation, @code{*@var{ptr}} contains 0.
10592 It should be only used for operands of type @code{bool} or @code{char} and
10593 in conjunction with @code{__atomic_test_and_set}.
10594 For other types it may only clear partially. If the type is not @code{bool}
10595 prefer using @code{__atomic_store}.
10597 The valid memory order variants are
10598 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and
10599 @code{__ATOMIC_RELEASE}.
10603 @deftypefn {Built-in Function} void __atomic_thread_fence (int memorder)
10605 This built-in function acts as a synchronization fence between threads
10606 based on the specified memory order.
10608 All memory orders are valid.
10612 @deftypefn {Built-in Function} void __atomic_signal_fence (int memorder)
10614 This built-in function acts as a synchronization fence between a thread
10615 and signal handlers based in the same thread.
10617 All memory orders are valid.
10621 @deftypefn {Built-in Function} bool __atomic_always_lock_free (size_t size, void *ptr)
10623 This built-in function returns true if objects of @var{size} bytes always
10624 generate lock-free atomic instructions for the target architecture.
10625 @var{size} must resolve to a compile-time constant and the result also
10626 resolves to a compile-time constant.
10628 @var{ptr} is an optional pointer to the object that may be used to determine
10629 alignment. A value of 0 indicates typical alignment should be used. The
10630 compiler may also ignore this parameter.
10633 if (__atomic_always_lock_free (sizeof (long long), 0))
10638 @deftypefn {Built-in Function} bool __atomic_is_lock_free (size_t size, void *ptr)
10640 This built-in function returns true if objects of @var{size} bytes always
10641 generate lock-free atomic instructions for the target architecture. If
10642 the built-in function is not known to be lock-free, a call is made to a
10643 runtime routine named @code{__atomic_is_lock_free}.
10645 @var{ptr} is an optional pointer to the object that may be used to determine
10646 alignment. A value of 0 indicates typical alignment should be used. The
10647 compiler may also ignore this parameter.
10650 @node Integer Overflow Builtins
10651 @section Built-in Functions to Perform Arithmetic with Overflow Checking
10653 The following built-in functions allow performing simple arithmetic operations
10654 together with checking whether the operations overflowed.
10656 @deftypefn {Built-in Function} bool __builtin_add_overflow (@var{type1} a, @var{type2} b, @var{type3} *res)
10657 @deftypefnx {Built-in Function} bool __builtin_sadd_overflow (int a, int b, int *res)
10658 @deftypefnx {Built-in Function} bool __builtin_saddl_overflow (long int a, long int b, long int *res)
10659 @deftypefnx {Built-in Function} bool __builtin_saddll_overflow (long long int a, long long int b, long long int *res)
10660 @deftypefnx {Built-in Function} bool __builtin_uadd_overflow (unsigned int a, unsigned int b, unsigned int *res)
10661 @deftypefnx {Built-in Function} bool __builtin_uaddl_overflow (unsigned long int a, unsigned long int b, unsigned long int *res)
10662 @deftypefnx {Built-in Function} bool __builtin_uaddll_overflow (unsigned long long int a, unsigned long long int b, unsigned long long int *res)
10664 These built-in functions promote the first two operands into infinite precision signed
10665 type and perform addition on those promoted operands. The result is then
10666 cast to the type the third pointer argument points to and stored there.
10667 If the stored result is equal to the infinite precision result, the built-in
10668 functions return false, otherwise they return true. As the addition is
10669 performed in infinite signed precision, these built-in functions have fully defined
10670 behavior for all argument values.
10672 The first built-in function allows arbitrary integral types for operands and
10673 the result type must be pointer to some integral type other than enumerated or
10674 boolean type, the rest of the built-in functions have explicit integer types.
10676 The compiler will attempt to use hardware instructions to implement
10677 these built-in functions where possible, like conditional jump on overflow
10678 after addition, conditional jump on carry etc.
10682 @deftypefn {Built-in Function} bool __builtin_sub_overflow (@var{type1} a, @var{type2} b, @var{type3} *res)
10683 @deftypefnx {Built-in Function} bool __builtin_ssub_overflow (int a, int b, int *res)
10684 @deftypefnx {Built-in Function} bool __builtin_ssubl_overflow (long int a, long int b, long int *res)
10685 @deftypefnx {Built-in Function} bool __builtin_ssubll_overflow (long long int a, long long int b, long long int *res)
10686 @deftypefnx {Built-in Function} bool __builtin_usub_overflow (unsigned int a, unsigned int b, unsigned int *res)
10687 @deftypefnx {Built-in Function} bool __builtin_usubl_overflow (unsigned long int a, unsigned long int b, unsigned long int *res)
10688 @deftypefnx {Built-in Function} bool __builtin_usubll_overflow (unsigned long long int a, unsigned long long int b, unsigned long long int *res)
10690 These built-in functions are similar to the add overflow checking built-in
10691 functions above, except they perform subtraction, subtract the second argument
10692 from the first one, instead of addition.
10696 @deftypefn {Built-in Function} bool __builtin_mul_overflow (@var{type1} a, @var{type2} b, @var{type3} *res)
10697 @deftypefnx {Built-in Function} bool __builtin_smul_overflow (int a, int b, int *res)
10698 @deftypefnx {Built-in Function} bool __builtin_smull_overflow (long int a, long int b, long int *res)
10699 @deftypefnx {Built-in Function} bool __builtin_smulll_overflow (long long int a, long long int b, long long int *res)
10700 @deftypefnx {Built-in Function} bool __builtin_umul_overflow (unsigned int a, unsigned int b, unsigned int *res)
10701 @deftypefnx {Built-in Function} bool __builtin_umull_overflow (unsigned long int a, unsigned long int b, unsigned long int *res)
10702 @deftypefnx {Built-in Function} bool __builtin_umulll_overflow (unsigned long long int a, unsigned long long int b, unsigned long long int *res)
10704 These built-in functions are similar to the add overflow checking built-in
10705 functions above, except they perform multiplication, instead of addition.
10709 The following built-in functions allow checking if simple arithmetic operation
10712 @deftypefn {Built-in Function} bool __builtin_add_overflow_p (@var{type1} a, @var{type2} b, @var{type3} c)
10713 @deftypefnx {Built-in Function} bool __builtin_sub_overflow_p (@var{type1} a, @var{type2} b, @var{type3} c)
10714 @deftypefnx {Built-in Function} bool __builtin_mul_overflow_p (@var{type1} a, @var{type2} b, @var{type3} c)
10716 These built-in functions are similar to @code{__builtin_add_overflow},
10717 @code{__builtin_sub_overflow}, or @code{__builtin_mul_overflow}, except that
10718 they don't store the result of the arithmetic operation anywhere and the
10719 last argument is not a pointer, but some expression with integral type other
10720 than enumerated or boolean type.
10722 The built-in functions promote the first two operands into infinite precision signed type
10723 and perform addition on those promoted operands. The result is then
10724 cast to the type of the third argument. If the cast result is equal to the infinite
10725 precision result, the built-in functions return false, otherwise they return true.
10726 The value of the third argument is ignored, just the side effects in the third argument
10727 are evaluated, and no integral argument promotions are performed on the last argument.
10728 If the third argument is a bit-field, the type used for the result cast has the
10729 precision and signedness of the given bit-field, rather than precision and signedness
10730 of the underlying type.
10732 For example, the following macro can be used to portably check, at
10733 compile-time, whether or not adding two constant integers will overflow,
10734 and perform the addition only when it is known to be safe and not to trigger
10735 a @option{-Woverflow} warning.
10738 #define INT_ADD_OVERFLOW_P(a, b) \
10739 __builtin_add_overflow_p (a, b, (__typeof__ ((a) + (b))) 0)
10742 A = INT_MAX, B = 3,
10743 C = INT_ADD_OVERFLOW_P (A, B) ? 0 : A + B,
10744 D = __builtin_add_overflow_p (1, SCHAR_MAX, (signed char) 0)
10748 The compiler will attempt to use hardware instructions to implement
10749 these built-in functions where possible, like conditional jump on overflow
10750 after addition, conditional jump on carry etc.
10754 @node x86 specific memory model extensions for transactional memory
10755 @section x86-Specific Memory Model Extensions for Transactional Memory
10757 The x86 architecture supports additional memory ordering flags
10758 to mark critical sections for hardware lock elision.
10759 These must be specified in addition to an existing memory order to
10763 @item __ATOMIC_HLE_ACQUIRE
10764 Start lock elision on a lock variable.
10765 Memory order must be @code{__ATOMIC_ACQUIRE} or stronger.
10766 @item __ATOMIC_HLE_RELEASE
10767 End lock elision on a lock variable.
10768 Memory order must be @code{__ATOMIC_RELEASE} or stronger.
10771 When a lock acquire fails, it is required for good performance to abort
10772 the transaction quickly. This can be done with a @code{_mm_pause}.
10775 #include <immintrin.h> // For _mm_pause
10779 /* Acquire lock with lock elision */
10780 while (__atomic_exchange_n(&lockvar, 1, __ATOMIC_ACQUIRE|__ATOMIC_HLE_ACQUIRE))
10781 _mm_pause(); /* Abort failed transaction */
10783 /* Free lock with lock elision */
10784 __atomic_store_n(&lockvar, 0, __ATOMIC_RELEASE|__ATOMIC_HLE_RELEASE);
10787 @node Object Size Checking
10788 @section Object Size Checking Built-in Functions
10789 @findex __builtin_object_size
10790 @findex __builtin___memcpy_chk
10791 @findex __builtin___mempcpy_chk
10792 @findex __builtin___memmove_chk
10793 @findex __builtin___memset_chk
10794 @findex __builtin___strcpy_chk
10795 @findex __builtin___stpcpy_chk
10796 @findex __builtin___strncpy_chk
10797 @findex __builtin___strcat_chk
10798 @findex __builtin___strncat_chk
10799 @findex __builtin___sprintf_chk
10800 @findex __builtin___snprintf_chk
10801 @findex __builtin___vsprintf_chk
10802 @findex __builtin___vsnprintf_chk
10803 @findex __builtin___printf_chk
10804 @findex __builtin___vprintf_chk
10805 @findex __builtin___fprintf_chk
10806 @findex __builtin___vfprintf_chk
10808 GCC implements a limited buffer overflow protection mechanism that can
10809 prevent some buffer overflow attacks by determining the sizes of objects
10810 into which data is about to be written and preventing the writes when
10811 the size isn't sufficient. The built-in functions described below yield
10812 the best results when used together and when optimization is enabled.
10813 For example, to detect object sizes across function boundaries or to
10814 follow pointer assignments through non-trivial control flow they rely
10815 on various optimization passes enabled with @option{-O2}. However, to
10816 a limited extent, they can be used without optimization as well.
10818 @deftypefn {Built-in Function} {size_t} __builtin_object_size (const void * @var{ptr}, int @var{type})
10819 is a built-in construct that returns a constant number of bytes from
10820 @var{ptr} to the end of the object @var{ptr} pointer points to
10821 (if known at compile time). @code{__builtin_object_size} never evaluates
10822 its arguments for side effects. If there are any side effects in them, it
10823 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
10824 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
10825 point to and all of them are known at compile time, the returned number
10826 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
10827 0 and minimum if nonzero. If it is not possible to determine which objects
10828 @var{ptr} points to at compile time, @code{__builtin_object_size} should
10829 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
10830 for @var{type} 2 or 3.
10832 @var{type} is an integer constant from 0 to 3. If the least significant
10833 bit is clear, objects are whole variables, if it is set, a closest
10834 surrounding subobject is considered the object a pointer points to.
10835 The second bit determines if maximum or minimum of remaining bytes
10839 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
10840 char *p = &var.buf1[1], *q = &var.b;
10842 /* Here the object p points to is var. */
10843 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
10844 /* The subobject p points to is var.buf1. */
10845 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
10846 /* The object q points to is var. */
10847 assert (__builtin_object_size (q, 0)
10848 == (char *) (&var + 1) - (char *) &var.b);
10849 /* The subobject q points to is var.b. */
10850 assert (__builtin_object_size (q, 1) == sizeof (var.b));
10854 There are built-in functions added for many common string operation
10855 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
10856 built-in is provided. This built-in has an additional last argument,
10857 which is the number of bytes remaining in the object the @var{dest}
10858 argument points to or @code{(size_t) -1} if the size is not known.
10860 The built-in functions are optimized into the normal string functions
10861 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
10862 it is known at compile time that the destination object will not
10863 be overflowed. If the compiler can determine at compile time that the
10864 object will always be overflowed, it issues a warning.
10866 The intended use can be e.g.@:
10870 #define bos0(dest) __builtin_object_size (dest, 0)
10871 #define memcpy(dest, src, n) \
10872 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
10876 /* It is unknown what object p points to, so this is optimized
10877 into plain memcpy - no checking is possible. */
10878 memcpy (p, "abcde", n);
10879 /* Destination is known and length too. It is known at compile
10880 time there will be no overflow. */
10881 memcpy (&buf[5], "abcde", 5);
10882 /* Destination is known, but the length is not known at compile time.
10883 This will result in __memcpy_chk call that can check for overflow
10885 memcpy (&buf[5], "abcde", n);
10886 /* Destination is known and it is known at compile time there will
10887 be overflow. There will be a warning and __memcpy_chk call that
10888 will abort the program at run time. */
10889 memcpy (&buf[6], "abcde", 5);
10892 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
10893 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
10894 @code{strcat} and @code{strncat}.
10896 There are also checking built-in functions for formatted output functions.
10898 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
10899 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
10900 const char *fmt, ...);
10901 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
10903 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
10904 const char *fmt, va_list ap);
10907 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
10908 etc.@: functions and can contain implementation specific flags on what
10909 additional security measures the checking function might take, such as
10910 handling @code{%n} differently.
10912 The @var{os} argument is the object size @var{s} points to, like in the
10913 other built-in functions. There is a small difference in the behavior
10914 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
10915 optimized into the non-checking functions only if @var{flag} is 0, otherwise
10916 the checking function is called with @var{os} argument set to
10917 @code{(size_t) -1}.
10919 In addition to this, there are checking built-in functions
10920 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
10921 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
10922 These have just one additional argument, @var{flag}, right before
10923 format string @var{fmt}. If the compiler is able to optimize them to
10924 @code{fputc} etc.@: functions, it does, otherwise the checking function
10925 is called and the @var{flag} argument passed to it.
10927 @node Pointer Bounds Checker builtins
10928 @section Pointer Bounds Checker Built-in Functions
10929 @cindex Pointer Bounds Checker builtins
10930 @findex __builtin___bnd_set_ptr_bounds
10931 @findex __builtin___bnd_narrow_ptr_bounds
10932 @findex __builtin___bnd_copy_ptr_bounds
10933 @findex __builtin___bnd_init_ptr_bounds
10934 @findex __builtin___bnd_null_ptr_bounds
10935 @findex __builtin___bnd_store_ptr_bounds
10936 @findex __builtin___bnd_chk_ptr_lbounds
10937 @findex __builtin___bnd_chk_ptr_ubounds
10938 @findex __builtin___bnd_chk_ptr_bounds
10939 @findex __builtin___bnd_get_ptr_lbound
10940 @findex __builtin___bnd_get_ptr_ubound
10942 GCC provides a set of built-in functions to control Pointer Bounds Checker
10943 instrumentation. Note that all Pointer Bounds Checker builtins can be used
10944 even if you compile with Pointer Bounds Checker off
10945 (@option{-fno-check-pointer-bounds}).
10946 The behavior may differ in such case as documented below.
10948 @deftypefn {Built-in Function} {void *} __builtin___bnd_set_ptr_bounds (const void *@var{q}, size_t @var{size})
10950 This built-in function returns a new pointer with the value of @var{q}, and
10951 associate it with the bounds [@var{q}, @var{q}+@var{size}-1]. With Pointer
10952 Bounds Checker off, the built-in function just returns the first argument.
10955 extern void *__wrap_malloc (size_t n)
10957 void *p = (void *)__real_malloc (n);
10958 if (!p) return __builtin___bnd_null_ptr_bounds (p);
10959 return __builtin___bnd_set_ptr_bounds (p, n);
10965 @deftypefn {Built-in Function} {void *} __builtin___bnd_narrow_ptr_bounds (const void *@var{p}, const void *@var{q}, size_t @var{size})
10967 This built-in function returns a new pointer with the value of @var{p}
10968 and associates it with the narrowed bounds formed by the intersection
10969 of bounds associated with @var{q} and the bounds
10970 [@var{p}, @var{p} + @var{size} - 1].
10971 With Pointer Bounds Checker off, the built-in function just returns the first
10975 void init_objects (object *objs, size_t size)
10978 /* Initialize objects one-by-one passing pointers with bounds of
10979 an object, not the full array of objects. */
10980 for (i = 0; i < size; i++)
10981 init_object (__builtin___bnd_narrow_ptr_bounds (objs + i, objs,
10988 @deftypefn {Built-in Function} {void *} __builtin___bnd_copy_ptr_bounds (const void *@var{q}, const void *@var{r})
10990 This built-in function returns a new pointer with the value of @var{q},
10991 and associates it with the bounds already associated with pointer @var{r}.
10992 With Pointer Bounds Checker off, the built-in function just returns the first
10996 /* Here is a way to get pointer to object's field but
10997 still with the full object's bounds. */
10998 int *field_ptr = __builtin___bnd_copy_ptr_bounds (&objptr->int_field,
11004 @deftypefn {Built-in Function} {void *} __builtin___bnd_init_ptr_bounds (const void *@var{q})
11006 This built-in function returns a new pointer with the value of @var{q}, and
11007 associates it with INIT (allowing full memory access) bounds. With Pointer
11008 Bounds Checker off, the built-in function just returns the first argument.
11012 @deftypefn {Built-in Function} {void *} __builtin___bnd_null_ptr_bounds (const void *@var{q})
11014 This built-in function returns a new pointer with the value of @var{q}, and
11015 associates it with NULL (allowing no memory access) bounds. With Pointer
11016 Bounds Checker off, the built-in function just returns the first argument.
11020 @deftypefn {Built-in Function} void __builtin___bnd_store_ptr_bounds (const void **@var{ptr_addr}, const void *@var{ptr_val})
11022 This built-in function stores the bounds associated with pointer @var{ptr_val}
11023 and location @var{ptr_addr} into Bounds Table. This can be useful to propagate
11024 bounds from legacy code without touching the associated pointer's memory when
11025 pointers are copied as integers. With Pointer Bounds Checker off, the built-in
11026 function call is ignored.
11030 @deftypefn {Built-in Function} void __builtin___bnd_chk_ptr_lbounds (const void *@var{q})
11032 This built-in function checks if the pointer @var{q} is within the lower
11033 bound of its associated bounds. With Pointer Bounds Checker off, the built-in
11034 function call is ignored.
11037 extern void *__wrap_memset (void *dst, int c, size_t len)
11041 __builtin___bnd_chk_ptr_lbounds (dst);
11042 __builtin___bnd_chk_ptr_ubounds ((char *)dst + len - 1);
11043 __real_memset (dst, c, len);
11051 @deftypefn {Built-in Function} void __builtin___bnd_chk_ptr_ubounds (const void *@var{q})
11053 This built-in function checks if the pointer @var{q} is within the upper
11054 bound of its associated bounds. With Pointer Bounds Checker off, the built-in
11055 function call is ignored.
11059 @deftypefn {Built-in Function} void __builtin___bnd_chk_ptr_bounds (const void *@var{q}, size_t @var{size})
11061 This built-in function checks if [@var{q}, @var{q} + @var{size} - 1] is within
11062 the lower and upper bounds associated with @var{q}. With Pointer Bounds Checker
11063 off, the built-in function call is ignored.
11066 extern void *__wrap_memcpy (void *dst, const void *src, size_t n)
11070 __bnd_chk_ptr_bounds (dst, n);
11071 __bnd_chk_ptr_bounds (src, n);
11072 __real_memcpy (dst, src, n);
11080 @deftypefn {Built-in Function} {const void *} __builtin___bnd_get_ptr_lbound (const void *@var{q})
11082 This built-in function returns the lower bound associated
11083 with the pointer @var{q}, as a pointer value.
11084 This is useful for debugging using @code{printf}.
11085 With Pointer Bounds Checker off, the built-in function returns 0.
11088 void *lb = __builtin___bnd_get_ptr_lbound (q);
11089 void *ub = __builtin___bnd_get_ptr_ubound (q);
11090 printf ("q = %p lb(q) = %p ub(q) = %p", q, lb, ub);
11095 @deftypefn {Built-in Function} {const void *} __builtin___bnd_get_ptr_ubound (const void *@var{q})
11097 This built-in function returns the upper bound (which is a pointer) associated
11098 with the pointer @var{q}. With Pointer Bounds Checker off,
11099 the built-in function returns -1.
11103 @node Other Builtins
11104 @section Other Built-in Functions Provided by GCC
11105 @cindex built-in functions
11106 @findex __builtin_alloca
11107 @findex __builtin_alloca_with_align
11108 @findex __builtin_alloca_with_align_and_max
11109 @findex __builtin_call_with_static_chain
11110 @findex __builtin_extend_pointer
11111 @findex __builtin_fpclassify
11112 @findex __builtin_isfinite
11113 @findex __builtin_isnormal
11114 @findex __builtin_isgreater
11115 @findex __builtin_isgreaterequal
11116 @findex __builtin_isinf_sign
11117 @findex __builtin_isless
11118 @findex __builtin_islessequal
11119 @findex __builtin_islessgreater
11120 @findex __builtin_isunordered
11121 @findex __builtin_powi
11122 @findex __builtin_powif
11123 @findex __builtin_powil
11284 @findex fprintf_unlocked
11286 @findex fputs_unlocked
11394 @findex nexttowardf
11395 @findex nexttowardl
11403 @findex printf_unlocked
11433 @findex signbitd128
11434 @findex significand
11435 @findex significandf
11436 @findex significandl
11464 @findex strncasecmp
11507 GCC provides a large number of built-in functions other than the ones
11508 mentioned above. Some of these are for internal use in the processing
11509 of exceptions or variable-length argument lists and are not
11510 documented here because they may change from time to time; we do not
11511 recommend general use of these functions.
11513 The remaining functions are provided for optimization purposes.
11515 With the exception of built-ins that have library equivalents such as
11516 the standard C library functions discussed below, or that expand to
11517 library calls, GCC built-in functions are always expanded inline and
11518 thus do not have corresponding entry points and their address cannot
11519 be obtained. Attempting to use them in an expression other than
11520 a function call results in a compile-time error.
11522 @opindex fno-builtin
11523 GCC includes built-in versions of many of the functions in the standard
11524 C library. These functions come in two forms: one whose names start with
11525 the @code{__builtin_} prefix, and the other without. Both forms have the
11526 same type (including prototype), the same address (when their address is
11527 taken), and the same meaning as the C library functions even if you specify
11528 the @option{-fno-builtin} option @pxref{C Dialect Options}). Many of these
11529 functions are only optimized in certain cases; if they are not optimized in
11530 a particular case, a call to the library function is emitted.
11534 Outside strict ISO C mode (@option{-ansi}, @option{-std=c90},
11535 @option{-std=c99} or @option{-std=c11}), the functions
11536 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
11537 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
11538 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
11539 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
11540 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
11541 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
11542 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
11543 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
11544 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
11545 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
11546 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
11547 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
11548 @code{signbitd64}, @code{signbitd128}, @code{significandf},
11549 @code{significandl}, @code{significand}, @code{sincosf},
11550 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
11551 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
11552 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
11553 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
11555 may be handled as built-in functions.
11556 All these functions have corresponding versions
11557 prefixed with @code{__builtin_}, which may be used even in strict C90
11560 The ISO C99 functions
11561 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
11562 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
11563 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
11564 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
11565 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
11566 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
11567 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
11568 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
11569 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
11570 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
11571 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
11572 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
11573 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
11574 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
11575 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
11576 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
11577 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
11578 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
11579 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
11580 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
11581 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
11582 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
11583 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
11584 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
11585 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
11586 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
11587 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
11588 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
11589 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
11590 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
11591 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
11592 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
11593 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
11594 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
11595 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
11596 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
11597 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
11598 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
11599 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
11600 are handled as built-in functions
11601 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
11603 There are also built-in versions of the ISO C99 functions
11604 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
11605 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
11606 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
11607 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
11608 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
11609 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
11610 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
11611 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
11612 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
11613 that are recognized in any mode since ISO C90 reserves these names for
11614 the purpose to which ISO C99 puts them. All these functions have
11615 corresponding versions prefixed with @code{__builtin_}.
11617 There are also built-in functions @code{__builtin_fabsf@var{n}},
11618 @code{__builtin_fabsf@var{n}x}, @code{__builtin_copysignf@var{n}} and
11619 @code{__builtin_copysignf@var{n}x}, corresponding to the TS 18661-3
11620 functions @code{fabsf@var{n}}, @code{fabsf@var{n}x},
11621 @code{copysignf@var{n}} and @code{copysignf@var{n}x}, for supported
11622 types @code{_Float@var{n}} and @code{_Float@var{n}x}.
11624 There are also GNU extension functions @code{clog10}, @code{clog10f} and
11625 @code{clog10l} which names are reserved by ISO C99 for future use.
11626 All these functions have versions prefixed with @code{__builtin_}.
11628 The ISO C94 functions
11629 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
11630 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
11631 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
11633 are handled as built-in functions
11634 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
11636 The ISO C90 functions
11637 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
11638 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
11639 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
11640 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
11641 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
11642 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
11643 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
11644 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
11645 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
11646 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
11647 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
11648 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
11649 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
11650 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
11651 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
11652 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
11653 are all recognized as built-in functions unless
11654 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
11655 is specified for an individual function). All of these functions have
11656 corresponding versions prefixed with @code{__builtin_}.
11658 GCC provides built-in versions of the ISO C99 floating-point comparison
11659 macros that avoid raising exceptions for unordered operands. They have
11660 the same names as the standard macros ( @code{isgreater},
11661 @code{isgreaterequal}, @code{isless}, @code{islessequal},
11662 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
11663 prefixed. We intend for a library implementor to be able to simply
11664 @code{#define} each standard macro to its built-in equivalent.
11665 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
11666 @code{isinf_sign}, @code{isnormal} and @code{signbit} built-ins used with
11667 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
11668 built-in functions appear both with and without the @code{__builtin_} prefix.
11670 @deftypefn {Built-in Function} void *__builtin_alloca (size_t size)
11671 The @code{__builtin_alloca} function must be called at block scope.
11672 The function allocates an object @var{size} bytes large on the stack
11673 of the calling function. The object is aligned on the default stack
11674 alignment boundary for the target determined by the
11675 @code{__BIGGEST_ALIGNMENT__} macro. The @code{__builtin_alloca}
11676 function returns a pointer to the first byte of the allocated object.
11677 The lifetime of the allocated object ends just before the calling
11678 function returns to its caller. This is so even when
11679 @code{__builtin_alloca} is called within a nested block.
11681 For example, the following function allocates eight objects of @code{n}
11682 bytes each on the stack, storing a pointer to each in consecutive elements
11683 of the array @code{a}. It then passes the array to function @code{g}
11684 which can safely use the storage pointed to by each of the array elements.
11687 void f (unsigned n)
11690 for (int i = 0; i != 8; ++i)
11691 a [i] = __builtin_alloca (n);
11693 g (a, n); // @r{safe}
11697 Since the @code{__builtin_alloca} function doesn't validate its argument
11698 it is the responsibility of its caller to make sure the argument doesn't
11699 cause it to exceed the stack size limit.
11700 The @code{__builtin_alloca} function is provided to make it possible to
11701 allocate on the stack arrays of bytes with an upper bound that may be
11702 computed at run time. Since C99 Variable Length Arrays offer
11703 similar functionality under a portable, more convenient, and safer
11704 interface they are recommended instead, in both C99 and C++ programs
11705 where GCC provides them as an extension.
11706 @xref{Variable Length}, for details.
11710 @deftypefn {Built-in Function} void *__builtin_alloca_with_align (size_t size, size_t alignment)
11711 The @code{__builtin_alloca_with_align} function must be called at block
11712 scope. The function allocates an object @var{size} bytes large on
11713 the stack of the calling function. The allocated object is aligned on
11714 the boundary specified by the argument @var{alignment} whose unit is given
11715 in bits (not bytes). The @var{size} argument must be positive and not
11716 exceed the stack size limit. The @var{alignment} argument must be a constant
11717 integer expression that evaluates to a power of 2 greater than or equal to
11718 @code{CHAR_BIT} and less than some unspecified maximum. Invocations
11719 with other values are rejected with an error indicating the valid bounds.
11720 The function returns a pointer to the first byte of the allocated object.
11721 The lifetime of the allocated object ends at the end of the block in which
11722 the function was called. The allocated storage is released no later than
11723 just before the calling function returns to its caller, but may be released
11724 at the end of the block in which the function was called.
11726 For example, in the following function the call to @code{g} is unsafe
11727 because when @code{overalign} is non-zero, the space allocated by
11728 @code{__builtin_alloca_with_align} may have been released at the end
11729 of the @code{if} statement in which it was called.
11732 void f (unsigned n, bool overalign)
11736 p = __builtin_alloca_with_align (n, 64 /* bits */);
11738 p = __builtin_alloc (n);
11740 g (p, n); // @r{unsafe}
11744 Since the @code{__builtin_alloca_with_align} function doesn't validate its
11745 @var{size} argument it is the responsibility of its caller to make sure
11746 the argument doesn't cause it to exceed the stack size limit.
11747 The @code{__builtin_alloca_with_align} function is provided to make
11748 it possible to allocate on the stack overaligned arrays of bytes with
11749 an upper bound that may be computed at run time. Since C99
11750 Variable Length Arrays offer the same functionality under
11751 a portable, more convenient, and safer interface they are recommended
11752 instead, in both C99 and C++ programs where GCC provides them as
11753 an extension. @xref{Variable Length}, for details.
11757 @deftypefn {Built-in Function} void *__builtin_alloca_with_align_and_max (size_t size, size_t alignment, size_t max_size)
11758 Similar to @code{__builtin_alloca_with_align} but takes an extra argument
11759 specifying an upper bound for @var{size} in case its value cannot be computed
11760 at compile time, for use by @option{-fstack-usage}, @option{-Wstack-usage}
11761 and @option{-Walloca-larger-than}. @var{max_size} must be a constant integer
11762 expression, it has no effect on code generation and no attempt is made to
11763 check its compatibility with @var{size}.
11767 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
11769 You can use the built-in function @code{__builtin_types_compatible_p} to
11770 determine whether two types are the same.
11772 This built-in function returns 1 if the unqualified versions of the
11773 types @var{type1} and @var{type2} (which are types, not expressions) are
11774 compatible, 0 otherwise. The result of this built-in function can be
11775 used in integer constant expressions.
11777 This built-in function ignores top level qualifiers (e.g., @code{const},
11778 @code{volatile}). For example, @code{int} is equivalent to @code{const
11781 The type @code{int[]} and @code{int[5]} are compatible. On the other
11782 hand, @code{int} and @code{char *} are not compatible, even if the size
11783 of their types, on the particular architecture are the same. Also, the
11784 amount of pointer indirection is taken into account when determining
11785 similarity. Consequently, @code{short *} is not similar to
11786 @code{short **}. Furthermore, two types that are typedefed are
11787 considered compatible if their underlying types are compatible.
11789 An @code{enum} type is not considered to be compatible with another
11790 @code{enum} type even if both are compatible with the same integer
11791 type; this is what the C standard specifies.
11792 For example, @code{enum @{foo, bar@}} is not similar to
11793 @code{enum @{hot, dog@}}.
11795 You typically use this function in code whose execution varies
11796 depending on the arguments' types. For example:
11801 typeof (x) tmp = (x); \
11802 if (__builtin_types_compatible_p (typeof (x), long double)) \
11803 tmp = foo_long_double (tmp); \
11804 else if (__builtin_types_compatible_p (typeof (x), double)) \
11805 tmp = foo_double (tmp); \
11806 else if (__builtin_types_compatible_p (typeof (x), float)) \
11807 tmp = foo_float (tmp); \
11814 @emph{Note:} This construct is only available for C@.
11818 @deftypefn {Built-in Function} @var{type} __builtin_call_with_static_chain (@var{call_exp}, @var{pointer_exp})
11820 The @var{call_exp} expression must be a function call, and the
11821 @var{pointer_exp} expression must be a pointer. The @var{pointer_exp}
11822 is passed to the function call in the target's static chain location.
11823 The result of builtin is the result of the function call.
11825 @emph{Note:} This builtin is only available for C@.
11826 This builtin can be used to call Go closures from C.
11830 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
11832 You can use the built-in function @code{__builtin_choose_expr} to
11833 evaluate code depending on the value of a constant expression. This
11834 built-in function returns @var{exp1} if @var{const_exp}, which is an
11835 integer constant expression, is nonzero. Otherwise it returns @var{exp2}.
11837 This built-in function is analogous to the @samp{? :} operator in C,
11838 except that the expression returned has its type unaltered by promotion
11839 rules. Also, the built-in function does not evaluate the expression
11840 that is not chosen. For example, if @var{const_exp} evaluates to true,
11841 @var{exp2} is not evaluated even if it has side effects.
11843 This built-in function can return an lvalue if the chosen argument is an
11846 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
11847 type. Similarly, if @var{exp2} is returned, its return type is the same
11854 __builtin_choose_expr ( \
11855 __builtin_types_compatible_p (typeof (x), double), \
11857 __builtin_choose_expr ( \
11858 __builtin_types_compatible_p (typeof (x), float), \
11860 /* @r{The void expression results in a compile-time error} \
11861 @r{when assigning the result to something.} */ \
11865 @emph{Note:} This construct is only available for C@. Furthermore, the
11866 unused expression (@var{exp1} or @var{exp2} depending on the value of
11867 @var{const_exp}) may still generate syntax errors. This may change in
11872 @deftypefn {Built-in Function} @var{type} __builtin_tgmath (@var{functions}, @var{arguments})
11874 The built-in function @code{__builtin_tgmath}, available only for C
11875 and Objective-C, calls a function determined according to the rules of
11876 @code{<tgmath.h>} macros. It is intended to be used in
11877 implementations of that header, so that expansions of macros from that
11878 header only expand each of their arguments once, to avoid problems
11879 when calls to such macros are nested inside the arguments of other
11880 calls to such macros; in addition, it results in better diagnostics
11881 for invalid calls to @code{<tgmath.h>} macros than implementations
11882 using other GNU C language features. For example, the @code{pow}
11883 type-generic macro might be defined as:
11886 #define pow(a, b) __builtin_tgmath (powf, pow, powl, \
11887 cpowf, cpow, cpowl, a, b)
11890 The arguments to @code{__builtin_tgmath} are at least two pointers to
11891 functions, followed by the arguments to the type-generic macro (which
11892 will be passed as arguments to the selected function). All the
11893 pointers to functions must be pointers to prototyped functions, none
11894 of which may have variable arguments, and all of which must have the
11895 same number of parameters; the number of parameters of the first
11896 function determines how many arguments to @code{__builtin_tgmath} are
11897 interpreted as function pointers, and how many as the arguments to the
11900 The types of the specified functions must all be different, but
11901 related to each other in the same way as a set of functions that may
11902 be selected between by a macro in @code{<tgmath.h>}. This means that
11903 the functions are parameterized by a floating-point type @var{t},
11904 different for each such function. The function return types may all
11905 be the same type, or they may be @var{t} for each function, or they
11906 may be the real type corresponding to @var{t} for each function (if
11907 some of the types @var{t} are complex). Likewise, for each parameter
11908 position, the type of the parameter in that position may always be the
11909 same type, or may be @var{t} for each function (this case must apply
11910 for at least one parameter position), or may be the real type
11911 corresponding to @var{t} for each function.
11913 The standard rules for @code{<tgmath.h>} macros are used to find a
11914 common type @var{u} from the types of the arguments for parameters
11915 whose types vary between the functions; complex integer types (a GNU
11916 extension) are treated like @code{_Complex double} for this purpose
11917 (or @code{_Complex _Float64} if all the function return types are the
11918 same @code{_Float@var{n}} or @code{_Float@var{n}x} type).
11919 If the function return types vary, or are all the same integer type,
11920 the function called is the one for which @var{t} is @var{u}, and it is
11921 an error if there is no such function. If the function return types
11922 are all the same floating-point type, the type-generic macro is taken
11923 to be one of those from TS 18661 that rounds the result to a narrower
11924 type; if there is a function for which @var{t} is @var{u}, it is
11925 called, and otherwise the first function, if any, for which @var{t}
11926 has at least the range and precision of @var{u} is called, and it is
11927 an error if there is no such function.
11931 @deftypefn {Built-in Function} @var{type} __builtin_complex (@var{real}, @var{imag})
11933 The built-in function @code{__builtin_complex} is provided for use in
11934 implementing the ISO C11 macros @code{CMPLXF}, @code{CMPLX} and
11935 @code{CMPLXL}. @var{real} and @var{imag} must have the same type, a
11936 real binary floating-point type, and the result has the corresponding
11937 complex type with real and imaginary parts @var{real} and @var{imag}.
11938 Unlike @samp{@var{real} + I * @var{imag}}, this works even when
11939 infinities, NaNs and negative zeros are involved.
11943 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
11944 You can use the built-in function @code{__builtin_constant_p} to
11945 determine if a value is known to be constant at compile time and hence
11946 that GCC can perform constant-folding on expressions involving that
11947 value. The argument of the function is the value to test. The function
11948 returns the integer 1 if the argument is known to be a compile-time
11949 constant and 0 if it is not known to be a compile-time constant. A
11950 return of 0 does not indicate that the value is @emph{not} a constant,
11951 but merely that GCC cannot prove it is a constant with the specified
11952 value of the @option{-O} option.
11954 You typically use this function in an embedded application where
11955 memory is a critical resource. If you have some complex calculation,
11956 you may want it to be folded if it involves constants, but need to call
11957 a function if it does not. For example:
11960 #define Scale_Value(X) \
11961 (__builtin_constant_p (X) \
11962 ? ((X) * SCALE + OFFSET) : Scale (X))
11965 You may use this built-in function in either a macro or an inline
11966 function. However, if you use it in an inlined function and pass an
11967 argument of the function as the argument to the built-in, GCC
11968 never returns 1 when you call the inline function with a string constant
11969 or compound literal (@pxref{Compound Literals}) and does not return 1
11970 when you pass a constant numeric value to the inline function unless you
11971 specify the @option{-O} option.
11973 You may also use @code{__builtin_constant_p} in initializers for static
11974 data. For instance, you can write
11977 static const int table[] = @{
11978 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
11984 This is an acceptable initializer even if @var{EXPRESSION} is not a
11985 constant expression, including the case where
11986 @code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be
11987 folded to a constant but @var{EXPRESSION} contains operands that are
11988 not otherwise permitted in a static initializer (for example,
11989 @code{0 && foo ()}). GCC must be more conservative about evaluating the
11990 built-in in this case, because it has no opportunity to perform
11994 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
11995 @opindex fprofile-arcs
11996 You may use @code{__builtin_expect} to provide the compiler with
11997 branch prediction information. In general, you should prefer to
11998 use actual profile feedback for this (@option{-fprofile-arcs}), as
11999 programmers are notoriously bad at predicting how their programs
12000 actually perform. However, there are applications in which this
12001 data is hard to collect.
12003 The return value is the value of @var{exp}, which should be an integral
12004 expression. The semantics of the built-in are that it is expected that
12005 @var{exp} == @var{c}. For example:
12008 if (__builtin_expect (x, 0))
12013 indicates that we do not expect to call @code{foo}, since
12014 we expect @code{x} to be zero. Since you are limited to integral
12015 expressions for @var{exp}, you should use constructions such as
12018 if (__builtin_expect (ptr != NULL, 1))
12023 when testing pointer or floating-point values.
12026 @deftypefn {Built-in Function} void __builtin_trap (void)
12027 This function causes the program to exit abnormally. GCC implements
12028 this function by using a target-dependent mechanism (such as
12029 intentionally executing an illegal instruction) or by calling
12030 @code{abort}. The mechanism used may vary from release to release so
12031 you should not rely on any particular implementation.
12034 @deftypefn {Built-in Function} void __builtin_unreachable (void)
12035 If control flow reaches the point of the @code{__builtin_unreachable},
12036 the program is undefined. It is useful in situations where the
12037 compiler cannot deduce the unreachability of the code.
12039 One such case is immediately following an @code{asm} statement that
12040 either never terminates, or one that transfers control elsewhere
12041 and never returns. In this example, without the
12042 @code{__builtin_unreachable}, GCC issues a warning that control
12043 reaches the end of a non-void function. It also generates code
12044 to return after the @code{asm}.
12047 int f (int c, int v)
12055 asm("jmp error_handler");
12056 __builtin_unreachable ();
12062 Because the @code{asm} statement unconditionally transfers control out
12063 of the function, control never reaches the end of the function
12064 body. The @code{__builtin_unreachable} is in fact unreachable and
12065 communicates this fact to the compiler.
12067 Another use for @code{__builtin_unreachable} is following a call a
12068 function that never returns but that is not declared
12069 @code{__attribute__((noreturn))}, as in this example:
12072 void function_that_never_returns (void);
12082 function_that_never_returns ();
12083 __builtin_unreachable ();
12090 @deftypefn {Built-in Function} {void *} __builtin_assume_aligned (const void *@var{exp}, size_t @var{align}, ...)
12091 This function returns its first argument, and allows the compiler
12092 to assume that the returned pointer is at least @var{align} bytes
12093 aligned. This built-in can have either two or three arguments,
12094 if it has three, the third argument should have integer type, and
12095 if it is nonzero means misalignment offset. For example:
12098 void *x = __builtin_assume_aligned (arg, 16);
12102 means that the compiler can assume @code{x}, set to @code{arg}, is at least
12103 16-byte aligned, while:
12106 void *x = __builtin_assume_aligned (arg, 32, 8);
12110 means that the compiler can assume for @code{x}, set to @code{arg}, that
12111 @code{(char *) x - 8} is 32-byte aligned.
12114 @deftypefn {Built-in Function} int __builtin_LINE ()
12115 This function is the equivalent of the preprocessor @code{__LINE__}
12116 macro and returns a constant integer expression that evaluates to
12117 the line number of the invocation of the built-in. When used as a C++
12118 default argument for a function @var{F}, it returns the line number
12119 of the call to @var{F}.
12122 @deftypefn {Built-in Function} {const char *} __builtin_FUNCTION ()
12123 This function is the equivalent of the @code{__FUNCTION__} symbol
12124 and returns an address constant pointing to the name of the function
12125 from which the built-in was invoked, or the empty string if
12126 the invocation is not at function scope. When used as a C++ default
12127 argument for a function @var{F}, it returns the name of @var{F}'s
12128 caller or the empty string if the call was not made at function
12132 @deftypefn {Built-in Function} {const char *} __builtin_FILE ()
12133 This function is the equivalent of the preprocessor @code{__FILE__}
12134 macro and returns an address constant pointing to the file name
12135 containing the invocation of the built-in, or the empty string if
12136 the invocation is not at function scope. When used as a C++ default
12137 argument for a function @var{F}, it returns the file name of the call
12138 to @var{F} or the empty string if the call was not made at function
12141 For example, in the following, each call to function @code{foo} will
12142 print a line similar to @code{"file.c:123: foo: message"} with the name
12143 of the file and the line number of the @code{printf} call, the name of
12144 the function @code{foo}, followed by the word @code{message}.
12148 function (const char *func = __builtin_FUNCTION ())
12155 printf ("%s:%i: %s: message\n", file (), line (), function ());
12161 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
12162 This function is used to flush the processor's instruction cache for
12163 the region of memory between @var{begin} inclusive and @var{end}
12164 exclusive. Some targets require that the instruction cache be
12165 flushed, after modifying memory containing code, in order to obtain
12166 deterministic behavior.
12168 If the target does not require instruction cache flushes,
12169 @code{__builtin___clear_cache} has no effect. Otherwise either
12170 instructions are emitted in-line to clear the instruction cache or a
12171 call to the @code{__clear_cache} function in libgcc is made.
12174 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
12175 This function is used to minimize cache-miss latency by moving data into
12176 a cache before it is accessed.
12177 You can insert calls to @code{__builtin_prefetch} into code for which
12178 you know addresses of data in memory that is likely to be accessed soon.
12179 If the target supports them, data prefetch instructions are generated.
12180 If the prefetch is done early enough before the access then the data will
12181 be in the cache by the time it is accessed.
12183 The value of @var{addr} is the address of the memory to prefetch.
12184 There are two optional arguments, @var{rw} and @var{locality}.
12185 The value of @var{rw} is a compile-time constant one or zero; one
12186 means that the prefetch is preparing for a write to the memory address
12187 and zero, the default, means that the prefetch is preparing for a read.
12188 The value @var{locality} must be a compile-time constant integer between
12189 zero and three. A value of zero means that the data has no temporal
12190 locality, so it need not be left in the cache after the access. A value
12191 of three means that the data has a high degree of temporal locality and
12192 should be left in all levels of cache possible. Values of one and two
12193 mean, respectively, a low or moderate degree of temporal locality. The
12197 for (i = 0; i < n; i++)
12199 a[i] = a[i] + b[i];
12200 __builtin_prefetch (&a[i+j], 1, 1);
12201 __builtin_prefetch (&b[i+j], 0, 1);
12206 Data prefetch does not generate faults if @var{addr} is invalid, but
12207 the address expression itself must be valid. For example, a prefetch
12208 of @code{p->next} does not fault if @code{p->next} is not a valid
12209 address, but evaluation faults if @code{p} is not a valid address.
12211 If the target does not support data prefetch, the address expression
12212 is evaluated if it includes side effects but no other code is generated
12213 and GCC does not issue a warning.
12216 @deftypefn {Built-in Function} double __builtin_huge_val (void)
12217 Returns a positive infinity, if supported by the floating-point format,
12218 else @code{DBL_MAX}. This function is suitable for implementing the
12219 ISO C macro @code{HUGE_VAL}.
12222 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
12223 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
12226 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
12227 Similar to @code{__builtin_huge_val}, except the return
12228 type is @code{long double}.
12231 @deftypefn {Built-in Function} _Float@var{n} __builtin_huge_valf@var{n} (void)
12232 Similar to @code{__builtin_huge_val}, except the return type is
12233 @code{_Float@var{n}}.
12236 @deftypefn {Built-in Function} _Float@var{n}x __builtin_huge_valf@var{n}x (void)
12237 Similar to @code{__builtin_huge_val}, except the return type is
12238 @code{_Float@var{n}x}.
12241 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
12242 This built-in implements the C99 fpclassify functionality. The first
12243 five int arguments should be the target library's notion of the
12244 possible FP classes and are used for return values. They must be
12245 constant values and they must appear in this order: @code{FP_NAN},
12246 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
12247 @code{FP_ZERO}. The ellipsis is for exactly one floating-point value
12248 to classify. GCC treats the last argument as type-generic, which
12249 means it does not do default promotion from float to double.
12252 @deftypefn {Built-in Function} double __builtin_inf (void)
12253 Similar to @code{__builtin_huge_val}, except a warning is generated
12254 if the target floating-point format does not support infinities.
12257 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
12258 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
12261 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
12262 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
12265 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
12266 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
12269 @deftypefn {Built-in Function} float __builtin_inff (void)
12270 Similar to @code{__builtin_inf}, except the return type is @code{float}.
12271 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
12274 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
12275 Similar to @code{__builtin_inf}, except the return
12276 type is @code{long double}.
12279 @deftypefn {Built-in Function} _Float@var{n} __builtin_inff@var{n} (void)
12280 Similar to @code{__builtin_inf}, except the return
12281 type is @code{_Float@var{n}}.
12284 @deftypefn {Built-in Function} _Float@var{n} __builtin_inff@var{n}x (void)
12285 Similar to @code{__builtin_inf}, except the return
12286 type is @code{_Float@var{n}x}.
12289 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
12290 Similar to @code{isinf}, except the return value is -1 for
12291 an argument of @code{-Inf} and 1 for an argument of @code{+Inf}.
12292 Note while the parameter list is an
12293 ellipsis, this function only accepts exactly one floating-point
12294 argument. GCC treats this parameter as type-generic, which means it
12295 does not do default promotion from float to double.
12298 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
12299 This is an implementation of the ISO C99 function @code{nan}.
12301 Since ISO C99 defines this function in terms of @code{strtod}, which we
12302 do not implement, a description of the parsing is in order. The string
12303 is parsed as by @code{strtol}; that is, the base is recognized by
12304 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
12305 in the significand such that the least significant bit of the number
12306 is at the least significant bit of the significand. The number is
12307 truncated to fit the significand field provided. The significand is
12308 forced to be a quiet NaN@.
12310 This function, if given a string literal all of which would have been
12311 consumed by @code{strtol}, is evaluated early enough that it is considered a
12312 compile-time constant.
12315 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
12316 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
12319 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
12320 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
12323 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
12324 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
12327 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
12328 Similar to @code{__builtin_nan}, except the return type is @code{float}.
12331 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
12332 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
12335 @deftypefn {Built-in Function} _Float@var{n} __builtin_nanf@var{n} (const char *str)
12336 Similar to @code{__builtin_nan}, except the return type is
12337 @code{_Float@var{n}}.
12340 @deftypefn {Built-in Function} _Float@var{n}x __builtin_nanf@var{n}x (const char *str)
12341 Similar to @code{__builtin_nan}, except the return type is
12342 @code{_Float@var{n}x}.
12345 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
12346 Similar to @code{__builtin_nan}, except the significand is forced
12347 to be a signaling NaN@. The @code{nans} function is proposed by
12348 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
12351 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
12352 Similar to @code{__builtin_nans}, except the return type is @code{float}.
12355 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
12356 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
12359 @deftypefn {Built-in Function} _Float@var{n} __builtin_nansf@var{n} (const char *str)
12360 Similar to @code{__builtin_nans}, except the return type is
12361 @code{_Float@var{n}}.
12364 @deftypefn {Built-in Function} _Float@var{n}x __builtin_nansf@var{n}x (const char *str)
12365 Similar to @code{__builtin_nans}, except the return type is
12366 @code{_Float@var{n}x}.
12369 @deftypefn {Built-in Function} int __builtin_ffs (int x)
12370 Returns one plus the index of the least significant 1-bit of @var{x}, or
12371 if @var{x} is zero, returns zero.
12374 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
12375 Returns the number of leading 0-bits in @var{x}, starting at the most
12376 significant bit position. If @var{x} is 0, the result is undefined.
12379 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
12380 Returns the number of trailing 0-bits in @var{x}, starting at the least
12381 significant bit position. If @var{x} is 0, the result is undefined.
12384 @deftypefn {Built-in Function} int __builtin_clrsb (int x)
12385 Returns the number of leading redundant sign bits in @var{x}, i.e.@: the
12386 number of bits following the most significant bit that are identical
12387 to it. There are no special cases for 0 or other values.
12390 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
12391 Returns the number of 1-bits in @var{x}.
12394 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
12395 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
12399 @deftypefn {Built-in Function} int __builtin_ffsl (long)
12400 Similar to @code{__builtin_ffs}, except the argument type is
12404 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
12405 Similar to @code{__builtin_clz}, except the argument type is
12406 @code{unsigned long}.
12409 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
12410 Similar to @code{__builtin_ctz}, except the argument type is
12411 @code{unsigned long}.
12414 @deftypefn {Built-in Function} int __builtin_clrsbl (long)
12415 Similar to @code{__builtin_clrsb}, except the argument type is
12419 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
12420 Similar to @code{__builtin_popcount}, except the argument type is
12421 @code{unsigned long}.
12424 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
12425 Similar to @code{__builtin_parity}, except the argument type is
12426 @code{unsigned long}.
12429 @deftypefn {Built-in Function} int __builtin_ffsll (long long)
12430 Similar to @code{__builtin_ffs}, except the argument type is
12434 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
12435 Similar to @code{__builtin_clz}, except the argument type is
12436 @code{unsigned long long}.
12439 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
12440 Similar to @code{__builtin_ctz}, except the argument type is
12441 @code{unsigned long long}.
12444 @deftypefn {Built-in Function} int __builtin_clrsbll (long long)
12445 Similar to @code{__builtin_clrsb}, except the argument type is
12449 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
12450 Similar to @code{__builtin_popcount}, except the argument type is
12451 @code{unsigned long long}.
12454 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
12455 Similar to @code{__builtin_parity}, except the argument type is
12456 @code{unsigned long long}.
12459 @deftypefn {Built-in Function} double __builtin_powi (double, int)
12460 Returns the first argument raised to the power of the second. Unlike the
12461 @code{pow} function no guarantees about precision and rounding are made.
12464 @deftypefn {Built-in Function} float __builtin_powif (float, int)
12465 Similar to @code{__builtin_powi}, except the argument and return types
12469 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
12470 Similar to @code{__builtin_powi}, except the argument and return types
12471 are @code{long double}.
12474 @deftypefn {Built-in Function} uint16_t __builtin_bswap16 (uint16_t x)
12475 Returns @var{x} with the order of the bytes reversed; for example,
12476 @code{0xaabb} becomes @code{0xbbaa}. Byte here always means
12480 @deftypefn {Built-in Function} uint32_t __builtin_bswap32 (uint32_t x)
12481 Similar to @code{__builtin_bswap16}, except the argument and return types
12485 @deftypefn {Built-in Function} uint64_t __builtin_bswap64 (uint64_t x)
12486 Similar to @code{__builtin_bswap32}, except the argument and return types
12490 @deftypefn {Built-in Function} Pmode __builtin_extend_pointer (void * x)
12491 On targets where the user visible pointer size is smaller than the size
12492 of an actual hardware address this function returns the extended user
12493 pointer. Targets where this is true included ILP32 mode on x86_64 or
12494 Aarch64. This function is mainly useful when writing inline assembly
12498 @deftypefn {Built-in Function} int __builtin_goacc_parlevel_id (int x)
12499 Returns the openacc gang, worker or vector id depending on whether @var{x} is
12503 @deftypefn {Built-in Function} int __builtin_goacc_parlevel_size (int x)
12504 Returns the openacc gang, worker or vector size depending on whether @var{x} is
12508 @node Target Builtins
12509 @section Built-in Functions Specific to Particular Target Machines
12511 On some target machines, GCC supports many built-in functions specific
12512 to those machines. Generally these generate calls to specific machine
12513 instructions, but allow the compiler to schedule those calls.
12516 * AArch64 Built-in Functions::
12517 * Alpha Built-in Functions::
12518 * Altera Nios II Built-in Functions::
12519 * ARC Built-in Functions::
12520 * ARC SIMD Built-in Functions::
12521 * ARM iWMMXt Built-in Functions::
12522 * ARM C Language Extensions (ACLE)::
12523 * ARM Floating Point Status and Control Intrinsics::
12524 * ARM ARMv8-M Security Extensions::
12525 * AVR Built-in Functions::
12526 * Blackfin Built-in Functions::
12527 * FR-V Built-in Functions::
12528 * MIPS DSP Built-in Functions::
12529 * MIPS Paired-Single Support::
12530 * MIPS Loongson Built-in Functions::
12531 * MIPS SIMD Architecture (MSA) Support::
12532 * Other MIPS Built-in Functions::
12533 * MSP430 Built-in Functions::
12534 * NDS32 Built-in Functions::
12535 * picoChip Built-in Functions::
12536 * Basic PowerPC Built-in Functions::
12537 * PowerPC AltiVec/VSX Built-in Functions::
12538 * PowerPC Hardware Transactional Memory Built-in Functions::
12539 * PowerPC Atomic Memory Operation Functions::
12540 * RX Built-in Functions::
12541 * S/390 System z Built-in Functions::
12542 * SH Built-in Functions::
12543 * SPARC VIS Built-in Functions::
12544 * SPU Built-in Functions::
12545 * TI C6X Built-in Functions::
12546 * TILE-Gx Built-in Functions::
12547 * TILEPro Built-in Functions::
12548 * x86 Built-in Functions::
12549 * x86 transactional memory intrinsics::
12550 * x86 control-flow protection intrinsics::
12553 @node AArch64 Built-in Functions
12554 @subsection AArch64 Built-in Functions
12556 These built-in functions are available for the AArch64 family of
12559 unsigned int __builtin_aarch64_get_fpcr ()
12560 void __builtin_aarch64_set_fpcr (unsigned int)
12561 unsigned int __builtin_aarch64_get_fpsr ()
12562 void __builtin_aarch64_set_fpsr (unsigned int)
12565 @node Alpha Built-in Functions
12566 @subsection Alpha Built-in Functions
12568 These built-in functions are available for the Alpha family of
12569 processors, depending on the command-line switches used.
12571 The following built-in functions are always available. They
12572 all generate the machine instruction that is part of the name.
12575 long __builtin_alpha_implver (void)
12576 long __builtin_alpha_rpcc (void)
12577 long __builtin_alpha_amask (long)
12578 long __builtin_alpha_cmpbge (long, long)
12579 long __builtin_alpha_extbl (long, long)
12580 long __builtin_alpha_extwl (long, long)
12581 long __builtin_alpha_extll (long, long)
12582 long __builtin_alpha_extql (long, long)
12583 long __builtin_alpha_extwh (long, long)
12584 long __builtin_alpha_extlh (long, long)
12585 long __builtin_alpha_extqh (long, long)
12586 long __builtin_alpha_insbl (long, long)
12587 long __builtin_alpha_inswl (long, long)
12588 long __builtin_alpha_insll (long, long)
12589 long __builtin_alpha_insql (long, long)
12590 long __builtin_alpha_inswh (long, long)
12591 long __builtin_alpha_inslh (long, long)
12592 long __builtin_alpha_insqh (long, long)
12593 long __builtin_alpha_mskbl (long, long)
12594 long __builtin_alpha_mskwl (long, long)
12595 long __builtin_alpha_mskll (long, long)
12596 long __builtin_alpha_mskql (long, long)
12597 long __builtin_alpha_mskwh (long, long)
12598 long __builtin_alpha_msklh (long, long)
12599 long __builtin_alpha_mskqh (long, long)
12600 long __builtin_alpha_umulh (long, long)
12601 long __builtin_alpha_zap (long, long)
12602 long __builtin_alpha_zapnot (long, long)
12605 The following built-in functions are always with @option{-mmax}
12606 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
12607 later. They all generate the machine instruction that is part
12611 long __builtin_alpha_pklb (long)
12612 long __builtin_alpha_pkwb (long)
12613 long __builtin_alpha_unpkbl (long)
12614 long __builtin_alpha_unpkbw (long)
12615 long __builtin_alpha_minub8 (long, long)
12616 long __builtin_alpha_minsb8 (long, long)
12617 long __builtin_alpha_minuw4 (long, long)
12618 long __builtin_alpha_minsw4 (long, long)
12619 long __builtin_alpha_maxub8 (long, long)
12620 long __builtin_alpha_maxsb8 (long, long)
12621 long __builtin_alpha_maxuw4 (long, long)
12622 long __builtin_alpha_maxsw4 (long, long)
12623 long __builtin_alpha_perr (long, long)
12626 The following built-in functions are always with @option{-mcix}
12627 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
12628 later. They all generate the machine instruction that is part
12632 long __builtin_alpha_cttz (long)
12633 long __builtin_alpha_ctlz (long)
12634 long __builtin_alpha_ctpop (long)
12637 The following built-in functions are available on systems that use the OSF/1
12638 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
12639 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
12640 @code{rdval} and @code{wrval}.
12643 void *__builtin_thread_pointer (void)
12644 void __builtin_set_thread_pointer (void *)
12647 @node Altera Nios II Built-in Functions
12648 @subsection Altera Nios II Built-in Functions
12650 These built-in functions are available for the Altera Nios II
12651 family of processors.
12653 The following built-in functions are always available. They
12654 all generate the machine instruction that is part of the name.
12657 int __builtin_ldbio (volatile const void *)
12658 int __builtin_ldbuio (volatile const void *)
12659 int __builtin_ldhio (volatile const void *)
12660 int __builtin_ldhuio (volatile const void *)
12661 int __builtin_ldwio (volatile const void *)
12662 void __builtin_stbio (volatile void *, int)
12663 void __builtin_sthio (volatile void *, int)
12664 void __builtin_stwio (volatile void *, int)
12665 void __builtin_sync (void)
12666 int __builtin_rdctl (int)
12667 int __builtin_rdprs (int, int)
12668 void __builtin_wrctl (int, int)
12669 void __builtin_flushd (volatile void *)
12670 void __builtin_flushda (volatile void *)
12671 int __builtin_wrpie (int);
12672 void __builtin_eni (int);
12673 int __builtin_ldex (volatile const void *)
12674 int __builtin_stex (volatile void *, int)
12675 int __builtin_ldsex (volatile const void *)
12676 int __builtin_stsex (volatile void *, int)
12679 The following built-in functions are always available. They
12680 all generate a Nios II Custom Instruction. The name of the
12681 function represents the types that the function takes and
12682 returns. The letter before the @code{n} is the return type
12683 or void if absent. The @code{n} represents the first parameter
12684 to all the custom instructions, the custom instruction number.
12685 The two letters after the @code{n} represent the up to two
12686 parameters to the function.
12688 The letters represent the following data types:
12691 @code{void} for return type and no parameter for parameter types.
12694 @code{int} for return type and parameter type
12697 @code{float} for return type and parameter type
12700 @code{void *} for return type and parameter type
12704 And the function names are:
12706 void __builtin_custom_n (void)
12707 void __builtin_custom_ni (int)
12708 void __builtin_custom_nf (float)
12709 void __builtin_custom_np (void *)
12710 void __builtin_custom_nii (int, int)
12711 void __builtin_custom_nif (int, float)
12712 void __builtin_custom_nip (int, void *)
12713 void __builtin_custom_nfi (float, int)
12714 void __builtin_custom_nff (float, float)
12715 void __builtin_custom_nfp (float, void *)
12716 void __builtin_custom_npi (void *, int)
12717 void __builtin_custom_npf (void *, float)
12718 void __builtin_custom_npp (void *, void *)
12719 int __builtin_custom_in (void)
12720 int __builtin_custom_ini (int)
12721 int __builtin_custom_inf (float)
12722 int __builtin_custom_inp (void *)
12723 int __builtin_custom_inii (int, int)
12724 int __builtin_custom_inif (int, float)
12725 int __builtin_custom_inip (int, void *)
12726 int __builtin_custom_infi (float, int)
12727 int __builtin_custom_inff (float, float)
12728 int __builtin_custom_infp (float, void *)
12729 int __builtin_custom_inpi (void *, int)
12730 int __builtin_custom_inpf (void *, float)
12731 int __builtin_custom_inpp (void *, void *)
12732 float __builtin_custom_fn (void)
12733 float __builtin_custom_fni (int)
12734 float __builtin_custom_fnf (float)
12735 float __builtin_custom_fnp (void *)
12736 float __builtin_custom_fnii (int, int)
12737 float __builtin_custom_fnif (int, float)
12738 float __builtin_custom_fnip (int, void *)
12739 float __builtin_custom_fnfi (float, int)
12740 float __builtin_custom_fnff (float, float)
12741 float __builtin_custom_fnfp (float, void *)
12742 float __builtin_custom_fnpi (void *, int)
12743 float __builtin_custom_fnpf (void *, float)
12744 float __builtin_custom_fnpp (void *, void *)
12745 void * __builtin_custom_pn (void)
12746 void * __builtin_custom_pni (int)
12747 void * __builtin_custom_pnf (float)
12748 void * __builtin_custom_pnp (void *)
12749 void * __builtin_custom_pnii (int, int)
12750 void * __builtin_custom_pnif (int, float)
12751 void * __builtin_custom_pnip (int, void *)
12752 void * __builtin_custom_pnfi (float, int)
12753 void * __builtin_custom_pnff (float, float)
12754 void * __builtin_custom_pnfp (float, void *)
12755 void * __builtin_custom_pnpi (void *, int)
12756 void * __builtin_custom_pnpf (void *, float)
12757 void * __builtin_custom_pnpp (void *, void *)
12760 @node ARC Built-in Functions
12761 @subsection ARC Built-in Functions
12763 The following built-in functions are provided for ARC targets. The
12764 built-ins generate the corresponding assembly instructions. In the
12765 examples given below, the generated code often requires an operand or
12766 result to be in a register. Where necessary further code will be
12767 generated to ensure this is true, but for brevity this is not
12768 described in each case.
12770 @emph{Note:} Using a built-in to generate an instruction not supported
12771 by a target may cause problems. At present the compiler is not
12772 guaranteed to detect such misuse, and as a result an internal compiler
12773 error may be generated.
12775 @deftypefn {Built-in Function} int __builtin_arc_aligned (void *@var{val}, int @var{alignval})
12776 Return 1 if @var{val} is known to have the byte alignment given
12777 by @var{alignval}, otherwise return 0.
12778 Note that this is different from
12780 __alignof__(*(char *)@var{val}) >= alignval
12782 because __alignof__ sees only the type of the dereference, whereas
12783 __builtin_arc_align uses alignment information from the pointer
12784 as well as from the pointed-to type.
12785 The information available will depend on optimization level.
12788 @deftypefn {Built-in Function} void __builtin_arc_brk (void)
12795 @deftypefn {Built-in Function} {unsigned int} __builtin_arc_core_read (unsigned int @var{regno})
12796 The operand is the number of a register to be read. Generates:
12798 mov @var{dest}, r@var{regno}
12800 where the value in @var{dest} will be the result returned from the
12804 @deftypefn {Built-in Function} void __builtin_arc_core_write (unsigned int @var{regno}, unsigned int @var{val})
12805 The first operand is the number of a register to be written, the
12806 second operand is a compile time constant to write into that
12807 register. Generates:
12809 mov r@var{regno}, @var{val}
12813 @deftypefn {Built-in Function} int __builtin_arc_divaw (int @var{a}, int @var{b})
12814 Only available if either @option{-mcpu=ARC700} or @option{-meA} is set.
12817 divaw @var{dest}, @var{a}, @var{b}
12819 where the value in @var{dest} will be the result returned from the
12823 @deftypefn {Built-in Function} void __builtin_arc_flag (unsigned int @var{a})
12830 @deftypefn {Built-in Function} {unsigned int} __builtin_arc_lr (unsigned int @var{auxr})
12831 The operand, @var{auxv}, is the address of an auxiliary register and
12832 must be a compile time constant. Generates:
12834 lr @var{dest}, [@var{auxr}]
12836 Where the value in @var{dest} will be the result returned from the
12840 @deftypefn {Built-in Function} void __builtin_arc_mul64 (int @var{a}, int @var{b})
12841 Only available with @option{-mmul64}. Generates:
12843 mul64 @var{a}, @var{b}
12847 @deftypefn {Built-in Function} void __builtin_arc_mulu64 (unsigned int @var{a}, unsigned int @var{b})
12848 Only available with @option{-mmul64}. Generates:
12850 mulu64 @var{a}, @var{b}
12854 @deftypefn {Built-in Function} void __builtin_arc_nop (void)
12861 @deftypefn {Built-in Function} int __builtin_arc_norm (int @var{src})
12862 Only valid if the @samp{norm} instruction is available through the
12863 @option{-mnorm} option or by default with @option{-mcpu=ARC700}.
12866 norm @var{dest}, @var{src}
12868 Where the value in @var{dest} will be the result returned from the
12872 @deftypefn {Built-in Function} {short int} __builtin_arc_normw (short int @var{src})
12873 Only valid if the @samp{normw} instruction is available through the
12874 @option{-mnorm} option or by default with @option{-mcpu=ARC700}.
12877 normw @var{dest}, @var{src}
12879 Where the value in @var{dest} will be the result returned from the
12883 @deftypefn {Built-in Function} void __builtin_arc_rtie (void)
12890 @deftypefn {Built-in Function} void __builtin_arc_sleep (int @var{a}
12897 @deftypefn {Built-in Function} void __builtin_arc_sr (unsigned int @var{auxr}, unsigned int @var{val})
12898 The first argument, @var{auxv}, is the address of an auxiliary
12899 register, the second argument, @var{val}, is a compile time constant
12900 to be written to the register. Generates:
12902 sr @var{auxr}, [@var{val}]
12906 @deftypefn {Built-in Function} int __builtin_arc_swap (int @var{src})
12907 Only valid with @option{-mswap}. Generates:
12909 swap @var{dest}, @var{src}
12911 Where the value in @var{dest} will be the result returned from the
12915 @deftypefn {Built-in Function} void __builtin_arc_swi (void)
12922 @deftypefn {Built-in Function} void __builtin_arc_sync (void)
12923 Only available with @option{-mcpu=ARC700}. Generates:
12929 @deftypefn {Built-in Function} void __builtin_arc_trap_s (unsigned int @var{c})
12930 Only available with @option{-mcpu=ARC700}. Generates:
12936 @deftypefn {Built-in Function} void __builtin_arc_unimp_s (void)
12937 Only available with @option{-mcpu=ARC700}. Generates:
12943 The instructions generated by the following builtins are not
12944 considered as candidates for scheduling. They are not moved around by
12945 the compiler during scheduling, and thus can be expected to appear
12946 where they are put in the C code:
12948 __builtin_arc_brk()
12949 __builtin_arc_core_read()
12950 __builtin_arc_core_write()
12951 __builtin_arc_flag()
12953 __builtin_arc_sleep()
12955 __builtin_arc_swi()
12958 @node ARC SIMD Built-in Functions
12959 @subsection ARC SIMD Built-in Functions
12961 SIMD builtins provided by the compiler can be used to generate the
12962 vector instructions. This section describes the available builtins
12963 and their usage in programs. With the @option{-msimd} option, the
12964 compiler provides 128-bit vector types, which can be specified using
12965 the @code{vector_size} attribute. The header file @file{arc-simd.h}
12966 can be included to use the following predefined types:
12968 typedef int __v4si __attribute__((vector_size(16)));
12969 typedef short __v8hi __attribute__((vector_size(16)));
12972 These types can be used to define 128-bit variables. The built-in
12973 functions listed in the following section can be used on these
12974 variables to generate the vector operations.
12976 For all builtins, @code{__builtin_arc_@var{someinsn}}, the header file
12977 @file{arc-simd.h} also provides equivalent macros called
12978 @code{_@var{someinsn}} that can be used for programming ease and
12979 improved readability. The following macros for DMA control are also
12982 #define _setup_dma_in_channel_reg _vdiwr
12983 #define _setup_dma_out_channel_reg _vdowr
12986 The following is a complete list of all the SIMD built-ins provided
12987 for ARC, grouped by calling signature.
12989 The following take two @code{__v8hi} arguments and return a
12990 @code{__v8hi} result:
12992 __v8hi __builtin_arc_vaddaw (__v8hi, __v8hi)
12993 __v8hi __builtin_arc_vaddw (__v8hi, __v8hi)
12994 __v8hi __builtin_arc_vand (__v8hi, __v8hi)
12995 __v8hi __builtin_arc_vandaw (__v8hi, __v8hi)
12996 __v8hi __builtin_arc_vavb (__v8hi, __v8hi)
12997 __v8hi __builtin_arc_vavrb (__v8hi, __v8hi)
12998 __v8hi __builtin_arc_vbic (__v8hi, __v8hi)
12999 __v8hi __builtin_arc_vbicaw (__v8hi, __v8hi)
13000 __v8hi __builtin_arc_vdifaw (__v8hi, __v8hi)
13001 __v8hi __builtin_arc_vdifw (__v8hi, __v8hi)
13002 __v8hi __builtin_arc_veqw (__v8hi, __v8hi)
13003 __v8hi __builtin_arc_vh264f (__v8hi, __v8hi)
13004 __v8hi __builtin_arc_vh264ft (__v8hi, __v8hi)
13005 __v8hi __builtin_arc_vh264fw (__v8hi, __v8hi)
13006 __v8hi __builtin_arc_vlew (__v8hi, __v8hi)
13007 __v8hi __builtin_arc_vltw (__v8hi, __v8hi)
13008 __v8hi __builtin_arc_vmaxaw (__v8hi, __v8hi)
13009 __v8hi __builtin_arc_vmaxw (__v8hi, __v8hi)
13010 __v8hi __builtin_arc_vminaw (__v8hi, __v8hi)
13011 __v8hi __builtin_arc_vminw (__v8hi, __v8hi)
13012 __v8hi __builtin_arc_vmr1aw (__v8hi, __v8hi)
13013 __v8hi __builtin_arc_vmr1w (__v8hi, __v8hi)
13014 __v8hi __builtin_arc_vmr2aw (__v8hi, __v8hi)
13015 __v8hi __builtin_arc_vmr2w (__v8hi, __v8hi)
13016 __v8hi __builtin_arc_vmr3aw (__v8hi, __v8hi)
13017 __v8hi __builtin_arc_vmr3w (__v8hi, __v8hi)
13018 __v8hi __builtin_arc_vmr4aw (__v8hi, __v8hi)
13019 __v8hi __builtin_arc_vmr4w (__v8hi, __v8hi)
13020 __v8hi __builtin_arc_vmr5aw (__v8hi, __v8hi)
13021 __v8hi __builtin_arc_vmr5w (__v8hi, __v8hi)
13022 __v8hi __builtin_arc_vmr6aw (__v8hi, __v8hi)
13023 __v8hi __builtin_arc_vmr6w (__v8hi, __v8hi)
13024 __v8hi __builtin_arc_vmr7aw (__v8hi, __v8hi)
13025 __v8hi __builtin_arc_vmr7w (__v8hi, __v8hi)
13026 __v8hi __builtin_arc_vmrb (__v8hi, __v8hi)
13027 __v8hi __builtin_arc_vmulaw (__v8hi, __v8hi)
13028 __v8hi __builtin_arc_vmulfaw (__v8hi, __v8hi)
13029 __v8hi __builtin_arc_vmulfw (__v8hi, __v8hi)
13030 __v8hi __builtin_arc_vmulw (__v8hi, __v8hi)
13031 __v8hi __builtin_arc_vnew (__v8hi, __v8hi)
13032 __v8hi __builtin_arc_vor (__v8hi, __v8hi)
13033 __v8hi __builtin_arc_vsubaw (__v8hi, __v8hi)
13034 __v8hi __builtin_arc_vsubw (__v8hi, __v8hi)
13035 __v8hi __builtin_arc_vsummw (__v8hi, __v8hi)
13036 __v8hi __builtin_arc_vvc1f (__v8hi, __v8hi)
13037 __v8hi __builtin_arc_vvc1ft (__v8hi, __v8hi)
13038 __v8hi __builtin_arc_vxor (__v8hi, __v8hi)
13039 __v8hi __builtin_arc_vxoraw (__v8hi, __v8hi)
13042 The following take one @code{__v8hi} and one @code{int} argument and return a
13043 @code{__v8hi} result:
13046 __v8hi __builtin_arc_vbaddw (__v8hi, int)
13047 __v8hi __builtin_arc_vbmaxw (__v8hi, int)
13048 __v8hi __builtin_arc_vbminw (__v8hi, int)
13049 __v8hi __builtin_arc_vbmulaw (__v8hi, int)
13050 __v8hi __builtin_arc_vbmulfw (__v8hi, int)
13051 __v8hi __builtin_arc_vbmulw (__v8hi, int)
13052 __v8hi __builtin_arc_vbrsubw (__v8hi, int)
13053 __v8hi __builtin_arc_vbsubw (__v8hi, int)
13056 The following take one @code{__v8hi} argument and one @code{int} argument which
13057 must be a 3-bit compile time constant indicating a register number
13058 I0-I7. They return a @code{__v8hi} result.
13060 __v8hi __builtin_arc_vasrw (__v8hi, const int)
13061 __v8hi __builtin_arc_vsr8 (__v8hi, const int)
13062 __v8hi __builtin_arc_vsr8aw (__v8hi, const int)
13065 The following take one @code{__v8hi} argument and one @code{int}
13066 argument which must be a 6-bit compile time constant. They return a
13067 @code{__v8hi} result.
13069 __v8hi __builtin_arc_vasrpwbi (__v8hi, const int)
13070 __v8hi __builtin_arc_vasrrpwbi (__v8hi, const int)
13071 __v8hi __builtin_arc_vasrrwi (__v8hi, const int)
13072 __v8hi __builtin_arc_vasrsrwi (__v8hi, const int)
13073 __v8hi __builtin_arc_vasrwi (__v8hi, const int)
13074 __v8hi __builtin_arc_vsr8awi (__v8hi, const int)
13075 __v8hi __builtin_arc_vsr8i (__v8hi, const int)
13078 The following take one @code{__v8hi} argument and one @code{int} argument which
13079 must be a 8-bit compile time constant. They return a @code{__v8hi}
13082 __v8hi __builtin_arc_vd6tapf (__v8hi, const int)
13083 __v8hi __builtin_arc_vmvaw (__v8hi, const int)
13084 __v8hi __builtin_arc_vmvw (__v8hi, const int)
13085 __v8hi __builtin_arc_vmvzw (__v8hi, const int)
13088 The following take two @code{int} arguments, the second of which which
13089 must be a 8-bit compile time constant. They return a @code{__v8hi}
13092 __v8hi __builtin_arc_vmovaw (int, const int)
13093 __v8hi __builtin_arc_vmovw (int, const int)
13094 __v8hi __builtin_arc_vmovzw (int, const int)
13097 The following take a single @code{__v8hi} argument and return a
13098 @code{__v8hi} result:
13100 __v8hi __builtin_arc_vabsaw (__v8hi)
13101 __v8hi __builtin_arc_vabsw (__v8hi)
13102 __v8hi __builtin_arc_vaddsuw (__v8hi)
13103 __v8hi __builtin_arc_vexch1 (__v8hi)
13104 __v8hi __builtin_arc_vexch2 (__v8hi)
13105 __v8hi __builtin_arc_vexch4 (__v8hi)
13106 __v8hi __builtin_arc_vsignw (__v8hi)
13107 __v8hi __builtin_arc_vupbaw (__v8hi)
13108 __v8hi __builtin_arc_vupbw (__v8hi)
13109 __v8hi __builtin_arc_vupsbaw (__v8hi)
13110 __v8hi __builtin_arc_vupsbw (__v8hi)
13113 The following take two @code{int} arguments and return no result:
13115 void __builtin_arc_vdirun (int, int)
13116 void __builtin_arc_vdorun (int, int)
13119 The following take two @code{int} arguments and return no result. The
13120 first argument must a 3-bit compile time constant indicating one of
13121 the DR0-DR7 DMA setup channels:
13123 void __builtin_arc_vdiwr (const int, int)
13124 void __builtin_arc_vdowr (const int, int)
13127 The following take an @code{int} argument and return no result:
13129 void __builtin_arc_vendrec (int)
13130 void __builtin_arc_vrec (int)
13131 void __builtin_arc_vrecrun (int)
13132 void __builtin_arc_vrun (int)
13135 The following take a @code{__v8hi} argument and two @code{int}
13136 arguments and return a @code{__v8hi} result. The second argument must
13137 be a 3-bit compile time constants, indicating one the registers I0-I7,
13138 and the third argument must be an 8-bit compile time constant.
13140 @emph{Note:} Although the equivalent hardware instructions do not take
13141 an SIMD register as an operand, these builtins overwrite the relevant
13142 bits of the @code{__v8hi} register provided as the first argument with
13143 the value loaded from the @code{[Ib, u8]} location in the SDM.
13146 __v8hi __builtin_arc_vld32 (__v8hi, const int, const int)
13147 __v8hi __builtin_arc_vld32wh (__v8hi, const int, const int)
13148 __v8hi __builtin_arc_vld32wl (__v8hi, const int, const int)
13149 __v8hi __builtin_arc_vld64 (__v8hi, const int, const int)
13152 The following take two @code{int} arguments and return a @code{__v8hi}
13153 result. The first argument must be a 3-bit compile time constants,
13154 indicating one the registers I0-I7, and the second argument must be an
13155 8-bit compile time constant.
13158 __v8hi __builtin_arc_vld128 (const int, const int)
13159 __v8hi __builtin_arc_vld64w (const int, const int)
13162 The following take a @code{__v8hi} argument and two @code{int}
13163 arguments and return no result. The second argument must be a 3-bit
13164 compile time constants, indicating one the registers I0-I7, and the
13165 third argument must be an 8-bit compile time constant.
13168 void __builtin_arc_vst128 (__v8hi, const int, const int)
13169 void __builtin_arc_vst64 (__v8hi, const int, const int)
13172 The following take a @code{__v8hi} argument and three @code{int}
13173 arguments and return no result. The second argument must be a 3-bit
13174 compile-time constant, identifying the 16-bit sub-register to be
13175 stored, the third argument must be a 3-bit compile time constants,
13176 indicating one the registers I0-I7, and the fourth argument must be an
13177 8-bit compile time constant.
13180 void __builtin_arc_vst16_n (__v8hi, const int, const int, const int)
13181 void __builtin_arc_vst32_n (__v8hi, const int, const int, const int)
13184 @node ARM iWMMXt Built-in Functions
13185 @subsection ARM iWMMXt Built-in Functions
13187 These built-in functions are available for the ARM family of
13188 processors when the @option{-mcpu=iwmmxt} switch is used:
13191 typedef int v2si __attribute__ ((vector_size (8)));
13192 typedef short v4hi __attribute__ ((vector_size (8)));
13193 typedef char v8qi __attribute__ ((vector_size (8)));
13195 int __builtin_arm_getwcgr0 (void)
13196 void __builtin_arm_setwcgr0 (int)
13197 int __builtin_arm_getwcgr1 (void)
13198 void __builtin_arm_setwcgr1 (int)
13199 int __builtin_arm_getwcgr2 (void)
13200 void __builtin_arm_setwcgr2 (int)
13201 int __builtin_arm_getwcgr3 (void)
13202 void __builtin_arm_setwcgr3 (int)
13203 int __builtin_arm_textrmsb (v8qi, int)
13204 int __builtin_arm_textrmsh (v4hi, int)
13205 int __builtin_arm_textrmsw (v2si, int)
13206 int __builtin_arm_textrmub (v8qi, int)
13207 int __builtin_arm_textrmuh (v4hi, int)
13208 int __builtin_arm_textrmuw (v2si, int)
13209 v8qi __builtin_arm_tinsrb (v8qi, int, int)
13210 v4hi __builtin_arm_tinsrh (v4hi, int, int)
13211 v2si __builtin_arm_tinsrw (v2si, int, int)
13212 long long __builtin_arm_tmia (long long, int, int)
13213 long long __builtin_arm_tmiabb (long long, int, int)
13214 long long __builtin_arm_tmiabt (long long, int, int)
13215 long long __builtin_arm_tmiaph (long long, int, int)
13216 long long __builtin_arm_tmiatb (long long, int, int)
13217 long long __builtin_arm_tmiatt (long long, int, int)
13218 int __builtin_arm_tmovmskb (v8qi)
13219 int __builtin_arm_tmovmskh (v4hi)
13220 int __builtin_arm_tmovmskw (v2si)
13221 long long __builtin_arm_waccb (v8qi)
13222 long long __builtin_arm_wacch (v4hi)
13223 long long __builtin_arm_waccw (v2si)
13224 v8qi __builtin_arm_waddb (v8qi, v8qi)
13225 v8qi __builtin_arm_waddbss (v8qi, v8qi)
13226 v8qi __builtin_arm_waddbus (v8qi, v8qi)
13227 v4hi __builtin_arm_waddh (v4hi, v4hi)
13228 v4hi __builtin_arm_waddhss (v4hi, v4hi)
13229 v4hi __builtin_arm_waddhus (v4hi, v4hi)
13230 v2si __builtin_arm_waddw (v2si, v2si)
13231 v2si __builtin_arm_waddwss (v2si, v2si)
13232 v2si __builtin_arm_waddwus (v2si, v2si)
13233 v8qi __builtin_arm_walign (v8qi, v8qi, int)
13234 long long __builtin_arm_wand(long long, long long)
13235 long long __builtin_arm_wandn (long long, long long)
13236 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
13237 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
13238 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
13239 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
13240 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
13241 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
13242 v2si __builtin_arm_wcmpeqw (v2si, v2si)
13243 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
13244 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
13245 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
13246 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
13247 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
13248 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
13249 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
13250 long long __builtin_arm_wmacsz (v4hi, v4hi)
13251 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
13252 long long __builtin_arm_wmacuz (v4hi, v4hi)
13253 v4hi __builtin_arm_wmadds (v4hi, v4hi)
13254 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
13255 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
13256 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
13257 v2si __builtin_arm_wmaxsw (v2si, v2si)
13258 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
13259 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
13260 v2si __builtin_arm_wmaxuw (v2si, v2si)
13261 v8qi __builtin_arm_wminsb (v8qi, v8qi)
13262 v4hi __builtin_arm_wminsh (v4hi, v4hi)
13263 v2si __builtin_arm_wminsw (v2si, v2si)
13264 v8qi __builtin_arm_wminub (v8qi, v8qi)
13265 v4hi __builtin_arm_wminuh (v4hi, v4hi)
13266 v2si __builtin_arm_wminuw (v2si, v2si)
13267 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
13268 v4hi __builtin_arm_wmulul (v4hi, v4hi)
13269 v4hi __builtin_arm_wmulum (v4hi, v4hi)
13270 long long __builtin_arm_wor (long long, long long)
13271 v2si __builtin_arm_wpackdss (long long, long long)
13272 v2si __builtin_arm_wpackdus (long long, long long)
13273 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
13274 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
13275 v4hi __builtin_arm_wpackwss (v2si, v2si)
13276 v4hi __builtin_arm_wpackwus (v2si, v2si)
13277 long long __builtin_arm_wrord (long long, long long)
13278 long long __builtin_arm_wrordi (long long, int)
13279 v4hi __builtin_arm_wrorh (v4hi, long long)
13280 v4hi __builtin_arm_wrorhi (v4hi, int)
13281 v2si __builtin_arm_wrorw (v2si, long long)
13282 v2si __builtin_arm_wrorwi (v2si, int)
13283 v2si __builtin_arm_wsadb (v2si, v8qi, v8qi)
13284 v2si __builtin_arm_wsadbz (v8qi, v8qi)
13285 v2si __builtin_arm_wsadh (v2si, v4hi, v4hi)
13286 v2si __builtin_arm_wsadhz (v4hi, v4hi)
13287 v4hi __builtin_arm_wshufh (v4hi, int)
13288 long long __builtin_arm_wslld (long long, long long)
13289 long long __builtin_arm_wslldi (long long, int)
13290 v4hi __builtin_arm_wsllh (v4hi, long long)
13291 v4hi __builtin_arm_wsllhi (v4hi, int)
13292 v2si __builtin_arm_wsllw (v2si, long long)
13293 v2si __builtin_arm_wsllwi (v2si, int)
13294 long long __builtin_arm_wsrad (long long, long long)
13295 long long __builtin_arm_wsradi (long long, int)
13296 v4hi __builtin_arm_wsrah (v4hi, long long)
13297 v4hi __builtin_arm_wsrahi (v4hi, int)
13298 v2si __builtin_arm_wsraw (v2si, long long)
13299 v2si __builtin_arm_wsrawi (v2si, int)
13300 long long __builtin_arm_wsrld (long long, long long)
13301 long long __builtin_arm_wsrldi (long long, int)
13302 v4hi __builtin_arm_wsrlh (v4hi, long long)
13303 v4hi __builtin_arm_wsrlhi (v4hi, int)
13304 v2si __builtin_arm_wsrlw (v2si, long long)
13305 v2si __builtin_arm_wsrlwi (v2si, int)
13306 v8qi __builtin_arm_wsubb (v8qi, v8qi)
13307 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
13308 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
13309 v4hi __builtin_arm_wsubh (v4hi, v4hi)
13310 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
13311 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
13312 v2si __builtin_arm_wsubw (v2si, v2si)
13313 v2si __builtin_arm_wsubwss (v2si, v2si)
13314 v2si __builtin_arm_wsubwus (v2si, v2si)
13315 v4hi __builtin_arm_wunpckehsb (v8qi)
13316 v2si __builtin_arm_wunpckehsh (v4hi)
13317 long long __builtin_arm_wunpckehsw (v2si)
13318 v4hi __builtin_arm_wunpckehub (v8qi)
13319 v2si __builtin_arm_wunpckehuh (v4hi)
13320 long long __builtin_arm_wunpckehuw (v2si)
13321 v4hi __builtin_arm_wunpckelsb (v8qi)
13322 v2si __builtin_arm_wunpckelsh (v4hi)
13323 long long __builtin_arm_wunpckelsw (v2si)
13324 v4hi __builtin_arm_wunpckelub (v8qi)
13325 v2si __builtin_arm_wunpckeluh (v4hi)
13326 long long __builtin_arm_wunpckeluw (v2si)
13327 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
13328 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
13329 v2si __builtin_arm_wunpckihw (v2si, v2si)
13330 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
13331 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
13332 v2si __builtin_arm_wunpckilw (v2si, v2si)
13333 long long __builtin_arm_wxor (long long, long long)
13334 long long __builtin_arm_wzero ()
13338 @node ARM C Language Extensions (ACLE)
13339 @subsection ARM C Language Extensions (ACLE)
13341 GCC implements extensions for C as described in the ARM C Language
13342 Extensions (ACLE) specification, which can be found at
13343 @uref{http://infocenter.arm.com/help/topic/com.arm.doc.ihi0053c/IHI0053C_acle_2_0.pdf}.
13345 As a part of ACLE, GCC implements extensions for Advanced SIMD as described in
13346 the ARM C Language Extensions Specification. The complete list of Advanced SIMD
13347 intrinsics can be found at
13348 @uref{http://infocenter.arm.com/help/topic/com.arm.doc.ihi0073a/IHI0073A_arm_neon_intrinsics_ref.pdf}.
13349 The built-in intrinsics for the Advanced SIMD extension are available when
13352 Currently, ARM and AArch64 back ends do not support ACLE 2.0 fully. Both
13353 back ends support CRC32 intrinsics and the ARM back end supports the
13354 Coprocessor intrinsics, all from @file{arm_acle.h}. The ARM back end's 16-bit
13355 floating-point Advanced SIMD intrinsics currently comply to ACLE v1.1.
13356 AArch64's back end does not have support for 16-bit floating point Advanced SIMD
13359 See @ref{ARM Options} and @ref{AArch64 Options} for more information on the
13360 availability of extensions.
13362 @node ARM Floating Point Status and Control Intrinsics
13363 @subsection ARM Floating Point Status and Control Intrinsics
13365 These built-in functions are available for the ARM family of
13366 processors with floating-point unit.
13369 unsigned int __builtin_arm_get_fpscr ()
13370 void __builtin_arm_set_fpscr (unsigned int)
13373 @node ARM ARMv8-M Security Extensions
13374 @subsection ARM ARMv8-M Security Extensions
13376 GCC implements the ARMv8-M Security Extensions as described in the ARMv8-M
13377 Security Extensions: Requirements on Development Tools Engineering
13378 Specification, which can be found at
13379 @uref{http://infocenter.arm.com/help/topic/com.arm.doc.ecm0359818/ECM0359818_armv8m_security_extensions_reqs_on_dev_tools_1_0.pdf}.
13381 As part of the Security Extensions GCC implements two new function attributes:
13382 @code{cmse_nonsecure_entry} and @code{cmse_nonsecure_call}.
13384 As part of the Security Extensions GCC implements the intrinsics below. FPTR
13385 is used here to mean any function pointer type.
13388 cmse_address_info_t cmse_TT (void *)
13389 cmse_address_info_t cmse_TT_fptr (FPTR)
13390 cmse_address_info_t cmse_TTT (void *)
13391 cmse_address_info_t cmse_TTT_fptr (FPTR)
13392 cmse_address_info_t cmse_TTA (void *)
13393 cmse_address_info_t cmse_TTA_fptr (FPTR)
13394 cmse_address_info_t cmse_TTAT (void *)
13395 cmse_address_info_t cmse_TTAT_fptr (FPTR)
13396 void * cmse_check_address_range (void *, size_t, int)
13397 typeof(p) cmse_nsfptr_create (FPTR p)
13398 intptr_t cmse_is_nsfptr (FPTR)
13399 int cmse_nonsecure_caller (void)
13402 @node AVR Built-in Functions
13403 @subsection AVR Built-in Functions
13405 For each built-in function for AVR, there is an equally named,
13406 uppercase built-in macro defined. That way users can easily query if
13407 or if not a specific built-in is implemented or not. For example, if
13408 @code{__builtin_avr_nop} is available the macro
13409 @code{__BUILTIN_AVR_NOP} is defined to @code{1} and undefined otherwise.
13413 @item void __builtin_avr_nop (void)
13414 @itemx void __builtin_avr_sei (void)
13415 @itemx void __builtin_avr_cli (void)
13416 @itemx void __builtin_avr_sleep (void)
13417 @itemx void __builtin_avr_wdr (void)
13418 @itemx unsigned char __builtin_avr_swap (unsigned char)
13419 @itemx unsigned int __builtin_avr_fmul (unsigned char, unsigned char)
13420 @itemx int __builtin_avr_fmuls (char, char)
13421 @itemx int __builtin_avr_fmulsu (char, unsigned char)
13422 These built-in functions map to the respective machine
13423 instruction, i.e.@: @code{nop}, @code{sei}, @code{cli}, @code{sleep},
13424 @code{wdr}, @code{swap}, @code{fmul}, @code{fmuls}
13425 resp. @code{fmulsu}. The three @code{fmul*} built-ins are implemented
13426 as library call if no hardware multiplier is available.
13428 @item void __builtin_avr_delay_cycles (unsigned long ticks)
13429 Delay execution for @var{ticks} cycles. Note that this
13430 built-in does not take into account the effect of interrupts that
13431 might increase delay time. @var{ticks} must be a compile-time
13432 integer constant; delays with a variable number of cycles are not supported.
13434 @item char __builtin_avr_flash_segment (const __memx void*)
13435 This built-in takes a byte address to the 24-bit
13436 @ref{AVR Named Address Spaces,address space} @code{__memx} and returns
13437 the number of the flash segment (the 64 KiB chunk) where the address
13438 points to. Counting starts at @code{0}.
13439 If the address does not point to flash memory, return @code{-1}.
13441 @item uint8_t __builtin_avr_insert_bits (uint32_t map, uint8_t bits, uint8_t val)
13442 Insert bits from @var{bits} into @var{val} and return the resulting
13443 value. The nibbles of @var{map} determine how the insertion is
13444 performed: Let @var{X} be the @var{n}-th nibble of @var{map}
13446 @item If @var{X} is @code{0xf},
13447 then the @var{n}-th bit of @var{val} is returned unaltered.
13449 @item If X is in the range 0@dots{}7,
13450 then the @var{n}-th result bit is set to the @var{X}-th bit of @var{bits}
13452 @item If X is in the range 8@dots{}@code{0xe},
13453 then the @var{n}-th result bit is undefined.
13457 One typical use case for this built-in is adjusting input and
13458 output values to non-contiguous port layouts. Some examples:
13461 // same as val, bits is unused
13462 __builtin_avr_insert_bits (0xffffffff, bits, val)
13466 // same as bits, val is unused
13467 __builtin_avr_insert_bits (0x76543210, bits, val)
13471 // same as rotating bits by 4
13472 __builtin_avr_insert_bits (0x32107654, bits, 0)
13476 // high nibble of result is the high nibble of val
13477 // low nibble of result is the low nibble of bits
13478 __builtin_avr_insert_bits (0xffff3210, bits, val)
13482 // reverse the bit order of bits
13483 __builtin_avr_insert_bits (0x01234567, bits, 0)
13486 @item void __builtin_avr_nops (unsigned count)
13487 Insert @var{count} @code{NOP} instructions.
13488 The number of instructions must be a compile-time integer constant.
13493 There are many more AVR-specific built-in functions that are used to
13494 implement the ISO/IEC TR 18037 ``Embedded C'' fixed-point functions of
13495 section 7.18a.6. You don't need to use these built-ins directly.
13496 Instead, use the declarations as supplied by the @code{stdfix.h} header
13500 #include <stdfix.h>
13502 // Re-interpret the bit representation of unsigned 16-bit
13503 // integer @var{uval} as Q-format 0.16 value.
13504 unsigned fract get_bits (uint_ur_t uval)
13506 return urbits (uval);
13510 @node Blackfin Built-in Functions
13511 @subsection Blackfin Built-in Functions
13513 Currently, there are two Blackfin-specific built-in functions. These are
13514 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
13515 using inline assembly; by using these built-in functions the compiler can
13516 automatically add workarounds for hardware errata involving these
13517 instructions. These functions are named as follows:
13520 void __builtin_bfin_csync (void)
13521 void __builtin_bfin_ssync (void)
13524 @node FR-V Built-in Functions
13525 @subsection FR-V Built-in Functions
13527 GCC provides many FR-V-specific built-in functions. In general,
13528 these functions are intended to be compatible with those described
13529 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
13530 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
13531 @code{__MBTOHE}, the GCC forms of which pass 128-bit values by
13532 pointer rather than by value.
13534 Most of the functions are named after specific FR-V instructions.
13535 Such functions are said to be ``directly mapped'' and are summarized
13536 here in tabular form.
13540 * Directly-mapped Integer Functions::
13541 * Directly-mapped Media Functions::
13542 * Raw read/write Functions::
13543 * Other Built-in Functions::
13546 @node Argument Types
13547 @subsubsection Argument Types
13549 The arguments to the built-in functions can be divided into three groups:
13550 register numbers, compile-time constants and run-time values. In order
13551 to make this classification clear at a glance, the arguments and return
13552 values are given the following pseudo types:
13554 @multitable @columnfractions .20 .30 .15 .35
13555 @item Pseudo type @tab Real C type @tab Constant? @tab Description
13556 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
13557 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
13558 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
13559 @item @code{uw2} @tab @code{unsigned long long} @tab No
13560 @tab an unsigned doubleword
13561 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
13562 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
13563 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
13564 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
13567 These pseudo types are not defined by GCC, they are simply a notational
13568 convenience used in this manual.
13570 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
13571 and @code{sw2} are evaluated at run time. They correspond to
13572 register operands in the underlying FR-V instructions.
13574 @code{const} arguments represent immediate operands in the underlying
13575 FR-V instructions. They must be compile-time constants.
13577 @code{acc} arguments are evaluated at compile time and specify the number
13578 of an accumulator register. For example, an @code{acc} argument of 2
13579 selects the ACC2 register.
13581 @code{iacc} arguments are similar to @code{acc} arguments but specify the
13582 number of an IACC register. See @pxref{Other Built-in Functions}
13585 @node Directly-mapped Integer Functions
13586 @subsubsection Directly-Mapped Integer Functions
13588 The functions listed below map directly to FR-V I-type instructions.
13590 @multitable @columnfractions .45 .32 .23
13591 @item Function prototype @tab Example usage @tab Assembly output
13592 @item @code{sw1 __ADDSS (sw1, sw1)}
13593 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
13594 @tab @code{ADDSS @var{a},@var{b},@var{c}}
13595 @item @code{sw1 __SCAN (sw1, sw1)}
13596 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
13597 @tab @code{SCAN @var{a},@var{b},@var{c}}
13598 @item @code{sw1 __SCUTSS (sw1)}
13599 @tab @code{@var{b} = __SCUTSS (@var{a})}
13600 @tab @code{SCUTSS @var{a},@var{b}}
13601 @item @code{sw1 __SLASS (sw1, sw1)}
13602 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
13603 @tab @code{SLASS @var{a},@var{b},@var{c}}
13604 @item @code{void __SMASS (sw1, sw1)}
13605 @tab @code{__SMASS (@var{a}, @var{b})}
13606 @tab @code{SMASS @var{a},@var{b}}
13607 @item @code{void __SMSSS (sw1, sw1)}
13608 @tab @code{__SMSSS (@var{a}, @var{b})}
13609 @tab @code{SMSSS @var{a},@var{b}}
13610 @item @code{void __SMU (sw1, sw1)}
13611 @tab @code{__SMU (@var{a}, @var{b})}
13612 @tab @code{SMU @var{a},@var{b}}
13613 @item @code{sw2 __SMUL (sw1, sw1)}
13614 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
13615 @tab @code{SMUL @var{a},@var{b},@var{c}}
13616 @item @code{sw1 __SUBSS (sw1, sw1)}
13617 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
13618 @tab @code{SUBSS @var{a},@var{b},@var{c}}
13619 @item @code{uw2 __UMUL (uw1, uw1)}
13620 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
13621 @tab @code{UMUL @var{a},@var{b},@var{c}}
13624 @node Directly-mapped Media Functions
13625 @subsubsection Directly-Mapped Media Functions
13627 The functions listed below map directly to FR-V M-type instructions.
13629 @multitable @columnfractions .45 .32 .23
13630 @item Function prototype @tab Example usage @tab Assembly output
13631 @item @code{uw1 __MABSHS (sw1)}
13632 @tab @code{@var{b} = __MABSHS (@var{a})}
13633 @tab @code{MABSHS @var{a},@var{b}}
13634 @item @code{void __MADDACCS (acc, acc)}
13635 @tab @code{__MADDACCS (@var{b}, @var{a})}
13636 @tab @code{MADDACCS @var{a},@var{b}}
13637 @item @code{sw1 __MADDHSS (sw1, sw1)}
13638 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
13639 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
13640 @item @code{uw1 __MADDHUS (uw1, uw1)}
13641 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
13642 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
13643 @item @code{uw1 __MAND (uw1, uw1)}
13644 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
13645 @tab @code{MAND @var{a},@var{b},@var{c}}
13646 @item @code{void __MASACCS (acc, acc)}
13647 @tab @code{__MASACCS (@var{b}, @var{a})}
13648 @tab @code{MASACCS @var{a},@var{b}}
13649 @item @code{uw1 __MAVEH (uw1, uw1)}
13650 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
13651 @tab @code{MAVEH @var{a},@var{b},@var{c}}
13652 @item @code{uw2 __MBTOH (uw1)}
13653 @tab @code{@var{b} = __MBTOH (@var{a})}
13654 @tab @code{MBTOH @var{a},@var{b}}
13655 @item @code{void __MBTOHE (uw1 *, uw1)}
13656 @tab @code{__MBTOHE (&@var{b}, @var{a})}
13657 @tab @code{MBTOHE @var{a},@var{b}}
13658 @item @code{void __MCLRACC (acc)}
13659 @tab @code{__MCLRACC (@var{a})}
13660 @tab @code{MCLRACC @var{a}}
13661 @item @code{void __MCLRACCA (void)}
13662 @tab @code{__MCLRACCA ()}
13663 @tab @code{MCLRACCA}
13664 @item @code{uw1 __Mcop1 (uw1, uw1)}
13665 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
13666 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
13667 @item @code{uw1 __Mcop2 (uw1, uw1)}
13668 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
13669 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
13670 @item @code{uw1 __MCPLHI (uw2, const)}
13671 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
13672 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
13673 @item @code{uw1 __MCPLI (uw2, const)}
13674 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
13675 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
13676 @item @code{void __MCPXIS (acc, sw1, sw1)}
13677 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
13678 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
13679 @item @code{void __MCPXIU (acc, uw1, uw1)}
13680 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
13681 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
13682 @item @code{void __MCPXRS (acc, sw1, sw1)}
13683 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
13684 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
13685 @item @code{void __MCPXRU (acc, uw1, uw1)}
13686 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
13687 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
13688 @item @code{uw1 __MCUT (acc, uw1)}
13689 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
13690 @tab @code{MCUT @var{a},@var{b},@var{c}}
13691 @item @code{uw1 __MCUTSS (acc, sw1)}
13692 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
13693 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
13694 @item @code{void __MDADDACCS (acc, acc)}
13695 @tab @code{__MDADDACCS (@var{b}, @var{a})}
13696 @tab @code{MDADDACCS @var{a},@var{b}}
13697 @item @code{void __MDASACCS (acc, acc)}
13698 @tab @code{__MDASACCS (@var{b}, @var{a})}
13699 @tab @code{MDASACCS @var{a},@var{b}}
13700 @item @code{uw2 __MDCUTSSI (acc, const)}
13701 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
13702 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
13703 @item @code{uw2 __MDPACKH (uw2, uw2)}
13704 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
13705 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
13706 @item @code{uw2 __MDROTLI (uw2, const)}
13707 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
13708 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
13709 @item @code{void __MDSUBACCS (acc, acc)}
13710 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
13711 @tab @code{MDSUBACCS @var{a},@var{b}}
13712 @item @code{void __MDUNPACKH (uw1 *, uw2)}
13713 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
13714 @tab @code{MDUNPACKH @var{a},@var{b}}
13715 @item @code{uw2 __MEXPDHD (uw1, const)}
13716 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
13717 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
13718 @item @code{uw1 __MEXPDHW (uw1, const)}
13719 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
13720 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
13721 @item @code{uw1 __MHDSETH (uw1, const)}
13722 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
13723 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
13724 @item @code{sw1 __MHDSETS (const)}
13725 @tab @code{@var{b} = __MHDSETS (@var{a})}
13726 @tab @code{MHDSETS #@var{a},@var{b}}
13727 @item @code{uw1 __MHSETHIH (uw1, const)}
13728 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
13729 @tab @code{MHSETHIH #@var{a},@var{b}}
13730 @item @code{sw1 __MHSETHIS (sw1, const)}
13731 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
13732 @tab @code{MHSETHIS #@var{a},@var{b}}
13733 @item @code{uw1 __MHSETLOH (uw1, const)}
13734 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
13735 @tab @code{MHSETLOH #@var{a},@var{b}}
13736 @item @code{sw1 __MHSETLOS (sw1, const)}
13737 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
13738 @tab @code{MHSETLOS #@var{a},@var{b}}
13739 @item @code{uw1 __MHTOB (uw2)}
13740 @tab @code{@var{b} = __MHTOB (@var{a})}
13741 @tab @code{MHTOB @var{a},@var{b}}
13742 @item @code{void __MMACHS (acc, sw1, sw1)}
13743 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
13744 @tab @code{MMACHS @var{a},@var{b},@var{c}}
13745 @item @code{void __MMACHU (acc, uw1, uw1)}
13746 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
13747 @tab @code{MMACHU @var{a},@var{b},@var{c}}
13748 @item @code{void __MMRDHS (acc, sw1, sw1)}
13749 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
13750 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
13751 @item @code{void __MMRDHU (acc, uw1, uw1)}
13752 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
13753 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
13754 @item @code{void __MMULHS (acc, sw1, sw1)}
13755 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
13756 @tab @code{MMULHS @var{a},@var{b},@var{c}}
13757 @item @code{void __MMULHU (acc, uw1, uw1)}
13758 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
13759 @tab @code{MMULHU @var{a},@var{b},@var{c}}
13760 @item @code{void __MMULXHS (acc, sw1, sw1)}
13761 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
13762 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
13763 @item @code{void __MMULXHU (acc, uw1, uw1)}
13764 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
13765 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
13766 @item @code{uw1 __MNOT (uw1)}
13767 @tab @code{@var{b} = __MNOT (@var{a})}
13768 @tab @code{MNOT @var{a},@var{b}}
13769 @item @code{uw1 __MOR (uw1, uw1)}
13770 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
13771 @tab @code{MOR @var{a},@var{b},@var{c}}
13772 @item @code{uw1 __MPACKH (uh, uh)}
13773 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
13774 @tab @code{MPACKH @var{a},@var{b},@var{c}}
13775 @item @code{sw2 __MQADDHSS (sw2, sw2)}
13776 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
13777 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
13778 @item @code{uw2 __MQADDHUS (uw2, uw2)}
13779 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
13780 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
13781 @item @code{void __MQCPXIS (acc, sw2, sw2)}
13782 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
13783 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
13784 @item @code{void __MQCPXIU (acc, uw2, uw2)}
13785 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
13786 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
13787 @item @code{void __MQCPXRS (acc, sw2, sw2)}
13788 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
13789 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
13790 @item @code{void __MQCPXRU (acc, uw2, uw2)}
13791 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
13792 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
13793 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
13794 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
13795 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
13796 @item @code{sw2 __MQLMTHS (sw2, sw2)}
13797 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
13798 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
13799 @item @code{void __MQMACHS (acc, sw2, sw2)}
13800 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
13801 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
13802 @item @code{void __MQMACHU (acc, uw2, uw2)}
13803 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
13804 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
13805 @item @code{void __MQMACXHS (acc, sw2, sw2)}
13806 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
13807 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
13808 @item @code{void __MQMULHS (acc, sw2, sw2)}
13809 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
13810 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
13811 @item @code{void __MQMULHU (acc, uw2, uw2)}
13812 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
13813 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
13814 @item @code{void __MQMULXHS (acc, sw2, sw2)}
13815 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
13816 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
13817 @item @code{void __MQMULXHU (acc, uw2, uw2)}
13818 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
13819 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
13820 @item @code{sw2 __MQSATHS (sw2, sw2)}
13821 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
13822 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
13823 @item @code{uw2 __MQSLLHI (uw2, int)}
13824 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
13825 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
13826 @item @code{sw2 __MQSRAHI (sw2, int)}
13827 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
13828 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
13829 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
13830 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
13831 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
13832 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
13833 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
13834 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
13835 @item @code{void __MQXMACHS (acc, sw2, sw2)}
13836 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
13837 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
13838 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
13839 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
13840 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
13841 @item @code{uw1 __MRDACC (acc)}
13842 @tab @code{@var{b} = __MRDACC (@var{a})}
13843 @tab @code{MRDACC @var{a},@var{b}}
13844 @item @code{uw1 __MRDACCG (acc)}
13845 @tab @code{@var{b} = __MRDACCG (@var{a})}
13846 @tab @code{MRDACCG @var{a},@var{b}}
13847 @item @code{uw1 __MROTLI (uw1, const)}
13848 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
13849 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
13850 @item @code{uw1 __MROTRI (uw1, const)}
13851 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
13852 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
13853 @item @code{sw1 __MSATHS (sw1, sw1)}
13854 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
13855 @tab @code{MSATHS @var{a},@var{b},@var{c}}
13856 @item @code{uw1 __MSATHU (uw1, uw1)}
13857 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
13858 @tab @code{MSATHU @var{a},@var{b},@var{c}}
13859 @item @code{uw1 __MSLLHI (uw1, const)}
13860 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
13861 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
13862 @item @code{sw1 __MSRAHI (sw1, const)}
13863 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
13864 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
13865 @item @code{uw1 __MSRLHI (uw1, const)}
13866 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
13867 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
13868 @item @code{void __MSUBACCS (acc, acc)}
13869 @tab @code{__MSUBACCS (@var{b}, @var{a})}
13870 @tab @code{MSUBACCS @var{a},@var{b}}
13871 @item @code{sw1 __MSUBHSS (sw1, sw1)}
13872 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
13873 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
13874 @item @code{uw1 __MSUBHUS (uw1, uw1)}
13875 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
13876 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
13877 @item @code{void __MTRAP (void)}
13878 @tab @code{__MTRAP ()}
13880 @item @code{uw2 __MUNPACKH (uw1)}
13881 @tab @code{@var{b} = __MUNPACKH (@var{a})}
13882 @tab @code{MUNPACKH @var{a},@var{b}}
13883 @item @code{uw1 __MWCUT (uw2, uw1)}
13884 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
13885 @tab @code{MWCUT @var{a},@var{b},@var{c}}
13886 @item @code{void __MWTACC (acc, uw1)}
13887 @tab @code{__MWTACC (@var{b}, @var{a})}
13888 @tab @code{MWTACC @var{a},@var{b}}
13889 @item @code{void __MWTACCG (acc, uw1)}
13890 @tab @code{__MWTACCG (@var{b}, @var{a})}
13891 @tab @code{MWTACCG @var{a},@var{b}}
13892 @item @code{uw1 __MXOR (uw1, uw1)}
13893 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
13894 @tab @code{MXOR @var{a},@var{b},@var{c}}
13897 @node Raw read/write Functions
13898 @subsubsection Raw Read/Write Functions
13900 This sections describes built-in functions related to read and write
13901 instructions to access memory. These functions generate
13902 @code{membar} instructions to flush the I/O load and stores where
13903 appropriate, as described in Fujitsu's manual described above.
13907 @item unsigned char __builtin_read8 (void *@var{data})
13908 @item unsigned short __builtin_read16 (void *@var{data})
13909 @item unsigned long __builtin_read32 (void *@var{data})
13910 @item unsigned long long __builtin_read64 (void *@var{data})
13912 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
13913 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
13914 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
13915 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
13918 @node Other Built-in Functions
13919 @subsubsection Other Built-in Functions
13921 This section describes built-in functions that are not named after
13922 a specific FR-V instruction.
13925 @item sw2 __IACCreadll (iacc @var{reg})
13926 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
13927 for future expansion and must be 0.
13929 @item sw1 __IACCreadl (iacc @var{reg})
13930 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
13931 Other values of @var{reg} are rejected as invalid.
13933 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
13934 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
13935 is reserved for future expansion and must be 0.
13937 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
13938 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
13939 is 1. Other values of @var{reg} are rejected as invalid.
13941 @item void __data_prefetch0 (const void *@var{x})
13942 Use the @code{dcpl} instruction to load the contents of address @var{x}
13943 into the data cache.
13945 @item void __data_prefetch (const void *@var{x})
13946 Use the @code{nldub} instruction to load the contents of address @var{x}
13947 into the data cache. The instruction is issued in slot I1@.
13950 @node MIPS DSP Built-in Functions
13951 @subsection MIPS DSP Built-in Functions
13953 The MIPS DSP Application-Specific Extension (ASE) includes new
13954 instructions that are designed to improve the performance of DSP and
13955 media applications. It provides instructions that operate on packed
13956 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
13958 GCC supports MIPS DSP operations using both the generic
13959 vector extensions (@pxref{Vector Extensions}) and a collection of
13960 MIPS-specific built-in functions. Both kinds of support are
13961 enabled by the @option{-mdsp} command-line option.
13963 Revision 2 of the ASE was introduced in the second half of 2006.
13964 This revision adds extra instructions to the original ASE, but is
13965 otherwise backwards-compatible with it. You can select revision 2
13966 using the command-line option @option{-mdspr2}; this option implies
13969 The SCOUNT and POS bits of the DSP control register are global. The
13970 WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and
13971 POS bits. During optimization, the compiler does not delete these
13972 instructions and it does not delete calls to functions containing
13973 these instructions.
13975 At present, GCC only provides support for operations on 32-bit
13976 vectors. The vector type associated with 8-bit integer data is
13977 usually called @code{v4i8}, the vector type associated with Q7
13978 is usually called @code{v4q7}, the vector type associated with 16-bit
13979 integer data is usually called @code{v2i16}, and the vector type
13980 associated with Q15 is usually called @code{v2q15}. They can be
13981 defined in C as follows:
13984 typedef signed char v4i8 __attribute__ ((vector_size(4)));
13985 typedef signed char v4q7 __attribute__ ((vector_size(4)));
13986 typedef short v2i16 __attribute__ ((vector_size(4)));
13987 typedef short v2q15 __attribute__ ((vector_size(4)));
13990 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
13991 initialized in the same way as aggregates. For example:
13994 v4i8 a = @{1, 2, 3, 4@};
13996 b = (v4i8) @{5, 6, 7, 8@};
13998 v2q15 c = @{0x0fcb, 0x3a75@};
14000 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
14003 @emph{Note:} The CPU's endianness determines the order in which values
14004 are packed. On little-endian targets, the first value is the least
14005 significant and the last value is the most significant. The opposite
14006 order applies to big-endian targets. For example, the code above
14007 sets the lowest byte of @code{a} to @code{1} on little-endian targets
14008 and @code{4} on big-endian targets.
14010 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
14011 representation. As shown in this example, the integer representation
14012 of a Q7 value can be obtained by multiplying the fractional value by
14013 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
14014 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
14017 The table below lists the @code{v4i8} and @code{v2q15} operations for which
14018 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
14019 and @code{c} and @code{d} are @code{v2q15} values.
14021 @multitable @columnfractions .50 .50
14022 @item C code @tab MIPS instruction
14023 @item @code{a + b} @tab @code{addu.qb}
14024 @item @code{c + d} @tab @code{addq.ph}
14025 @item @code{a - b} @tab @code{subu.qb}
14026 @item @code{c - d} @tab @code{subq.ph}
14029 The table below lists the @code{v2i16} operation for which
14030 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
14031 @code{v2i16} values.
14033 @multitable @columnfractions .50 .50
14034 @item C code @tab MIPS instruction
14035 @item @code{e * f} @tab @code{mul.ph}
14038 It is easier to describe the DSP built-in functions if we first define
14039 the following types:
14044 typedef unsigned int ui32;
14045 typedef long long a64;
14048 @code{q31} and @code{i32} are actually the same as @code{int}, but we
14049 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
14050 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
14051 @code{long long}, but we use @code{a64} to indicate values that are
14052 placed in one of the four DSP accumulators (@code{$ac0},
14053 @code{$ac1}, @code{$ac2} or @code{$ac3}).
14055 Also, some built-in functions prefer or require immediate numbers as
14056 parameters, because the corresponding DSP instructions accept both immediate
14057 numbers and register operands, or accept immediate numbers only. The
14058 immediate parameters are listed as follows.
14066 imm0_255: 0 to 255.
14067 imm_n32_31: -32 to 31.
14068 imm_n512_511: -512 to 511.
14071 The following built-in functions map directly to a particular MIPS DSP
14072 instruction. Please refer to the architecture specification
14073 for details on what each instruction does.
14076 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
14077 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
14078 q31 __builtin_mips_addq_s_w (q31, q31)
14079 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
14080 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
14081 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
14082 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
14083 q31 __builtin_mips_subq_s_w (q31, q31)
14084 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
14085 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
14086 i32 __builtin_mips_addsc (i32, i32)
14087 i32 __builtin_mips_addwc (i32, i32)
14088 i32 __builtin_mips_modsub (i32, i32)
14089 i32 __builtin_mips_raddu_w_qb (v4i8)
14090 v2q15 __builtin_mips_absq_s_ph (v2q15)
14091 q31 __builtin_mips_absq_s_w (q31)
14092 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
14093 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
14094 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
14095 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
14096 q31 __builtin_mips_preceq_w_phl (v2q15)
14097 q31 __builtin_mips_preceq_w_phr (v2q15)
14098 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
14099 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
14100 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
14101 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
14102 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
14103 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
14104 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
14105 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
14106 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
14107 v4i8 __builtin_mips_shll_qb (v4i8, i32)
14108 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
14109 v2q15 __builtin_mips_shll_ph (v2q15, i32)
14110 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
14111 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
14112 q31 __builtin_mips_shll_s_w (q31, imm0_31)
14113 q31 __builtin_mips_shll_s_w (q31, i32)
14114 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
14115 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
14116 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
14117 v2q15 __builtin_mips_shra_ph (v2q15, i32)
14118 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
14119 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
14120 q31 __builtin_mips_shra_r_w (q31, imm0_31)
14121 q31 __builtin_mips_shra_r_w (q31, i32)
14122 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
14123 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
14124 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
14125 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
14126 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
14127 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
14128 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
14129 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
14130 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
14131 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
14132 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
14133 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
14134 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
14135 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
14136 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
14137 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
14138 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
14139 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
14140 i32 __builtin_mips_bitrev (i32)
14141 i32 __builtin_mips_insv (i32, i32)
14142 v4i8 __builtin_mips_repl_qb (imm0_255)
14143 v4i8 __builtin_mips_repl_qb (i32)
14144 v2q15 __builtin_mips_repl_ph (imm_n512_511)
14145 v2q15 __builtin_mips_repl_ph (i32)
14146 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
14147 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
14148 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
14149 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
14150 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
14151 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
14152 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
14153 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
14154 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
14155 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
14156 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
14157 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
14158 i32 __builtin_mips_extr_w (a64, imm0_31)
14159 i32 __builtin_mips_extr_w (a64, i32)
14160 i32 __builtin_mips_extr_r_w (a64, imm0_31)
14161 i32 __builtin_mips_extr_s_h (a64, i32)
14162 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
14163 i32 __builtin_mips_extr_rs_w (a64, i32)
14164 i32 __builtin_mips_extr_s_h (a64, imm0_31)
14165 i32 __builtin_mips_extr_r_w (a64, i32)
14166 i32 __builtin_mips_extp (a64, imm0_31)
14167 i32 __builtin_mips_extp (a64, i32)
14168 i32 __builtin_mips_extpdp (a64, imm0_31)
14169 i32 __builtin_mips_extpdp (a64, i32)
14170 a64 __builtin_mips_shilo (a64, imm_n32_31)
14171 a64 __builtin_mips_shilo (a64, i32)
14172 a64 __builtin_mips_mthlip (a64, i32)
14173 void __builtin_mips_wrdsp (i32, imm0_63)
14174 i32 __builtin_mips_rddsp (imm0_63)
14175 i32 __builtin_mips_lbux (void *, i32)
14176 i32 __builtin_mips_lhx (void *, i32)
14177 i32 __builtin_mips_lwx (void *, i32)
14178 a64 __builtin_mips_ldx (void *, i32) [MIPS64 only]
14179 i32 __builtin_mips_bposge32 (void)
14180 a64 __builtin_mips_madd (a64, i32, i32);
14181 a64 __builtin_mips_maddu (a64, ui32, ui32);
14182 a64 __builtin_mips_msub (a64, i32, i32);
14183 a64 __builtin_mips_msubu (a64, ui32, ui32);
14184 a64 __builtin_mips_mult (i32, i32);
14185 a64 __builtin_mips_multu (ui32, ui32);
14188 The following built-in functions map directly to a particular MIPS DSP REV 2
14189 instruction. Please refer to the architecture specification
14190 for details on what each instruction does.
14193 v4q7 __builtin_mips_absq_s_qb (v4q7);
14194 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
14195 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
14196 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
14197 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
14198 i32 __builtin_mips_append (i32, i32, imm0_31);
14199 i32 __builtin_mips_balign (i32, i32, imm0_3);
14200 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
14201 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
14202 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
14203 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
14204 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
14205 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
14206 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
14207 q31 __builtin_mips_mulq_rs_w (q31, q31);
14208 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
14209 q31 __builtin_mips_mulq_s_w (q31, q31);
14210 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
14211 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
14212 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
14213 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
14214 i32 __builtin_mips_prepend (i32, i32, imm0_31);
14215 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
14216 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
14217 v4i8 __builtin_mips_shra_qb (v4i8, i32);
14218 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
14219 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
14220 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
14221 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
14222 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
14223 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
14224 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
14225 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
14226 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
14227 q31 __builtin_mips_addqh_w (q31, q31);
14228 q31 __builtin_mips_addqh_r_w (q31, q31);
14229 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
14230 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
14231 q31 __builtin_mips_subqh_w (q31, q31);
14232 q31 __builtin_mips_subqh_r_w (q31, q31);
14233 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
14234 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
14235 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
14236 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
14237 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
14238 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
14242 @node MIPS Paired-Single Support
14243 @subsection MIPS Paired-Single Support
14245 The MIPS64 architecture includes a number of instructions that
14246 operate on pairs of single-precision floating-point values.
14247 Each pair is packed into a 64-bit floating-point register,
14248 with one element being designated the ``upper half'' and
14249 the other being designated the ``lower half''.
14251 GCC supports paired-single operations using both the generic
14252 vector extensions (@pxref{Vector Extensions}) and a collection of
14253 MIPS-specific built-in functions. Both kinds of support are
14254 enabled by the @option{-mpaired-single} command-line option.
14256 The vector type associated with paired-single values is usually
14257 called @code{v2sf}. It can be defined in C as follows:
14260 typedef float v2sf __attribute__ ((vector_size (8)));
14263 @code{v2sf} values are initialized in the same way as aggregates.
14267 v2sf a = @{1.5, 9.1@};
14270 b = (v2sf) @{e, f@};
14273 @emph{Note:} The CPU's endianness determines which value is stored in
14274 the upper half of a register and which value is stored in the lower half.
14275 On little-endian targets, the first value is the lower one and the second
14276 value is the upper one. The opposite order applies to big-endian targets.
14277 For example, the code above sets the lower half of @code{a} to
14278 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
14280 @node MIPS Loongson Built-in Functions
14281 @subsection MIPS Loongson Built-in Functions
14283 GCC provides intrinsics to access the SIMD instructions provided by the
14284 ST Microelectronics Loongson-2E and -2F processors. These intrinsics,
14285 available after inclusion of the @code{loongson.h} header file,
14286 operate on the following 64-bit vector types:
14289 @item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers;
14290 @item @code{uint16x4_t}, a vector of four unsigned 16-bit integers;
14291 @item @code{uint32x2_t}, a vector of two unsigned 32-bit integers;
14292 @item @code{int8x8_t}, a vector of eight signed 8-bit integers;
14293 @item @code{int16x4_t}, a vector of four signed 16-bit integers;
14294 @item @code{int32x2_t}, a vector of two signed 32-bit integers.
14297 The intrinsics provided are listed below; each is named after the
14298 machine instruction to which it corresponds, with suffixes added as
14299 appropriate to distinguish intrinsics that expand to the same machine
14300 instruction yet have different argument types. Refer to the architecture
14301 documentation for a description of the functionality of each
14305 int16x4_t packsswh (int32x2_t s, int32x2_t t);
14306 int8x8_t packsshb (int16x4_t s, int16x4_t t);
14307 uint8x8_t packushb (uint16x4_t s, uint16x4_t t);
14308 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t);
14309 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t);
14310 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t);
14311 int32x2_t paddw_s (int32x2_t s, int32x2_t t);
14312 int16x4_t paddh_s (int16x4_t s, int16x4_t t);
14313 int8x8_t paddb_s (int8x8_t s, int8x8_t t);
14314 uint64_t paddd_u (uint64_t s, uint64_t t);
14315 int64_t paddd_s (int64_t s, int64_t t);
14316 int16x4_t paddsh (int16x4_t s, int16x4_t t);
14317 int8x8_t paddsb (int8x8_t s, int8x8_t t);
14318 uint16x4_t paddush (uint16x4_t s, uint16x4_t t);
14319 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t);
14320 uint64_t pandn_ud (uint64_t s, uint64_t t);
14321 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t);
14322 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t);
14323 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t);
14324 int64_t pandn_sd (int64_t s, int64_t t);
14325 int32x2_t pandn_sw (int32x2_t s, int32x2_t t);
14326 int16x4_t pandn_sh (int16x4_t s, int16x4_t t);
14327 int8x8_t pandn_sb (int8x8_t s, int8x8_t t);
14328 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t);
14329 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t);
14330 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t);
14331 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t);
14332 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t);
14333 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t);
14334 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t);
14335 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t);
14336 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t);
14337 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t);
14338 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t);
14339 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t);
14340 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t);
14341 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t);
14342 uint16x4_t pextrh_u (uint16x4_t s, int field);
14343 int16x4_t pextrh_s (int16x4_t s, int field);
14344 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t);
14345 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t);
14346 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t);
14347 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t);
14348 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t);
14349 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t);
14350 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t);
14351 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t);
14352 int32x2_t pmaddhw (int16x4_t s, int16x4_t t);
14353 int16x4_t pmaxsh (int16x4_t s, int16x4_t t);
14354 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t);
14355 int16x4_t pminsh (int16x4_t s, int16x4_t t);
14356 uint8x8_t pminub (uint8x8_t s, uint8x8_t t);
14357 uint8x8_t pmovmskb_u (uint8x8_t s);
14358 int8x8_t pmovmskb_s (int8x8_t s);
14359 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t);
14360 int16x4_t pmulhh (int16x4_t s, int16x4_t t);
14361 int16x4_t pmullh (int16x4_t s, int16x4_t t);
14362 int64_t pmuluw (uint32x2_t s, uint32x2_t t);
14363 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t);
14364 uint16x4_t biadd (uint8x8_t s);
14365 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t);
14366 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order);
14367 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order);
14368 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount);
14369 int16x4_t psllh_s (int16x4_t s, uint8_t amount);
14370 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount);
14371 int32x2_t psllw_s (int32x2_t s, uint8_t amount);
14372 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount);
14373 int16x4_t psrlh_s (int16x4_t s, uint8_t amount);
14374 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount);
14375 int32x2_t psrlw_s (int32x2_t s, uint8_t amount);
14376 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount);
14377 int16x4_t psrah_s (int16x4_t s, uint8_t amount);
14378 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount);
14379 int32x2_t psraw_s (int32x2_t s, uint8_t amount);
14380 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t);
14381 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t);
14382 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t);
14383 int32x2_t psubw_s (int32x2_t s, int32x2_t t);
14384 int16x4_t psubh_s (int16x4_t s, int16x4_t t);
14385 int8x8_t psubb_s (int8x8_t s, int8x8_t t);
14386 uint64_t psubd_u (uint64_t s, uint64_t t);
14387 int64_t psubd_s (int64_t s, int64_t t);
14388 int16x4_t psubsh (int16x4_t s, int16x4_t t);
14389 int8x8_t psubsb (int8x8_t s, int8x8_t t);
14390 uint16x4_t psubush (uint16x4_t s, uint16x4_t t);
14391 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t);
14392 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t);
14393 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t);
14394 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t);
14395 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t);
14396 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t);
14397 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t);
14398 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t);
14399 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t);
14400 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t);
14401 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t);
14402 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t);
14403 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t);
14407 * Paired-Single Arithmetic::
14408 * Paired-Single Built-in Functions::
14409 * MIPS-3D Built-in Functions::
14412 @node Paired-Single Arithmetic
14413 @subsubsection Paired-Single Arithmetic
14415 The table below lists the @code{v2sf} operations for which hardware
14416 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
14417 values and @code{x} is an integral value.
14419 @multitable @columnfractions .50 .50
14420 @item C code @tab MIPS instruction
14421 @item @code{a + b} @tab @code{add.ps}
14422 @item @code{a - b} @tab @code{sub.ps}
14423 @item @code{-a} @tab @code{neg.ps}
14424 @item @code{a * b} @tab @code{mul.ps}
14425 @item @code{a * b + c} @tab @code{madd.ps}
14426 @item @code{a * b - c} @tab @code{msub.ps}
14427 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
14428 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
14429 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
14432 Note that the multiply-accumulate instructions can be disabled
14433 using the command-line option @code{-mno-fused-madd}.
14435 @node Paired-Single Built-in Functions
14436 @subsubsection Paired-Single Built-in Functions
14438 The following paired-single functions map directly to a particular
14439 MIPS instruction. Please refer to the architecture specification
14440 for details on what each instruction does.
14443 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
14444 Pair lower lower (@code{pll.ps}).
14446 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
14447 Pair upper lower (@code{pul.ps}).
14449 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
14450 Pair lower upper (@code{plu.ps}).
14452 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
14453 Pair upper upper (@code{puu.ps}).
14455 @item v2sf __builtin_mips_cvt_ps_s (float, float)
14456 Convert pair to paired single (@code{cvt.ps.s}).
14458 @item float __builtin_mips_cvt_s_pl (v2sf)
14459 Convert pair lower to single (@code{cvt.s.pl}).
14461 @item float __builtin_mips_cvt_s_pu (v2sf)
14462 Convert pair upper to single (@code{cvt.s.pu}).
14464 @item v2sf __builtin_mips_abs_ps (v2sf)
14465 Absolute value (@code{abs.ps}).
14467 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
14468 Align variable (@code{alnv.ps}).
14470 @emph{Note:} The value of the third parameter must be 0 or 4
14471 modulo 8, otherwise the result is unpredictable. Please read the
14472 instruction description for details.
14475 The following multi-instruction functions are also available.
14476 In each case, @var{cond} can be any of the 16 floating-point conditions:
14477 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
14478 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
14479 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
14482 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14483 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14484 Conditional move based on floating-point comparison (@code{c.@var{cond}.ps},
14485 @code{movt.ps}/@code{movf.ps}).
14487 The @code{movt} functions return the value @var{x} computed by:
14490 c.@var{cond}.ps @var{cc},@var{a},@var{b}
14491 mov.ps @var{x},@var{c}
14492 movt.ps @var{x},@var{d},@var{cc}
14495 The @code{movf} functions are similar but use @code{movf.ps} instead
14498 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14499 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14500 Comparison of two paired-single values (@code{c.@var{cond}.ps},
14501 @code{bc1t}/@code{bc1f}).
14503 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
14504 and return either the upper or lower half of the result. For example:
14508 if (__builtin_mips_upper_c_eq_ps (a, b))
14509 upper_halves_are_equal ();
14511 upper_halves_are_unequal ();
14513 if (__builtin_mips_lower_c_eq_ps (a, b))
14514 lower_halves_are_equal ();
14516 lower_halves_are_unequal ();
14520 @node MIPS-3D Built-in Functions
14521 @subsubsection MIPS-3D Built-in Functions
14523 The MIPS-3D Application-Specific Extension (ASE) includes additional
14524 paired-single instructions that are designed to improve the performance
14525 of 3D graphics operations. Support for these instructions is controlled
14526 by the @option{-mips3d} command-line option.
14528 The functions listed below map directly to a particular MIPS-3D
14529 instruction. Please refer to the architecture specification for
14530 more details on what each instruction does.
14533 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
14534 Reduction add (@code{addr.ps}).
14536 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
14537 Reduction multiply (@code{mulr.ps}).
14539 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
14540 Convert paired single to paired word (@code{cvt.pw.ps}).
14542 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
14543 Convert paired word to paired single (@code{cvt.ps.pw}).
14545 @item float __builtin_mips_recip1_s (float)
14546 @itemx double __builtin_mips_recip1_d (double)
14547 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
14548 Reduced-precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
14550 @item float __builtin_mips_recip2_s (float, float)
14551 @itemx double __builtin_mips_recip2_d (double, double)
14552 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
14553 Reduced-precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
14555 @item float __builtin_mips_rsqrt1_s (float)
14556 @itemx double __builtin_mips_rsqrt1_d (double)
14557 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
14558 Reduced-precision reciprocal square root (sequence step 1)
14559 (@code{rsqrt1.@var{fmt}}).
14561 @item float __builtin_mips_rsqrt2_s (float, float)
14562 @itemx double __builtin_mips_rsqrt2_d (double, double)
14563 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
14564 Reduced-precision reciprocal square root (sequence step 2)
14565 (@code{rsqrt2.@var{fmt}}).
14568 The following multi-instruction functions are also available.
14569 In each case, @var{cond} can be any of the 16 floating-point conditions:
14570 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
14571 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
14572 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
14575 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
14576 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
14577 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
14578 @code{bc1t}/@code{bc1f}).
14580 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
14581 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
14586 if (__builtin_mips_cabs_eq_s (a, b))
14592 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14593 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14594 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
14595 @code{bc1t}/@code{bc1f}).
14597 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
14598 and return either the upper or lower half of the result. For example:
14602 if (__builtin_mips_upper_cabs_eq_ps (a, b))
14603 upper_halves_are_equal ();
14605 upper_halves_are_unequal ();
14607 if (__builtin_mips_lower_cabs_eq_ps (a, b))
14608 lower_halves_are_equal ();
14610 lower_halves_are_unequal ();
14613 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14614 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14615 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
14616 @code{movt.ps}/@code{movf.ps}).
14618 The @code{movt} functions return the value @var{x} computed by:
14621 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
14622 mov.ps @var{x},@var{c}
14623 movt.ps @var{x},@var{d},@var{cc}
14626 The @code{movf} functions are similar but use @code{movf.ps} instead
14629 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14630 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14631 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14632 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
14633 Comparison of two paired-single values
14634 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
14635 @code{bc1any2t}/@code{bc1any2f}).
14637 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
14638 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
14639 result is true and the @code{all} forms return true if both results are true.
14644 if (__builtin_mips_any_c_eq_ps (a, b))
14649 if (__builtin_mips_all_c_eq_ps (a, b))
14655 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14656 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14657 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14658 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
14659 Comparison of four paired-single values
14660 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
14661 @code{bc1any4t}/@code{bc1any4f}).
14663 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
14664 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
14665 The @code{any} forms return true if any of the four results are true
14666 and the @code{all} forms return true if all four results are true.
14671 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
14676 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
14683 @node MIPS SIMD Architecture (MSA) Support
14684 @subsection MIPS SIMD Architecture (MSA) Support
14687 * MIPS SIMD Architecture Built-in Functions::
14690 GCC provides intrinsics to access the SIMD instructions provided by the
14691 MSA MIPS SIMD Architecture. The interface is made available by including
14692 @code{<msa.h>} and using @option{-mmsa -mhard-float -mfp64 -mnan=2008}.
14693 For each @code{__builtin_msa_*}, there is a shortened name of the intrinsic,
14696 MSA implements 128-bit wide vector registers, operating on 8-, 16-, 32- and
14697 64-bit integer, 16- and 32-bit fixed-point, or 32- and 64-bit floating point
14698 data elements. The following vectors typedefs are included in @code{msa.h}:
14700 @item @code{v16i8}, a vector of sixteen signed 8-bit integers;
14701 @item @code{v16u8}, a vector of sixteen unsigned 8-bit integers;
14702 @item @code{v8i16}, a vector of eight signed 16-bit integers;
14703 @item @code{v8u16}, a vector of eight unsigned 16-bit integers;
14704 @item @code{v4i32}, a vector of four signed 32-bit integers;
14705 @item @code{v4u32}, a vector of four unsigned 32-bit integers;
14706 @item @code{v2i64}, a vector of two signed 64-bit integers;
14707 @item @code{v2u64}, a vector of two unsigned 64-bit integers;
14708 @item @code{v4f32}, a vector of four 32-bit floats;
14709 @item @code{v2f64}, a vector of two 64-bit doubles.
14712 Instructions and corresponding built-ins may have additional restrictions and/or
14713 input/output values manipulated:
14715 @item @code{imm0_1}, an integer literal in range 0 to 1;
14716 @item @code{imm0_3}, an integer literal in range 0 to 3;
14717 @item @code{imm0_7}, an integer literal in range 0 to 7;
14718 @item @code{imm0_15}, an integer literal in range 0 to 15;
14719 @item @code{imm0_31}, an integer literal in range 0 to 31;
14720 @item @code{imm0_63}, an integer literal in range 0 to 63;
14721 @item @code{imm0_255}, an integer literal in range 0 to 255;
14722 @item @code{imm_n16_15}, an integer literal in range -16 to 15;
14723 @item @code{imm_n512_511}, an integer literal in range -512 to 511;
14724 @item @code{imm_n1024_1022}, an integer literal in range -512 to 511 left
14725 shifted by 1 bit, i.e., -1024, -1022, @dots{}, 1020, 1022;
14726 @item @code{imm_n2048_2044}, an integer literal in range -512 to 511 left
14727 shifted by 2 bits, i.e., -2048, -2044, @dots{}, 2040, 2044;
14728 @item @code{imm_n4096_4088}, an integer literal in range -512 to 511 left
14729 shifted by 3 bits, i.e., -4096, -4088, @dots{}, 4080, 4088;
14730 @item @code{imm1_4}, an integer literal in range 1 to 4;
14731 @item @code{i32, i64, u32, u64, f32, f64}, defined as follows:
14737 #if __LONG_MAX__ == __LONG_LONG_MAX__
14740 typedef long long i64;
14743 typedef unsigned int u32;
14744 #if __LONG_MAX__ == __LONG_LONG_MAX__
14745 typedef unsigned long u64;
14747 typedef unsigned long long u64;
14750 typedef double f64;
14755 @node MIPS SIMD Architecture Built-in Functions
14756 @subsubsection MIPS SIMD Architecture Built-in Functions
14758 The intrinsics provided are listed below; each is named after the
14759 machine instruction.
14762 v16i8 __builtin_msa_add_a_b (v16i8, v16i8);
14763 v8i16 __builtin_msa_add_a_h (v8i16, v8i16);
14764 v4i32 __builtin_msa_add_a_w (v4i32, v4i32);
14765 v2i64 __builtin_msa_add_a_d (v2i64, v2i64);
14767 v16i8 __builtin_msa_adds_a_b (v16i8, v16i8);
14768 v8i16 __builtin_msa_adds_a_h (v8i16, v8i16);
14769 v4i32 __builtin_msa_adds_a_w (v4i32, v4i32);
14770 v2i64 __builtin_msa_adds_a_d (v2i64, v2i64);
14772 v16i8 __builtin_msa_adds_s_b (v16i8, v16i8);
14773 v8i16 __builtin_msa_adds_s_h (v8i16, v8i16);
14774 v4i32 __builtin_msa_adds_s_w (v4i32, v4i32);
14775 v2i64 __builtin_msa_adds_s_d (v2i64, v2i64);
14777 v16u8 __builtin_msa_adds_u_b (v16u8, v16u8);
14778 v8u16 __builtin_msa_adds_u_h (v8u16, v8u16);
14779 v4u32 __builtin_msa_adds_u_w (v4u32, v4u32);
14780 v2u64 __builtin_msa_adds_u_d (v2u64, v2u64);
14782 v16i8 __builtin_msa_addv_b (v16i8, v16i8);
14783 v8i16 __builtin_msa_addv_h (v8i16, v8i16);
14784 v4i32 __builtin_msa_addv_w (v4i32, v4i32);
14785 v2i64 __builtin_msa_addv_d (v2i64, v2i64);
14787 v16i8 __builtin_msa_addvi_b (v16i8, imm0_31);
14788 v8i16 __builtin_msa_addvi_h (v8i16, imm0_31);
14789 v4i32 __builtin_msa_addvi_w (v4i32, imm0_31);
14790 v2i64 __builtin_msa_addvi_d (v2i64, imm0_31);
14792 v16u8 __builtin_msa_and_v (v16u8, v16u8);
14794 v16u8 __builtin_msa_andi_b (v16u8, imm0_255);
14796 v16i8 __builtin_msa_asub_s_b (v16i8, v16i8);
14797 v8i16 __builtin_msa_asub_s_h (v8i16, v8i16);
14798 v4i32 __builtin_msa_asub_s_w (v4i32, v4i32);
14799 v2i64 __builtin_msa_asub_s_d (v2i64, v2i64);
14801 v16u8 __builtin_msa_asub_u_b (v16u8, v16u8);
14802 v8u16 __builtin_msa_asub_u_h (v8u16, v8u16);
14803 v4u32 __builtin_msa_asub_u_w (v4u32, v4u32);
14804 v2u64 __builtin_msa_asub_u_d (v2u64, v2u64);
14806 v16i8 __builtin_msa_ave_s_b (v16i8, v16i8);
14807 v8i16 __builtin_msa_ave_s_h (v8i16, v8i16);
14808 v4i32 __builtin_msa_ave_s_w (v4i32, v4i32);
14809 v2i64 __builtin_msa_ave_s_d (v2i64, v2i64);
14811 v16u8 __builtin_msa_ave_u_b (v16u8, v16u8);
14812 v8u16 __builtin_msa_ave_u_h (v8u16, v8u16);
14813 v4u32 __builtin_msa_ave_u_w (v4u32, v4u32);
14814 v2u64 __builtin_msa_ave_u_d (v2u64, v2u64);
14816 v16i8 __builtin_msa_aver_s_b (v16i8, v16i8);
14817 v8i16 __builtin_msa_aver_s_h (v8i16, v8i16);
14818 v4i32 __builtin_msa_aver_s_w (v4i32, v4i32);
14819 v2i64 __builtin_msa_aver_s_d (v2i64, v2i64);
14821 v16u8 __builtin_msa_aver_u_b (v16u8, v16u8);
14822 v8u16 __builtin_msa_aver_u_h (v8u16, v8u16);
14823 v4u32 __builtin_msa_aver_u_w (v4u32, v4u32);
14824 v2u64 __builtin_msa_aver_u_d (v2u64, v2u64);
14826 v16u8 __builtin_msa_bclr_b (v16u8, v16u8);
14827 v8u16 __builtin_msa_bclr_h (v8u16, v8u16);
14828 v4u32 __builtin_msa_bclr_w (v4u32, v4u32);
14829 v2u64 __builtin_msa_bclr_d (v2u64, v2u64);
14831 v16u8 __builtin_msa_bclri_b (v16u8, imm0_7);
14832 v8u16 __builtin_msa_bclri_h (v8u16, imm0_15);
14833 v4u32 __builtin_msa_bclri_w (v4u32, imm0_31);
14834 v2u64 __builtin_msa_bclri_d (v2u64, imm0_63);
14836 v16u8 __builtin_msa_binsl_b (v16u8, v16u8, v16u8);
14837 v8u16 __builtin_msa_binsl_h (v8u16, v8u16, v8u16);
14838 v4u32 __builtin_msa_binsl_w (v4u32, v4u32, v4u32);
14839 v2u64 __builtin_msa_binsl_d (v2u64, v2u64, v2u64);
14841 v16u8 __builtin_msa_binsli_b (v16u8, v16u8, imm0_7);
14842 v8u16 __builtin_msa_binsli_h (v8u16, v8u16, imm0_15);
14843 v4u32 __builtin_msa_binsli_w (v4u32, v4u32, imm0_31);
14844 v2u64 __builtin_msa_binsli_d (v2u64, v2u64, imm0_63);
14846 v16u8 __builtin_msa_binsr_b (v16u8, v16u8, v16u8);
14847 v8u16 __builtin_msa_binsr_h (v8u16, v8u16, v8u16);
14848 v4u32 __builtin_msa_binsr_w (v4u32, v4u32, v4u32);
14849 v2u64 __builtin_msa_binsr_d (v2u64, v2u64, v2u64);
14851 v16u8 __builtin_msa_binsri_b (v16u8, v16u8, imm0_7);
14852 v8u16 __builtin_msa_binsri_h (v8u16, v8u16, imm0_15);
14853 v4u32 __builtin_msa_binsri_w (v4u32, v4u32, imm0_31);
14854 v2u64 __builtin_msa_binsri_d (v2u64, v2u64, imm0_63);
14856 v16u8 __builtin_msa_bmnz_v (v16u8, v16u8, v16u8);
14858 v16u8 __builtin_msa_bmnzi_b (v16u8, v16u8, imm0_255);
14860 v16u8 __builtin_msa_bmz_v (v16u8, v16u8, v16u8);
14862 v16u8 __builtin_msa_bmzi_b (v16u8, v16u8, imm0_255);
14864 v16u8 __builtin_msa_bneg_b (v16u8, v16u8);
14865 v8u16 __builtin_msa_bneg_h (v8u16, v8u16);
14866 v4u32 __builtin_msa_bneg_w (v4u32, v4u32);
14867 v2u64 __builtin_msa_bneg_d (v2u64, v2u64);
14869 v16u8 __builtin_msa_bnegi_b (v16u8, imm0_7);
14870 v8u16 __builtin_msa_bnegi_h (v8u16, imm0_15);
14871 v4u32 __builtin_msa_bnegi_w (v4u32, imm0_31);
14872 v2u64 __builtin_msa_bnegi_d (v2u64, imm0_63);
14874 i32 __builtin_msa_bnz_b (v16u8);
14875 i32 __builtin_msa_bnz_h (v8u16);
14876 i32 __builtin_msa_bnz_w (v4u32);
14877 i32 __builtin_msa_bnz_d (v2u64);
14879 i32 __builtin_msa_bnz_v (v16u8);
14881 v16u8 __builtin_msa_bsel_v (v16u8, v16u8, v16u8);
14883 v16u8 __builtin_msa_bseli_b (v16u8, v16u8, imm0_255);
14885 v16u8 __builtin_msa_bset_b (v16u8, v16u8);
14886 v8u16 __builtin_msa_bset_h (v8u16, v8u16);
14887 v4u32 __builtin_msa_bset_w (v4u32, v4u32);
14888 v2u64 __builtin_msa_bset_d (v2u64, v2u64);
14890 v16u8 __builtin_msa_bseti_b (v16u8, imm0_7);
14891 v8u16 __builtin_msa_bseti_h (v8u16, imm0_15);
14892 v4u32 __builtin_msa_bseti_w (v4u32, imm0_31);
14893 v2u64 __builtin_msa_bseti_d (v2u64, imm0_63);
14895 i32 __builtin_msa_bz_b (v16u8);
14896 i32 __builtin_msa_bz_h (v8u16);
14897 i32 __builtin_msa_bz_w (v4u32);
14898 i32 __builtin_msa_bz_d (v2u64);
14900 i32 __builtin_msa_bz_v (v16u8);
14902 v16i8 __builtin_msa_ceq_b (v16i8, v16i8);
14903 v8i16 __builtin_msa_ceq_h (v8i16, v8i16);
14904 v4i32 __builtin_msa_ceq_w (v4i32, v4i32);
14905 v2i64 __builtin_msa_ceq_d (v2i64, v2i64);
14907 v16i8 __builtin_msa_ceqi_b (v16i8, imm_n16_15);
14908 v8i16 __builtin_msa_ceqi_h (v8i16, imm_n16_15);
14909 v4i32 __builtin_msa_ceqi_w (v4i32, imm_n16_15);
14910 v2i64 __builtin_msa_ceqi_d (v2i64, imm_n16_15);
14912 i32 __builtin_msa_cfcmsa (imm0_31);
14914 v16i8 __builtin_msa_cle_s_b (v16i8, v16i8);
14915 v8i16 __builtin_msa_cle_s_h (v8i16, v8i16);
14916 v4i32 __builtin_msa_cle_s_w (v4i32, v4i32);
14917 v2i64 __builtin_msa_cle_s_d (v2i64, v2i64);
14919 v16i8 __builtin_msa_cle_u_b (v16u8, v16u8);
14920 v8i16 __builtin_msa_cle_u_h (v8u16, v8u16);
14921 v4i32 __builtin_msa_cle_u_w (v4u32, v4u32);
14922 v2i64 __builtin_msa_cle_u_d (v2u64, v2u64);
14924 v16i8 __builtin_msa_clei_s_b (v16i8, imm_n16_15);
14925 v8i16 __builtin_msa_clei_s_h (v8i16, imm_n16_15);
14926 v4i32 __builtin_msa_clei_s_w (v4i32, imm_n16_15);
14927 v2i64 __builtin_msa_clei_s_d (v2i64, imm_n16_15);
14929 v16i8 __builtin_msa_clei_u_b (v16u8, imm0_31);
14930 v8i16 __builtin_msa_clei_u_h (v8u16, imm0_31);
14931 v4i32 __builtin_msa_clei_u_w (v4u32, imm0_31);
14932 v2i64 __builtin_msa_clei_u_d (v2u64, imm0_31);
14934 v16i8 __builtin_msa_clt_s_b (v16i8, v16i8);
14935 v8i16 __builtin_msa_clt_s_h (v8i16, v8i16);
14936 v4i32 __builtin_msa_clt_s_w (v4i32, v4i32);
14937 v2i64 __builtin_msa_clt_s_d (v2i64, v2i64);
14939 v16i8 __builtin_msa_clt_u_b (v16u8, v16u8);
14940 v8i16 __builtin_msa_clt_u_h (v8u16, v8u16);
14941 v4i32 __builtin_msa_clt_u_w (v4u32, v4u32);
14942 v2i64 __builtin_msa_clt_u_d (v2u64, v2u64);
14944 v16i8 __builtin_msa_clti_s_b (v16i8, imm_n16_15);
14945 v8i16 __builtin_msa_clti_s_h (v8i16, imm_n16_15);
14946 v4i32 __builtin_msa_clti_s_w (v4i32, imm_n16_15);
14947 v2i64 __builtin_msa_clti_s_d (v2i64, imm_n16_15);
14949 v16i8 __builtin_msa_clti_u_b (v16u8, imm0_31);
14950 v8i16 __builtin_msa_clti_u_h (v8u16, imm0_31);
14951 v4i32 __builtin_msa_clti_u_w (v4u32, imm0_31);
14952 v2i64 __builtin_msa_clti_u_d (v2u64, imm0_31);
14954 i32 __builtin_msa_copy_s_b (v16i8, imm0_15);
14955 i32 __builtin_msa_copy_s_h (v8i16, imm0_7);
14956 i32 __builtin_msa_copy_s_w (v4i32, imm0_3);
14957 i64 __builtin_msa_copy_s_d (v2i64, imm0_1);
14959 u32 __builtin_msa_copy_u_b (v16i8, imm0_15);
14960 u32 __builtin_msa_copy_u_h (v8i16, imm0_7);
14961 u32 __builtin_msa_copy_u_w (v4i32, imm0_3);
14962 u64 __builtin_msa_copy_u_d (v2i64, imm0_1);
14964 void __builtin_msa_ctcmsa (imm0_31, i32);
14966 v16i8 __builtin_msa_div_s_b (v16i8, v16i8);
14967 v8i16 __builtin_msa_div_s_h (v8i16, v8i16);
14968 v4i32 __builtin_msa_div_s_w (v4i32, v4i32);
14969 v2i64 __builtin_msa_div_s_d (v2i64, v2i64);
14971 v16u8 __builtin_msa_div_u_b (v16u8, v16u8);
14972 v8u16 __builtin_msa_div_u_h (v8u16, v8u16);
14973 v4u32 __builtin_msa_div_u_w (v4u32, v4u32);
14974 v2u64 __builtin_msa_div_u_d (v2u64, v2u64);
14976 v8i16 __builtin_msa_dotp_s_h (v16i8, v16i8);
14977 v4i32 __builtin_msa_dotp_s_w (v8i16, v8i16);
14978 v2i64 __builtin_msa_dotp_s_d (v4i32, v4i32);
14980 v8u16 __builtin_msa_dotp_u_h (v16u8, v16u8);
14981 v4u32 __builtin_msa_dotp_u_w (v8u16, v8u16);
14982 v2u64 __builtin_msa_dotp_u_d (v4u32, v4u32);
14984 v8i16 __builtin_msa_dpadd_s_h (v8i16, v16i8, v16i8);
14985 v4i32 __builtin_msa_dpadd_s_w (v4i32, v8i16, v8i16);
14986 v2i64 __builtin_msa_dpadd_s_d (v2i64, v4i32, v4i32);
14988 v8u16 __builtin_msa_dpadd_u_h (v8u16, v16u8, v16u8);
14989 v4u32 __builtin_msa_dpadd_u_w (v4u32, v8u16, v8u16);
14990 v2u64 __builtin_msa_dpadd_u_d (v2u64, v4u32, v4u32);
14992 v8i16 __builtin_msa_dpsub_s_h (v8i16, v16i8, v16i8);
14993 v4i32 __builtin_msa_dpsub_s_w (v4i32, v8i16, v8i16);
14994 v2i64 __builtin_msa_dpsub_s_d (v2i64, v4i32, v4i32);
14996 v8i16 __builtin_msa_dpsub_u_h (v8i16, v16u8, v16u8);
14997 v4i32 __builtin_msa_dpsub_u_w (v4i32, v8u16, v8u16);
14998 v2i64 __builtin_msa_dpsub_u_d (v2i64, v4u32, v4u32);
15000 v4f32 __builtin_msa_fadd_w (v4f32, v4f32);
15001 v2f64 __builtin_msa_fadd_d (v2f64, v2f64);
15003 v4i32 __builtin_msa_fcaf_w (v4f32, v4f32);
15004 v2i64 __builtin_msa_fcaf_d (v2f64, v2f64);
15006 v4i32 __builtin_msa_fceq_w (v4f32, v4f32);
15007 v2i64 __builtin_msa_fceq_d (v2f64, v2f64);
15009 v4i32 __builtin_msa_fclass_w (v4f32);
15010 v2i64 __builtin_msa_fclass_d (v2f64);
15012 v4i32 __builtin_msa_fcle_w (v4f32, v4f32);
15013 v2i64 __builtin_msa_fcle_d (v2f64, v2f64);
15015 v4i32 __builtin_msa_fclt_w (v4f32, v4f32);
15016 v2i64 __builtin_msa_fclt_d (v2f64, v2f64);
15018 v4i32 __builtin_msa_fcne_w (v4f32, v4f32);
15019 v2i64 __builtin_msa_fcne_d (v2f64, v2f64);
15021 v4i32 __builtin_msa_fcor_w (v4f32, v4f32);
15022 v2i64 __builtin_msa_fcor_d (v2f64, v2f64);
15024 v4i32 __builtin_msa_fcueq_w (v4f32, v4f32);
15025 v2i64 __builtin_msa_fcueq_d (v2f64, v2f64);
15027 v4i32 __builtin_msa_fcule_w (v4f32, v4f32);
15028 v2i64 __builtin_msa_fcule_d (v2f64, v2f64);
15030 v4i32 __builtin_msa_fcult_w (v4f32, v4f32);
15031 v2i64 __builtin_msa_fcult_d (v2f64, v2f64);
15033 v4i32 __builtin_msa_fcun_w (v4f32, v4f32);
15034 v2i64 __builtin_msa_fcun_d (v2f64, v2f64);
15036 v4i32 __builtin_msa_fcune_w (v4f32, v4f32);
15037 v2i64 __builtin_msa_fcune_d (v2f64, v2f64);
15039 v4f32 __builtin_msa_fdiv_w (v4f32, v4f32);
15040 v2f64 __builtin_msa_fdiv_d (v2f64, v2f64);
15042 v8i16 __builtin_msa_fexdo_h (v4f32, v4f32);
15043 v4f32 __builtin_msa_fexdo_w (v2f64, v2f64);
15045 v4f32 __builtin_msa_fexp2_w (v4f32, v4i32);
15046 v2f64 __builtin_msa_fexp2_d (v2f64, v2i64);
15048 v4f32 __builtin_msa_fexupl_w (v8i16);
15049 v2f64 __builtin_msa_fexupl_d (v4f32);
15051 v4f32 __builtin_msa_fexupr_w (v8i16);
15052 v2f64 __builtin_msa_fexupr_d (v4f32);
15054 v4f32 __builtin_msa_ffint_s_w (v4i32);
15055 v2f64 __builtin_msa_ffint_s_d (v2i64);
15057 v4f32 __builtin_msa_ffint_u_w (v4u32);
15058 v2f64 __builtin_msa_ffint_u_d (v2u64);
15060 v4f32 __builtin_msa_ffql_w (v8i16);
15061 v2f64 __builtin_msa_ffql_d (v4i32);
15063 v4f32 __builtin_msa_ffqr_w (v8i16);
15064 v2f64 __builtin_msa_ffqr_d (v4i32);
15066 v16i8 __builtin_msa_fill_b (i32);
15067 v8i16 __builtin_msa_fill_h (i32);
15068 v4i32 __builtin_msa_fill_w (i32);
15069 v2i64 __builtin_msa_fill_d (i64);
15071 v4f32 __builtin_msa_flog2_w (v4f32);
15072 v2f64 __builtin_msa_flog2_d (v2f64);
15074 v4f32 __builtin_msa_fmadd_w (v4f32, v4f32, v4f32);
15075 v2f64 __builtin_msa_fmadd_d (v2f64, v2f64, v2f64);
15077 v4f32 __builtin_msa_fmax_w (v4f32, v4f32);
15078 v2f64 __builtin_msa_fmax_d (v2f64, v2f64);
15080 v4f32 __builtin_msa_fmax_a_w (v4f32, v4f32);
15081 v2f64 __builtin_msa_fmax_a_d (v2f64, v2f64);
15083 v4f32 __builtin_msa_fmin_w (v4f32, v4f32);
15084 v2f64 __builtin_msa_fmin_d (v2f64, v2f64);
15086 v4f32 __builtin_msa_fmin_a_w (v4f32, v4f32);
15087 v2f64 __builtin_msa_fmin_a_d (v2f64, v2f64);
15089 v4f32 __builtin_msa_fmsub_w (v4f32, v4f32, v4f32);
15090 v2f64 __builtin_msa_fmsub_d (v2f64, v2f64, v2f64);
15092 v4f32 __builtin_msa_fmul_w (v4f32, v4f32);
15093 v2f64 __builtin_msa_fmul_d (v2f64, v2f64);
15095 v4f32 __builtin_msa_frint_w (v4f32);
15096 v2f64 __builtin_msa_frint_d (v2f64);
15098 v4f32 __builtin_msa_frcp_w (v4f32);
15099 v2f64 __builtin_msa_frcp_d (v2f64);
15101 v4f32 __builtin_msa_frsqrt_w (v4f32);
15102 v2f64 __builtin_msa_frsqrt_d (v2f64);
15104 v4i32 __builtin_msa_fsaf_w (v4f32, v4f32);
15105 v2i64 __builtin_msa_fsaf_d (v2f64, v2f64);
15107 v4i32 __builtin_msa_fseq_w (v4f32, v4f32);
15108 v2i64 __builtin_msa_fseq_d (v2f64, v2f64);
15110 v4i32 __builtin_msa_fsle_w (v4f32, v4f32);
15111 v2i64 __builtin_msa_fsle_d (v2f64, v2f64);
15113 v4i32 __builtin_msa_fslt_w (v4f32, v4f32);
15114 v2i64 __builtin_msa_fslt_d (v2f64, v2f64);
15116 v4i32 __builtin_msa_fsne_w (v4f32, v4f32);
15117 v2i64 __builtin_msa_fsne_d (v2f64, v2f64);
15119 v4i32 __builtin_msa_fsor_w (v4f32, v4f32);
15120 v2i64 __builtin_msa_fsor_d (v2f64, v2f64);
15122 v4f32 __builtin_msa_fsqrt_w (v4f32);
15123 v2f64 __builtin_msa_fsqrt_d (v2f64);
15125 v4f32 __builtin_msa_fsub_w (v4f32, v4f32);
15126 v2f64 __builtin_msa_fsub_d (v2f64, v2f64);
15128 v4i32 __builtin_msa_fsueq_w (v4f32, v4f32);
15129 v2i64 __builtin_msa_fsueq_d (v2f64, v2f64);
15131 v4i32 __builtin_msa_fsule_w (v4f32, v4f32);
15132 v2i64 __builtin_msa_fsule_d (v2f64, v2f64);
15134 v4i32 __builtin_msa_fsult_w (v4f32, v4f32);
15135 v2i64 __builtin_msa_fsult_d (v2f64, v2f64);
15137 v4i32 __builtin_msa_fsun_w (v4f32, v4f32);
15138 v2i64 __builtin_msa_fsun_d (v2f64, v2f64);
15140 v4i32 __builtin_msa_fsune_w (v4f32, v4f32);
15141 v2i64 __builtin_msa_fsune_d (v2f64, v2f64);
15143 v4i32 __builtin_msa_ftint_s_w (v4f32);
15144 v2i64 __builtin_msa_ftint_s_d (v2f64);
15146 v4u32 __builtin_msa_ftint_u_w (v4f32);
15147 v2u64 __builtin_msa_ftint_u_d (v2f64);
15149 v8i16 __builtin_msa_ftq_h (v4f32, v4f32);
15150 v4i32 __builtin_msa_ftq_w (v2f64, v2f64);
15152 v4i32 __builtin_msa_ftrunc_s_w (v4f32);
15153 v2i64 __builtin_msa_ftrunc_s_d (v2f64);
15155 v4u32 __builtin_msa_ftrunc_u_w (v4f32);
15156 v2u64 __builtin_msa_ftrunc_u_d (v2f64);
15158 v8i16 __builtin_msa_hadd_s_h (v16i8, v16i8);
15159 v4i32 __builtin_msa_hadd_s_w (v8i16, v8i16);
15160 v2i64 __builtin_msa_hadd_s_d (v4i32, v4i32);
15162 v8u16 __builtin_msa_hadd_u_h (v16u8, v16u8);
15163 v4u32 __builtin_msa_hadd_u_w (v8u16, v8u16);
15164 v2u64 __builtin_msa_hadd_u_d (v4u32, v4u32);
15166 v8i16 __builtin_msa_hsub_s_h (v16i8, v16i8);
15167 v4i32 __builtin_msa_hsub_s_w (v8i16, v8i16);
15168 v2i64 __builtin_msa_hsub_s_d (v4i32, v4i32);
15170 v8i16 __builtin_msa_hsub_u_h (v16u8, v16u8);
15171 v4i32 __builtin_msa_hsub_u_w (v8u16, v8u16);
15172 v2i64 __builtin_msa_hsub_u_d (v4u32, v4u32);
15174 v16i8 __builtin_msa_ilvev_b (v16i8, v16i8);
15175 v8i16 __builtin_msa_ilvev_h (v8i16, v8i16);
15176 v4i32 __builtin_msa_ilvev_w (v4i32, v4i32);
15177 v2i64 __builtin_msa_ilvev_d (v2i64, v2i64);
15179 v16i8 __builtin_msa_ilvl_b (v16i8, v16i8);
15180 v8i16 __builtin_msa_ilvl_h (v8i16, v8i16);
15181 v4i32 __builtin_msa_ilvl_w (v4i32, v4i32);
15182 v2i64 __builtin_msa_ilvl_d (v2i64, v2i64);
15184 v16i8 __builtin_msa_ilvod_b (v16i8, v16i8);
15185 v8i16 __builtin_msa_ilvod_h (v8i16, v8i16);
15186 v4i32 __builtin_msa_ilvod_w (v4i32, v4i32);
15187 v2i64 __builtin_msa_ilvod_d (v2i64, v2i64);
15189 v16i8 __builtin_msa_ilvr_b (v16i8, v16i8);
15190 v8i16 __builtin_msa_ilvr_h (v8i16, v8i16);
15191 v4i32 __builtin_msa_ilvr_w (v4i32, v4i32);
15192 v2i64 __builtin_msa_ilvr_d (v2i64, v2i64);
15194 v16i8 __builtin_msa_insert_b (v16i8, imm0_15, i32);
15195 v8i16 __builtin_msa_insert_h (v8i16, imm0_7, i32);
15196 v4i32 __builtin_msa_insert_w (v4i32, imm0_3, i32);
15197 v2i64 __builtin_msa_insert_d (v2i64, imm0_1, i64);
15199 v16i8 __builtin_msa_insve_b (v16i8, imm0_15, v16i8);
15200 v8i16 __builtin_msa_insve_h (v8i16, imm0_7, v8i16);
15201 v4i32 __builtin_msa_insve_w (v4i32, imm0_3, v4i32);
15202 v2i64 __builtin_msa_insve_d (v2i64, imm0_1, v2i64);
15204 v16i8 __builtin_msa_ld_b (void *, imm_n512_511);
15205 v8i16 __builtin_msa_ld_h (void *, imm_n1024_1022);
15206 v4i32 __builtin_msa_ld_w (void *, imm_n2048_2044);
15207 v2i64 __builtin_msa_ld_d (void *, imm_n4096_4088);
15209 v16i8 __builtin_msa_ldi_b (imm_n512_511);
15210 v8i16 __builtin_msa_ldi_h (imm_n512_511);
15211 v4i32 __builtin_msa_ldi_w (imm_n512_511);
15212 v2i64 __builtin_msa_ldi_d (imm_n512_511);
15214 v8i16 __builtin_msa_madd_q_h (v8i16, v8i16, v8i16);
15215 v4i32 __builtin_msa_madd_q_w (v4i32, v4i32, v4i32);
15217 v8i16 __builtin_msa_maddr_q_h (v8i16, v8i16, v8i16);
15218 v4i32 __builtin_msa_maddr_q_w (v4i32, v4i32, v4i32);
15220 v16i8 __builtin_msa_maddv_b (v16i8, v16i8, v16i8);
15221 v8i16 __builtin_msa_maddv_h (v8i16, v8i16, v8i16);
15222 v4i32 __builtin_msa_maddv_w (v4i32, v4i32, v4i32);
15223 v2i64 __builtin_msa_maddv_d (v2i64, v2i64, v2i64);
15225 v16i8 __builtin_msa_max_a_b (v16i8, v16i8);
15226 v8i16 __builtin_msa_max_a_h (v8i16, v8i16);
15227 v4i32 __builtin_msa_max_a_w (v4i32, v4i32);
15228 v2i64 __builtin_msa_max_a_d (v2i64, v2i64);
15230 v16i8 __builtin_msa_max_s_b (v16i8, v16i8);
15231 v8i16 __builtin_msa_max_s_h (v8i16, v8i16);
15232 v4i32 __builtin_msa_max_s_w (v4i32, v4i32);
15233 v2i64 __builtin_msa_max_s_d (v2i64, v2i64);
15235 v16u8 __builtin_msa_max_u_b (v16u8, v16u8);
15236 v8u16 __builtin_msa_max_u_h (v8u16, v8u16);
15237 v4u32 __builtin_msa_max_u_w (v4u32, v4u32);
15238 v2u64 __builtin_msa_max_u_d (v2u64, v2u64);
15240 v16i8 __builtin_msa_maxi_s_b (v16i8, imm_n16_15);
15241 v8i16 __builtin_msa_maxi_s_h (v8i16, imm_n16_15);
15242 v4i32 __builtin_msa_maxi_s_w (v4i32, imm_n16_15);
15243 v2i64 __builtin_msa_maxi_s_d (v2i64, imm_n16_15);
15245 v16u8 __builtin_msa_maxi_u_b (v16u8, imm0_31);
15246 v8u16 __builtin_msa_maxi_u_h (v8u16, imm0_31);
15247 v4u32 __builtin_msa_maxi_u_w (v4u32, imm0_31);
15248 v2u64 __builtin_msa_maxi_u_d (v2u64, imm0_31);
15250 v16i8 __builtin_msa_min_a_b (v16i8, v16i8);
15251 v8i16 __builtin_msa_min_a_h (v8i16, v8i16);
15252 v4i32 __builtin_msa_min_a_w (v4i32, v4i32);
15253 v2i64 __builtin_msa_min_a_d (v2i64, v2i64);
15255 v16i8 __builtin_msa_min_s_b (v16i8, v16i8);
15256 v8i16 __builtin_msa_min_s_h (v8i16, v8i16);
15257 v4i32 __builtin_msa_min_s_w (v4i32, v4i32);
15258 v2i64 __builtin_msa_min_s_d (v2i64, v2i64);
15260 v16u8 __builtin_msa_min_u_b (v16u8, v16u8);
15261 v8u16 __builtin_msa_min_u_h (v8u16, v8u16);
15262 v4u32 __builtin_msa_min_u_w (v4u32, v4u32);
15263 v2u64 __builtin_msa_min_u_d (v2u64, v2u64);
15265 v16i8 __builtin_msa_mini_s_b (v16i8, imm_n16_15);
15266 v8i16 __builtin_msa_mini_s_h (v8i16, imm_n16_15);
15267 v4i32 __builtin_msa_mini_s_w (v4i32, imm_n16_15);
15268 v2i64 __builtin_msa_mini_s_d (v2i64, imm_n16_15);
15270 v16u8 __builtin_msa_mini_u_b (v16u8, imm0_31);
15271 v8u16 __builtin_msa_mini_u_h (v8u16, imm0_31);
15272 v4u32 __builtin_msa_mini_u_w (v4u32, imm0_31);
15273 v2u64 __builtin_msa_mini_u_d (v2u64, imm0_31);
15275 v16i8 __builtin_msa_mod_s_b (v16i8, v16i8);
15276 v8i16 __builtin_msa_mod_s_h (v8i16, v8i16);
15277 v4i32 __builtin_msa_mod_s_w (v4i32, v4i32);
15278 v2i64 __builtin_msa_mod_s_d (v2i64, v2i64);
15280 v16u8 __builtin_msa_mod_u_b (v16u8, v16u8);
15281 v8u16 __builtin_msa_mod_u_h (v8u16, v8u16);
15282 v4u32 __builtin_msa_mod_u_w (v4u32, v4u32);
15283 v2u64 __builtin_msa_mod_u_d (v2u64, v2u64);
15285 v16i8 __builtin_msa_move_v (v16i8);
15287 v8i16 __builtin_msa_msub_q_h (v8i16, v8i16, v8i16);
15288 v4i32 __builtin_msa_msub_q_w (v4i32, v4i32, v4i32);
15290 v8i16 __builtin_msa_msubr_q_h (v8i16, v8i16, v8i16);
15291 v4i32 __builtin_msa_msubr_q_w (v4i32, v4i32, v4i32);
15293 v16i8 __builtin_msa_msubv_b (v16i8, v16i8, v16i8);
15294 v8i16 __builtin_msa_msubv_h (v8i16, v8i16, v8i16);
15295 v4i32 __builtin_msa_msubv_w (v4i32, v4i32, v4i32);
15296 v2i64 __builtin_msa_msubv_d (v2i64, v2i64, v2i64);
15298 v8i16 __builtin_msa_mul_q_h (v8i16, v8i16);
15299 v4i32 __builtin_msa_mul_q_w (v4i32, v4i32);
15301 v8i16 __builtin_msa_mulr_q_h (v8i16, v8i16);
15302 v4i32 __builtin_msa_mulr_q_w (v4i32, v4i32);
15304 v16i8 __builtin_msa_mulv_b (v16i8, v16i8);
15305 v8i16 __builtin_msa_mulv_h (v8i16, v8i16);
15306 v4i32 __builtin_msa_mulv_w (v4i32, v4i32);
15307 v2i64 __builtin_msa_mulv_d (v2i64, v2i64);
15309 v16i8 __builtin_msa_nloc_b (v16i8);
15310 v8i16 __builtin_msa_nloc_h (v8i16);
15311 v4i32 __builtin_msa_nloc_w (v4i32);
15312 v2i64 __builtin_msa_nloc_d (v2i64);
15314 v16i8 __builtin_msa_nlzc_b (v16i8);
15315 v8i16 __builtin_msa_nlzc_h (v8i16);
15316 v4i32 __builtin_msa_nlzc_w (v4i32);
15317 v2i64 __builtin_msa_nlzc_d (v2i64);
15319 v16u8 __builtin_msa_nor_v (v16u8, v16u8);
15321 v16u8 __builtin_msa_nori_b (v16u8, imm0_255);
15323 v16u8 __builtin_msa_or_v (v16u8, v16u8);
15325 v16u8 __builtin_msa_ori_b (v16u8, imm0_255);
15327 v16i8 __builtin_msa_pckev_b (v16i8, v16i8);
15328 v8i16 __builtin_msa_pckev_h (v8i16, v8i16);
15329 v4i32 __builtin_msa_pckev_w (v4i32, v4i32);
15330 v2i64 __builtin_msa_pckev_d (v2i64, v2i64);
15332 v16i8 __builtin_msa_pckod_b (v16i8, v16i8);
15333 v8i16 __builtin_msa_pckod_h (v8i16, v8i16);
15334 v4i32 __builtin_msa_pckod_w (v4i32, v4i32);
15335 v2i64 __builtin_msa_pckod_d (v2i64, v2i64);
15337 v16i8 __builtin_msa_pcnt_b (v16i8);
15338 v8i16 __builtin_msa_pcnt_h (v8i16);
15339 v4i32 __builtin_msa_pcnt_w (v4i32);
15340 v2i64 __builtin_msa_pcnt_d (v2i64);
15342 v16i8 __builtin_msa_sat_s_b (v16i8, imm0_7);
15343 v8i16 __builtin_msa_sat_s_h (v8i16, imm0_15);
15344 v4i32 __builtin_msa_sat_s_w (v4i32, imm0_31);
15345 v2i64 __builtin_msa_sat_s_d (v2i64, imm0_63);
15347 v16u8 __builtin_msa_sat_u_b (v16u8, imm0_7);
15348 v8u16 __builtin_msa_sat_u_h (v8u16, imm0_15);
15349 v4u32 __builtin_msa_sat_u_w (v4u32, imm0_31);
15350 v2u64 __builtin_msa_sat_u_d (v2u64, imm0_63);
15352 v16i8 __builtin_msa_shf_b (v16i8, imm0_255);
15353 v8i16 __builtin_msa_shf_h (v8i16, imm0_255);
15354 v4i32 __builtin_msa_shf_w (v4i32, imm0_255);
15356 v16i8 __builtin_msa_sld_b (v16i8, v16i8, i32);
15357 v8i16 __builtin_msa_sld_h (v8i16, v8i16, i32);
15358 v4i32 __builtin_msa_sld_w (v4i32, v4i32, i32);
15359 v2i64 __builtin_msa_sld_d (v2i64, v2i64, i32);
15361 v16i8 __builtin_msa_sldi_b (v16i8, v16i8, imm0_15);
15362 v8i16 __builtin_msa_sldi_h (v8i16, v8i16, imm0_7);
15363 v4i32 __builtin_msa_sldi_w (v4i32, v4i32, imm0_3);
15364 v2i64 __builtin_msa_sldi_d (v2i64, v2i64, imm0_1);
15366 v16i8 __builtin_msa_sll_b (v16i8, v16i8);
15367 v8i16 __builtin_msa_sll_h (v8i16, v8i16);
15368 v4i32 __builtin_msa_sll_w (v4i32, v4i32);
15369 v2i64 __builtin_msa_sll_d (v2i64, v2i64);
15371 v16i8 __builtin_msa_slli_b (v16i8, imm0_7);
15372 v8i16 __builtin_msa_slli_h (v8i16, imm0_15);
15373 v4i32 __builtin_msa_slli_w (v4i32, imm0_31);
15374 v2i64 __builtin_msa_slli_d (v2i64, imm0_63);
15376 v16i8 __builtin_msa_splat_b (v16i8, i32);
15377 v8i16 __builtin_msa_splat_h (v8i16, i32);
15378 v4i32 __builtin_msa_splat_w (v4i32, i32);
15379 v2i64 __builtin_msa_splat_d (v2i64, i32);
15381 v16i8 __builtin_msa_splati_b (v16i8, imm0_15);
15382 v8i16 __builtin_msa_splati_h (v8i16, imm0_7);
15383 v4i32 __builtin_msa_splati_w (v4i32, imm0_3);
15384 v2i64 __builtin_msa_splati_d (v2i64, imm0_1);
15386 v16i8 __builtin_msa_sra_b (v16i8, v16i8);
15387 v8i16 __builtin_msa_sra_h (v8i16, v8i16);
15388 v4i32 __builtin_msa_sra_w (v4i32, v4i32);
15389 v2i64 __builtin_msa_sra_d (v2i64, v2i64);
15391 v16i8 __builtin_msa_srai_b (v16i8, imm0_7);
15392 v8i16 __builtin_msa_srai_h (v8i16, imm0_15);
15393 v4i32 __builtin_msa_srai_w (v4i32, imm0_31);
15394 v2i64 __builtin_msa_srai_d (v2i64, imm0_63);
15396 v16i8 __builtin_msa_srar_b (v16i8, v16i8);
15397 v8i16 __builtin_msa_srar_h (v8i16, v8i16);
15398 v4i32 __builtin_msa_srar_w (v4i32, v4i32);
15399 v2i64 __builtin_msa_srar_d (v2i64, v2i64);
15401 v16i8 __builtin_msa_srari_b (v16i8, imm0_7);
15402 v8i16 __builtin_msa_srari_h (v8i16, imm0_15);
15403 v4i32 __builtin_msa_srari_w (v4i32, imm0_31);
15404 v2i64 __builtin_msa_srari_d (v2i64, imm0_63);
15406 v16i8 __builtin_msa_srl_b (v16i8, v16i8);
15407 v8i16 __builtin_msa_srl_h (v8i16, v8i16);
15408 v4i32 __builtin_msa_srl_w (v4i32, v4i32);
15409 v2i64 __builtin_msa_srl_d (v2i64, v2i64);
15411 v16i8 __builtin_msa_srli_b (v16i8, imm0_7);
15412 v8i16 __builtin_msa_srli_h (v8i16, imm0_15);
15413 v4i32 __builtin_msa_srli_w (v4i32, imm0_31);
15414 v2i64 __builtin_msa_srli_d (v2i64, imm0_63);
15416 v16i8 __builtin_msa_srlr_b (v16i8, v16i8);
15417 v8i16 __builtin_msa_srlr_h (v8i16, v8i16);
15418 v4i32 __builtin_msa_srlr_w (v4i32, v4i32);
15419 v2i64 __builtin_msa_srlr_d (v2i64, v2i64);
15421 v16i8 __builtin_msa_srlri_b (v16i8, imm0_7);
15422 v8i16 __builtin_msa_srlri_h (v8i16, imm0_15);
15423 v4i32 __builtin_msa_srlri_w (v4i32, imm0_31);
15424 v2i64 __builtin_msa_srlri_d (v2i64, imm0_63);
15426 void __builtin_msa_st_b (v16i8, void *, imm_n512_511);
15427 void __builtin_msa_st_h (v8i16, void *, imm_n1024_1022);
15428 void __builtin_msa_st_w (v4i32, void *, imm_n2048_2044);
15429 void __builtin_msa_st_d (v2i64, void *, imm_n4096_4088);
15431 v16i8 __builtin_msa_subs_s_b (v16i8, v16i8);
15432 v8i16 __builtin_msa_subs_s_h (v8i16, v8i16);
15433 v4i32 __builtin_msa_subs_s_w (v4i32, v4i32);
15434 v2i64 __builtin_msa_subs_s_d (v2i64, v2i64);
15436 v16u8 __builtin_msa_subs_u_b (v16u8, v16u8);
15437 v8u16 __builtin_msa_subs_u_h (v8u16, v8u16);
15438 v4u32 __builtin_msa_subs_u_w (v4u32, v4u32);
15439 v2u64 __builtin_msa_subs_u_d (v2u64, v2u64);
15441 v16u8 __builtin_msa_subsus_u_b (v16u8, v16i8);
15442 v8u16 __builtin_msa_subsus_u_h (v8u16, v8i16);
15443 v4u32 __builtin_msa_subsus_u_w (v4u32, v4i32);
15444 v2u64 __builtin_msa_subsus_u_d (v2u64, v2i64);
15446 v16i8 __builtin_msa_subsuu_s_b (v16u8, v16u8);
15447 v8i16 __builtin_msa_subsuu_s_h (v8u16, v8u16);
15448 v4i32 __builtin_msa_subsuu_s_w (v4u32, v4u32);
15449 v2i64 __builtin_msa_subsuu_s_d (v2u64, v2u64);
15451 v16i8 __builtin_msa_subv_b (v16i8, v16i8);
15452 v8i16 __builtin_msa_subv_h (v8i16, v8i16);
15453 v4i32 __builtin_msa_subv_w (v4i32, v4i32);
15454 v2i64 __builtin_msa_subv_d (v2i64, v2i64);
15456 v16i8 __builtin_msa_subvi_b (v16i8, imm0_31);
15457 v8i16 __builtin_msa_subvi_h (v8i16, imm0_31);
15458 v4i32 __builtin_msa_subvi_w (v4i32, imm0_31);
15459 v2i64 __builtin_msa_subvi_d (v2i64, imm0_31);
15461 v16i8 __builtin_msa_vshf_b (v16i8, v16i8, v16i8);
15462 v8i16 __builtin_msa_vshf_h (v8i16, v8i16, v8i16);
15463 v4i32 __builtin_msa_vshf_w (v4i32, v4i32, v4i32);
15464 v2i64 __builtin_msa_vshf_d (v2i64, v2i64, v2i64);
15466 v16u8 __builtin_msa_xor_v (v16u8, v16u8);
15468 v16u8 __builtin_msa_xori_b (v16u8, imm0_255);
15471 @node Other MIPS Built-in Functions
15472 @subsection Other MIPS Built-in Functions
15474 GCC provides other MIPS-specific built-in functions:
15477 @item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr})
15478 Insert a @samp{cache} instruction with operands @var{op} and @var{addr}.
15479 GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE}
15480 when this function is available.
15482 @item unsigned int __builtin_mips_get_fcsr (void)
15483 @itemx void __builtin_mips_set_fcsr (unsigned int @var{value})
15484 Get and set the contents of the floating-point control and status register
15485 (FPU control register 31). These functions are only available in hard-float
15486 code but can be called in both MIPS16 and non-MIPS16 contexts.
15488 @code{__builtin_mips_set_fcsr} can be used to change any bit of the
15489 register except the condition codes, which GCC assumes are preserved.
15492 @node MSP430 Built-in Functions
15493 @subsection MSP430 Built-in Functions
15495 GCC provides a couple of special builtin functions to aid in the
15496 writing of interrupt handlers in C.
15499 @item __bic_SR_register_on_exit (int @var{mask})
15500 This clears the indicated bits in the saved copy of the status register
15501 currently residing on the stack. This only works inside interrupt
15502 handlers and the changes to the status register will only take affect
15503 once the handler returns.
15505 @item __bis_SR_register_on_exit (int @var{mask})
15506 This sets the indicated bits in the saved copy of the status register
15507 currently residing on the stack. This only works inside interrupt
15508 handlers and the changes to the status register will only take affect
15509 once the handler returns.
15511 @item __delay_cycles (long long @var{cycles})
15512 This inserts an instruction sequence that takes exactly @var{cycles}
15513 cycles (between 0 and about 17E9) to complete. The inserted sequence
15514 may use jumps, loops, or no-ops, and does not interfere with any other
15515 instructions. Note that @var{cycles} must be a compile-time constant
15516 integer - that is, you must pass a number, not a variable that may be
15517 optimized to a constant later. The number of cycles delayed by this
15521 @node NDS32 Built-in Functions
15522 @subsection NDS32 Built-in Functions
15524 These built-in functions are available for the NDS32 target:
15526 @deftypefn {Built-in Function} void __builtin_nds32_isync (int *@var{addr})
15527 Insert an ISYNC instruction into the instruction stream where
15528 @var{addr} is an instruction address for serialization.
15531 @deftypefn {Built-in Function} void __builtin_nds32_isb (void)
15532 Insert an ISB instruction into the instruction stream.
15535 @deftypefn {Built-in Function} int __builtin_nds32_mfsr (int @var{sr})
15536 Return the content of a system register which is mapped by @var{sr}.
15539 @deftypefn {Built-in Function} int __builtin_nds32_mfusr (int @var{usr})
15540 Return the content of a user space register which is mapped by @var{usr}.
15543 @deftypefn {Built-in Function} void __builtin_nds32_mtsr (int @var{value}, int @var{sr})
15544 Move the @var{value} to a system register which is mapped by @var{sr}.
15547 @deftypefn {Built-in Function} void __builtin_nds32_mtusr (int @var{value}, int @var{usr})
15548 Move the @var{value} to a user space register which is mapped by @var{usr}.
15551 @deftypefn {Built-in Function} void __builtin_nds32_setgie_en (void)
15552 Enable global interrupt.
15555 @deftypefn {Built-in Function} void __builtin_nds32_setgie_dis (void)
15556 Disable global interrupt.
15559 @node picoChip Built-in Functions
15560 @subsection picoChip Built-in Functions
15562 GCC provides an interface to selected machine instructions from the
15563 picoChip instruction set.
15566 @item int __builtin_sbc (int @var{value})
15567 Sign bit count. Return the number of consecutive bits in @var{value}
15568 that have the same value as the sign bit. The result is the number of
15569 leading sign bits minus one, giving the number of redundant sign bits in
15572 @item int __builtin_byteswap (int @var{value})
15573 Byte swap. Return the result of swapping the upper and lower bytes of
15576 @item int __builtin_brev (int @var{value})
15577 Bit reversal. Return the result of reversing the bits in
15578 @var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1,
15581 @item int __builtin_adds (int @var{x}, int @var{y})
15582 Saturating addition. Return the result of adding @var{x} and @var{y},
15583 storing the value 32767 if the result overflows.
15585 @item int __builtin_subs (int @var{x}, int @var{y})
15586 Saturating subtraction. Return the result of subtracting @var{y} from
15587 @var{x}, storing the value @minus{}32768 if the result overflows.
15589 @item void __builtin_halt (void)
15590 Halt. The processor stops execution. This built-in is useful for
15591 implementing assertions.
15595 @node Basic PowerPC Built-in Functions
15596 @subsection Basic PowerPC Built-in Functions
15599 * Basic PowerPC Built-in Functions Available on all Configurations::
15600 * Basic PowerPC Built-in Functions Available on ISA 2.05::
15601 * Basic PowerPC Built-in Functions Available on ISA 2.06::
15602 * Basic PowerPC Built-in Functions Available on ISA 2.07::
15603 * Basic PowerPC Built-in Functions Available on ISA 3.0::
15606 This section describes PowerPC built-in functions that do not require
15607 the inclusion of any special header files to declare prototypes or
15608 provide macro definitions. The sections that follow describe
15609 additional PowerPC built-in functions.
15611 @node Basic PowerPC Built-in Functions Available on all Configurations
15612 @subsubsection Basic PowerPC Built-in Functions Available on all Configurations
15614 @deftypefn {Built-in Function} void __builtin_cpu_init (void)
15615 This function is a @code{nop} on the PowerPC platform and is included solely
15616 to maintain API compatibility with the x86 builtins.
15619 @deftypefn {Built-in Function} int __builtin_cpu_is (const char *@var{cpuname})
15620 This function returns a value of @code{1} if the run-time CPU is of type
15621 @var{cpuname} and returns @code{0} otherwise
15623 The @code{__builtin_cpu_is} function requires GLIBC 2.23 or newer
15624 which exports the hardware capability bits. GCC defines the macro
15625 @code{__BUILTIN_CPU_SUPPORTS__} if the @code{__builtin_cpu_supports}
15626 built-in function is fully supported.
15628 If GCC was configured to use a GLIBC before 2.23, the built-in
15629 function @code{__builtin_cpu_is} always returns a 0 and the compiler
15632 The following CPU names can be detected:
15636 IBM POWER9 Server CPU.
15638 IBM POWER8 Server CPU.
15640 IBM POWER7 Server CPU.
15642 IBM POWER6 Server CPU (RAW mode).
15644 IBM POWER6 Server CPU (Architected mode).
15646 IBM POWER5+ Server CPU.
15648 IBM POWER5 Server CPU.
15650 IBM 970 Server CPU (ie, Apple G5).
15652 IBM POWER4 Server CPU.
15654 IBM A2 64-bit Embedded CPU
15656 IBM PowerPC 476FP 32-bit Embedded CPU.
15658 IBM PowerPC 464 32-bit Embedded CPU.
15660 PowerPC 440 32-bit Embedded CPU.
15662 PowerPC 405 32-bit Embedded CPU.
15664 IBM PowerPC Cell Broadband Engine Architecture CPU.
15667 Here is an example:
15669 #ifdef __BUILTIN_CPU_SUPPORTS__
15670 if (__builtin_cpu_is ("power8"))
15672 do_power8 (); // POWER8 specific implementation.
15677 do_generic (); // Generic implementation.
15682 @deftypefn {Built-in Function} int __builtin_cpu_supports (const char *@var{feature})
15683 This function returns a value of @code{1} if the run-time CPU supports the HWCAP
15684 feature @var{feature} and returns @code{0} otherwise.
15686 The @code{__builtin_cpu_supports} function requires GLIBC 2.23 or
15687 newer which exports the hardware capability bits. GCC defines the
15688 macro @code{__BUILTIN_CPU_SUPPORTS__} if the
15689 @code{__builtin_cpu_supports} built-in function is fully supported.
15691 If GCC was configured to use a GLIBC before 2.23, the built-in
15692 function @code{__builtin_cpu_suports} always returns a 0 and the
15693 compiler issues a warning.
15695 The following features can be
15700 4xx CPU has a Multiply Accumulator.
15702 CPU has a SIMD/Vector Unit.
15704 CPU supports ISA 2.05 (eg, POWER6)
15706 CPU supports ISA 2.06 (eg, POWER7)
15708 CPU supports ISA 2.07 (eg, POWER8)
15710 CPU supports ISA 3.0 (eg, POWER9)
15712 CPU supports the set of compatible performance monitoring events.
15714 CPU supports the Embedded ISA category.
15716 CPU has a CELL broadband engine.
15718 CPU supports the @code{darn} (deliver a random number) instruction.
15720 CPU has a decimal floating point unit.
15722 CPU supports the data stream control register.
15724 CPU supports event base branching.
15726 CPU has a SPE double precision floating point unit.
15728 CPU has a SPE single precision floating point unit.
15730 CPU has a floating point unit.
15732 CPU has hardware transaction memory instructions.
15734 Kernel aborts hardware transactions when a syscall is made.
15735 @item htm-no-suspend
15736 CPU supports hardware transaction memory but does not support the
15737 @code{tsuspend.} instruction.
15739 CPU supports icache snooping capabilities.
15741 CPU supports 128-bit IEEE binary floating point instructions.
15743 CPU supports the integer select instruction.
15745 CPU has a memory management unit.
15747 CPU does not have a timebase (eg, 601 and 403gx).
15749 CPU supports the PA Semi 6T CORE ISA.
15751 CPU supports ISA 2.00 (eg, POWER4)
15753 CPU supports ISA 2.02 (eg, POWER5)
15755 CPU supports ISA 2.03 (eg, POWER5+)
15757 CPU supports ISA 2.05 (eg, POWER6) extended opcodes mffgpr and mftgpr.
15759 CPU supports 32-bit mode execution.
15761 CPU supports the old POWER ISA (eg, 601)
15763 CPU supports 64-bit mode execution.
15765 CPU supports a little-endian mode that uses address swizzling.
15767 Kernel supports system call vectored.
15769 CPU support simultaneous multi-threading.
15771 CPU has a signal processing extension unit.
15773 CPU supports the target address register.
15775 CPU supports true little-endian mode.
15777 CPU has unified I/D cache.
15779 CPU supports the vector cryptography instructions.
15781 CPU supports the vector-scalar extension.
15784 Here is an example:
15786 #ifdef __BUILTIN_CPU_SUPPORTS__
15787 if (__builtin_cpu_supports ("fpu"))
15789 asm("fadd %0,%1,%2" : "=d"(dst) : "d"(src1), "d"(src2));
15794 dst = __fadd (src1, src2); // Software FP addition function.
15799 The following built-in functions are also available on all PowerPC
15802 uint64_t __builtin_ppc_get_timebase ();
15803 unsigned long __builtin_ppc_mftb ();
15806 The @code{__builtin_ppc_get_timebase} and @code{__builtin_ppc_mftb}
15807 functions generate instructions to read the Time Base Register. The
15808 @code{__builtin_ppc_get_timebase} function may generate multiple
15809 instructions and always returns the 64 bits of the Time Base Register.
15810 The @code{__builtin_ppc_mftb} function always generates one instruction and
15811 returns the Time Base Register value as an unsigned long, throwing away
15812 the most significant word on 32-bit environments.
15814 @node Basic PowerPC Built-in Functions Available on ISA 2.05
15815 @subsubsection Basic PowerPC Built-in Functions Available on ISA 2.05
15817 The basic built-in functions described in this section are
15818 available on the PowerPC family of processors starting with ISA 2.05
15819 or later. Unless specific options are explicitly disabled on the
15820 command line, specifying option @option{-mcpu=power6} has the effect of
15821 enabling the @option{-mpowerpc64}, @option{-mpowerpc-gpopt},
15822 @option{-mpowerpc-gfxopt}, @option{-mmfcrf}, @option{-mpopcntb},
15823 @option{-mfprnd}, @option{-mcmpb}, @option{-mhard-dfp}, and
15824 @option{-mrecip-precision} options. Specify the
15825 @option{-maltivec} and @option{-mfpgpr} options explicitly in
15826 combination with the above options if they are desired.
15828 The following functions require option @option{-mcmpb}.
15830 unsigned long long __builtin_cmpb (unsigned long long int, unsigned long long int);
15831 unsigned int __builtin_cmpb (unsigned int, unsigned int);
15834 The @code{__builtin_cmpb} function
15835 performs a byte-wise compare on the contents of its two arguments,
15836 returning the result of the byte-wise comparison as the returned
15837 value. For each byte comparison, the corresponding byte of the return
15838 value holds 0xff if the input bytes are equal and 0 if the input bytes
15839 are not equal. If either of the arguments to this built-in function
15840 is wider than 32 bits, the function call expands into the form that
15841 expects @code{unsigned long long int} arguments
15842 which is only available on 64-bit targets.
15844 The following built-in functions are available
15845 when hardware decimal floating point
15846 (@option{-mhard-dfp}) is available:
15848 _Decimal64 __builtin_ddedpd (int, _Decimal64);
15849 _Decimal128 __builtin_ddedpdq (int, _Decimal128);
15850 _Decimal64 __builtin_denbcd (int, _Decimal64);
15851 _Decimal128 __builtin_denbcdq (int, _Decimal128);
15852 _Decimal64 __builtin_diex (long long, _Decimal64);
15853 _Decimal128 _builtin_diexq (long long, _Decimal128);
15854 _Decimal64 __builtin_dscli (_Decimal64, int);
15855 _Decimal128 __builtin_dscliq (_Decimal128, int);
15856 _Decimal64 __builtin_dscri (_Decimal64, int);
15857 _Decimal128 __builtin_dscriq (_Decimal128, int);
15858 long long __builtin_dxex (_Decimal64);
15859 long long __builtin_dxexq (_Decimal128);
15860 _Decimal128 __builtin_pack_dec128 (unsigned long long, unsigned long long);
15861 unsigned long long __builtin_unpack_dec128 (_Decimal128, int);
15864 The following functions require @option{-mhard-float},
15865 @option{-mpowerpc-gfxopt}, and @option{-mpopcntb} options.
15868 double __builtin_recipdiv (double, double);
15869 float __builtin_recipdivf (float, float);
15870 double __builtin_rsqrt (double);
15871 float __builtin_rsqrtf (float);
15874 The @code{vec_rsqrt}, @code{__builtin_rsqrt}, and
15875 @code{__builtin_rsqrtf} functions generate multiple instructions to
15876 implement the reciprocal sqrt functionality using reciprocal sqrt
15877 estimate instructions.
15879 The @code{__builtin_recipdiv}, and @code{__builtin_recipdivf}
15880 functions generate multiple instructions to implement division using
15881 the reciprocal estimate instructions.
15883 The following functions require @option{-mhard-float} and
15884 @option{-mmultiple} options.
15887 long double __builtin_pack_longdouble (double, double);
15888 double __builtin_unpack_longdouble (long double, int);
15891 @node Basic PowerPC Built-in Functions Available on ISA 2.06
15892 @subsubsection Basic PowerPC Built-in Functions Available on ISA 2.06
15894 The basic built-in functions described in this section are
15895 available on the PowerPC family of processors starting with ISA 2.05
15896 or later. Unless specific options are explicitly disabled on the
15897 command line, specifying option @option{-mcpu=power7} has the effect of
15898 enabling all the same options as for @option{-mcpu=power6} in
15899 addition to the @option{-maltivec}, @option{-mpopcntd}, and
15900 @option{-mvsx} options.
15902 The following basic built-in functions require @option{-mpopcntd}:
15904 unsigned int __builtin_addg6s (unsigned int, unsigned int);
15905 long long __builtin_bpermd (long long, long long);
15906 unsigned int __builtin_cbcdtd (unsigned int);
15907 unsigned int __builtin_cdtbcd (unsigned int);
15908 long long __builtin_divde (long long, long long);
15909 unsigned long long __builtin_divdeu (unsigned long long, unsigned long long);
15910 int __builtin_divwe (int, int);
15911 unsigned int __builtin_divweu (unsigned int, unsigned int);
15912 vector __int128_t __builtin_pack_vector_int128 (long long, long long);
15913 void __builtin_rs6000_speculation_barrier (void);
15914 long long __builtin_unpack_vector_int128 (vector __int128_t, signed char);
15917 Of these, the @code{__builtin_divde} and @code{__builtin_divdeu} functions
15918 require a 64-bit environment.
15920 The following basic built-in functions, which are also supported on
15921 x86 targets, require @option{-mfloat128}.
15923 __float128 __builtin_fabsq (__float128);
15924 __float128 __builtin_copysignq (__float128, __float128);
15925 __float128 __builtin_infq (void);
15926 __float128 __builtin_huge_valq (void);
15927 __float128 __builtin_nanq (void);
15928 __float128 __builtin_nansq (void);
15930 __float128 __builtin_sqrtf128 (__float128);
15931 __float128 __builtin_fmaf128 (__float128, __float128, __float128);
15934 @node Basic PowerPC Built-in Functions Available on ISA 2.07
15935 @subsubsection Basic PowerPC Built-in Functions Available on ISA 2.07
15937 The basic built-in functions described in this section are
15938 available on the PowerPC family of processors starting with ISA 2.07
15939 or later. Unless specific options are explicitly disabled on the
15940 command line, specifying option @option{-mcpu=power8} has the effect of
15941 enabling all the same options as for @option{-mcpu=power7} in
15942 addition to the @option{-mpower8-fusion}, @option{-mpower8-vector},
15943 @option{-mcrypto}, @option{-mhtm}, @option{-mquad-memory}, and
15944 @option{-mquad-memory-atomic} options.
15946 This section intentionally empty.
15948 @node Basic PowerPC Built-in Functions Available on ISA 3.0
15949 @subsubsection Basic PowerPC Built-in Functions Available on ISA 3.0
15951 The basic built-in functions described in this section are
15952 available on the PowerPC family of processors starting with ISA 3.0
15953 or later. Unless specific options are explicitly disabled on the
15954 command line, specifying option @option{-mcpu=power9} has the effect of
15955 enabling all the same options as for @option{-mcpu=power8} in
15956 addition to the @option{-misel} option.
15958 The following built-in functions are available on Linux 64-bit systems
15959 that use the ISA 3.0 instruction set (@option{-mcpu=power9}):
15962 @item __float128 __builtin_addf128_round_to_odd (__float128, __float128)
15963 Perform a 128-bit IEEE floating point add using round to odd as the
15965 @findex __builtin_addf128_round_to_odd
15967 @item __float128 __builtin_subf128_round_to_odd (__float128, __float128)
15968 Perform a 128-bit IEEE floating point subtract using round to odd as
15970 @findex __builtin_subf128_round_to_odd
15972 @item __float128 __builtin_mulf128_round_to_odd (__float128, __float128)
15973 Perform a 128-bit IEEE floating point multiply using round to odd as
15975 @findex __builtin_mulf128_round_to_odd
15977 @item __float128 __builtin_divf128_round_to_odd (__float128, __float128)
15978 Perform a 128-bit IEEE floating point divide using round to odd as
15980 @findex __builtin_divf128_round_to_odd
15982 @item __float128 __builtin_sqrtf128_round_to_odd (__float128)
15983 Perform a 128-bit IEEE floating point square root using round to odd
15984 as the rounding mode.
15985 @findex __builtin_sqrtf128_round_to_odd
15987 @item __float128 __builtin_fmaf128_round_to_odd (__float128, __float128, __float128)
15988 Perform a 128-bit IEEE floating point fused multiply and add operation
15989 using round to odd as the rounding mode.
15990 @findex __builtin_fmaf128_round_to_odd
15992 @item double __builtin_truncf128_round_to_odd (__float128)
15993 Convert a 128-bit IEEE floating point value to @code{double} using
15994 round to odd as the rounding mode.
15995 @findex __builtin_truncf128_round_to_odd
15998 The following additional built-in functions are also available for the
15999 PowerPC family of processors, starting with ISA 3.0 or later:
16001 long long __builtin_darn (void);
16002 long long __builtin_darn_raw (void);
16003 int __builtin_darn_32 (void);
16006 The @code{__builtin_darn} and @code{__builtin_darn_raw}
16007 functions require a
16008 64-bit environment supporting ISA 3.0 or later.
16009 The @code{__builtin_darn} function provides a 64-bit conditioned
16010 random number. The @code{__builtin_darn_raw} function provides a
16011 64-bit raw random number. The @code{__builtin_darn_32} function
16012 provides a 32-bit conditioned random number.
16014 The following additional built-in functions are also available for the
16015 PowerPC family of processors, starting with ISA 3.0 or later:
16018 int __builtin_byte_in_set (unsigned char u, unsigned long long set);
16019 int __builtin_byte_in_range (unsigned char u, unsigned int range);
16020 int __builtin_byte_in_either_range (unsigned char u, unsigned int ranges);
16022 int __builtin_dfp_dtstsfi_lt (unsigned int comparison, _Decimal64 value);
16023 int __builtin_dfp_dtstsfi_lt (unsigned int comparison, _Decimal128 value);
16024 int __builtin_dfp_dtstsfi_lt_dd (unsigned int comparison, _Decimal64 value);
16025 int __builtin_dfp_dtstsfi_lt_td (unsigned int comparison, _Decimal128 value);
16027 int __builtin_dfp_dtstsfi_gt (unsigned int comparison, _Decimal64 value);
16028 int __builtin_dfp_dtstsfi_gt (unsigned int comparison, _Decimal128 value);
16029 int __builtin_dfp_dtstsfi_gt_dd (unsigned int comparison, _Decimal64 value);
16030 int __builtin_dfp_dtstsfi_gt_td (unsigned int comparison, _Decimal128 value);
16032 int __builtin_dfp_dtstsfi_eq (unsigned int comparison, _Decimal64 value);
16033 int __builtin_dfp_dtstsfi_eq (unsigned int comparison, _Decimal128 value);
16034 int __builtin_dfp_dtstsfi_eq_dd (unsigned int comparison, _Decimal64 value);
16035 int __builtin_dfp_dtstsfi_eq_td (unsigned int comparison, _Decimal128 value);
16037 int __builtin_dfp_dtstsfi_ov (unsigned int comparison, _Decimal64 value);
16038 int __builtin_dfp_dtstsfi_ov (unsigned int comparison, _Decimal128 value);
16039 int __builtin_dfp_dtstsfi_ov_dd (unsigned int comparison, _Decimal64 value);
16040 int __builtin_dfp_dtstsfi_ov_td (unsigned int comparison, _Decimal128 value);
16042 The @code{__builtin_byte_in_set} function requires a
16043 64-bit environment supporting ISA 3.0 or later. This function returns
16044 a non-zero value if and only if its @code{u} argument exactly equals one of
16045 the eight bytes contained within its 64-bit @code{set} argument.
16047 The @code{__builtin_byte_in_range} and
16048 @code{__builtin_byte_in_either_range} require an environment
16049 supporting ISA 3.0 or later. For these two functions, the
16050 @code{range} argument is encoded as 4 bytes, organized as
16051 @code{hi_1:lo_1:hi_2:lo_2}.
16052 The @code{__builtin_byte_in_range} function returns a
16053 non-zero value if and only if its @code{u} argument is within the
16054 range bounded between @code{lo_2} and @code{hi_2} inclusive.
16055 The @code{__builtin_byte_in_either_range} function returns non-zero if
16056 and only if its @code{u} argument is within either the range bounded
16057 between @code{lo_1} and @code{hi_1} inclusive or the range bounded
16058 between @code{lo_2} and @code{hi_2} inclusive.
16060 The @code{__builtin_dfp_dtstsfi_lt} function returns a non-zero value
16061 if and only if the number of signficant digits of its @code{value} argument
16062 is less than its @code{comparison} argument. The
16063 @code{__builtin_dfp_dtstsfi_lt_dd} and
16064 @code{__builtin_dfp_dtstsfi_lt_td} functions behave similarly, but
16065 require that the type of the @code{value} argument be
16066 @code{__Decimal64} and @code{__Decimal128} respectively.
16068 The @code{__builtin_dfp_dtstsfi_gt} function returns a non-zero value
16069 if and only if the number of signficant digits of its @code{value} argument
16070 is greater than its @code{comparison} argument. The
16071 @code{__builtin_dfp_dtstsfi_gt_dd} and
16072 @code{__builtin_dfp_dtstsfi_gt_td} functions behave similarly, but
16073 require that the type of the @code{value} argument be
16074 @code{__Decimal64} and @code{__Decimal128} respectively.
16076 The @code{__builtin_dfp_dtstsfi_eq} function returns a non-zero value
16077 if and only if the number of signficant digits of its @code{value} argument
16078 equals its @code{comparison} argument. The
16079 @code{__builtin_dfp_dtstsfi_eq_dd} and
16080 @code{__builtin_dfp_dtstsfi_eq_td} functions behave similarly, but
16081 require that the type of the @code{value} argument be
16082 @code{__Decimal64} and @code{__Decimal128} respectively.
16084 The @code{__builtin_dfp_dtstsfi_ov} function returns a non-zero value
16085 if and only if its @code{value} argument has an undefined number of
16086 significant digits, such as when @code{value} is an encoding of @code{NaN}.
16087 The @code{__builtin_dfp_dtstsfi_ov_dd} and
16088 @code{__builtin_dfp_dtstsfi_ov_td} functions behave similarly, but
16089 require that the type of the @code{value} argument be
16090 @code{__Decimal64} and @code{__Decimal128} respectively.
16094 @node PowerPC AltiVec/VSX Built-in Functions
16095 @subsection PowerPC AltiVec Built-in Functions
16097 GCC provides an interface for the PowerPC family of processors to access
16098 the AltiVec operations described in Motorola's AltiVec Programming
16099 Interface Manual. The interface is made available by including
16100 @code{<altivec.h>} and using @option{-maltivec} and
16101 @option{-mabi=altivec}. The interface supports the following vector
16105 vector unsigned char
16109 vector unsigned short
16110 vector signed short
16114 vector unsigned int
16120 If @option{-mvsx} is used the following additional vector types are
16124 vector unsigned long
16129 The long types are only implemented for 64-bit code generation, and
16130 the long type is only used in the floating point/integer conversion
16133 GCC's implementation of the high-level language interface available from
16134 C and C++ code differs from Motorola's documentation in several ways.
16139 A vector constant is a list of constant expressions within curly braces.
16142 A vector initializer requires no cast if the vector constant is of the
16143 same type as the variable it is initializing.
16146 If @code{signed} or @code{unsigned} is omitted, the signedness of the
16147 vector type is the default signedness of the base type. The default
16148 varies depending on the operating system, so a portable program should
16149 always specify the signedness.
16152 Compiling with @option{-maltivec} adds keywords @code{__vector},
16153 @code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and
16154 @code{bool}. When compiling ISO C, the context-sensitive substitution
16155 of the keywords @code{vector}, @code{pixel} and @code{bool} is
16156 disabled. To use them, you must include @code{<altivec.h>} instead.
16159 GCC allows using a @code{typedef} name as the type specifier for a
16163 For C, overloaded functions are implemented with macros so the following
16167 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
16171 Since @code{vec_add} is a macro, the vector constant in the example
16172 is treated as four separate arguments. Wrap the entire argument in
16173 parentheses for this to work.
16176 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
16177 Internally, GCC uses built-in functions to achieve the functionality in
16178 the aforementioned header file, but they are not supported and are
16179 subject to change without notice.
16181 GCC complies with the OpenPOWER 64-Bit ELF V2 ABI Specification,
16182 which may be found at
16183 @uref{http://openpowerfoundation.org/wp-content/uploads/resources/leabi-prd/content/index.html}.
16184 Appendix A of this document lists the vector API interfaces that must be
16185 provided by compliant compilers. Programmers should preferentially use
16186 the interfaces described therein. However, historically GCC has provided
16187 additional interfaces for access to vector instructions. These are
16188 briefly described below.
16190 The following interfaces are supported for the generic and specific
16191 AltiVec operations and the AltiVec predicates. In cases where there
16192 is a direct mapping between generic and specific operations, only the
16193 generic names are shown here, although the specific operations can also
16196 Arguments that are documented as @code{const int} require literal
16197 integral values within the range required for that operation.
16200 vector signed char vec_abs (vector signed char);
16201 vector signed short vec_abs (vector signed short);
16202 vector signed int vec_abs (vector signed int);
16203 vector float vec_abs (vector float);
16205 vector signed char vec_abss (vector signed char);
16206 vector signed short vec_abss (vector signed short);
16207 vector signed int vec_abss (vector signed int);
16209 vector signed char vec_add (vector bool char, vector signed char);
16210 vector signed char vec_add (vector signed char, vector bool char);
16211 vector signed char vec_add (vector signed char, vector signed char);
16212 vector unsigned char vec_add (vector bool char, vector unsigned char);
16213 vector unsigned char vec_add (vector unsigned char, vector bool char);
16214 vector unsigned char vec_add (vector unsigned char,
16215 vector unsigned char);
16216 vector signed short vec_add (vector bool short, vector signed short);
16217 vector signed short vec_add (vector signed short, vector bool short);
16218 vector signed short vec_add (vector signed short, vector signed short);
16219 vector unsigned short vec_add (vector bool short,
16220 vector unsigned short);
16221 vector unsigned short vec_add (vector unsigned short,
16222 vector bool short);
16223 vector unsigned short vec_add (vector unsigned short,
16224 vector unsigned short);
16225 vector signed int vec_add (vector bool int, vector signed int);
16226 vector signed int vec_add (vector signed int, vector bool int);
16227 vector signed int vec_add (vector signed int, vector signed int);
16228 vector unsigned int vec_add (vector bool int, vector unsigned int);
16229 vector unsigned int vec_add (vector unsigned int, vector bool int);
16230 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
16231 vector float vec_add (vector float, vector float);
16233 vector float vec_vaddfp (vector float, vector float);
16235 vector signed int vec_vadduwm (vector bool int, vector signed int);
16236 vector signed int vec_vadduwm (vector signed int, vector bool int);
16237 vector signed int vec_vadduwm (vector signed int, vector signed int);
16238 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
16239 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
16240 vector unsigned int vec_vadduwm (vector unsigned int,
16241 vector unsigned int);
16243 vector signed short vec_vadduhm (vector bool short,
16244 vector signed short);
16245 vector signed short vec_vadduhm (vector signed short,
16246 vector bool short);
16247 vector signed short vec_vadduhm (vector signed short,
16248 vector signed short);
16249 vector unsigned short vec_vadduhm (vector bool short,
16250 vector unsigned short);
16251 vector unsigned short vec_vadduhm (vector unsigned short,
16252 vector bool short);
16253 vector unsigned short vec_vadduhm (vector unsigned short,
16254 vector unsigned short);
16256 vector signed char vec_vaddubm (vector bool char, vector signed char);
16257 vector signed char vec_vaddubm (vector signed char, vector bool char);
16258 vector signed char vec_vaddubm (vector signed char, vector signed char);
16259 vector unsigned char vec_vaddubm (vector bool char,
16260 vector unsigned char);
16261 vector unsigned char vec_vaddubm (vector unsigned char,
16263 vector unsigned char vec_vaddubm (vector unsigned char,
16264 vector unsigned char);
16266 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
16268 vector unsigned char vec_adds (vector bool char, vector unsigned char);
16269 vector unsigned char vec_adds (vector unsigned char, vector bool char);
16270 vector unsigned char vec_adds (vector unsigned char,
16271 vector unsigned char);
16272 vector signed char vec_adds (vector bool char, vector signed char);
16273 vector signed char vec_adds (vector signed char, vector bool char);
16274 vector signed char vec_adds (vector signed char, vector signed char);
16275 vector unsigned short vec_adds (vector bool short,
16276 vector unsigned short);
16277 vector unsigned short vec_adds (vector unsigned short,
16278 vector bool short);
16279 vector unsigned short vec_adds (vector unsigned short,
16280 vector unsigned short);
16281 vector signed short vec_adds (vector bool short, vector signed short);
16282 vector signed short vec_adds (vector signed short, vector bool short);
16283 vector signed short vec_adds (vector signed short, vector signed short);
16284 vector unsigned int vec_adds (vector bool int, vector unsigned int);
16285 vector unsigned int vec_adds (vector unsigned int, vector bool int);
16286 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
16287 vector signed int vec_adds (vector bool int, vector signed int);
16288 vector signed int vec_adds (vector signed int, vector bool int);
16289 vector signed int vec_adds (vector signed int, vector signed int);
16291 vector signed int vec_vaddsws (vector bool int, vector signed int);
16292 vector signed int vec_vaddsws (vector signed int, vector bool int);
16293 vector signed int vec_vaddsws (vector signed int, vector signed int);
16295 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
16296 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
16297 vector unsigned int vec_vadduws (vector unsigned int,
16298 vector unsigned int);
16300 vector signed short vec_vaddshs (vector bool short,
16301 vector signed short);
16302 vector signed short vec_vaddshs (vector signed short,
16303 vector bool short);
16304 vector signed short vec_vaddshs (vector signed short,
16305 vector signed short);
16307 vector unsigned short vec_vadduhs (vector bool short,
16308 vector unsigned short);
16309 vector unsigned short vec_vadduhs (vector unsigned short,
16310 vector bool short);
16311 vector unsigned short vec_vadduhs (vector unsigned short,
16312 vector unsigned short);
16314 vector signed char vec_vaddsbs (vector bool char, vector signed char);
16315 vector signed char vec_vaddsbs (vector signed char, vector bool char);
16316 vector signed char vec_vaddsbs (vector signed char, vector signed char);
16318 vector unsigned char vec_vaddubs (vector bool char,
16319 vector unsigned char);
16320 vector unsigned char vec_vaddubs (vector unsigned char,
16322 vector unsigned char vec_vaddubs (vector unsigned char,
16323 vector unsigned char);
16325 vector float vec_and (vector float, vector float);
16326 vector float vec_and (vector float, vector bool int);
16327 vector float vec_and (vector bool int, vector float);
16328 vector bool long long vec_and (vector bool long long int,
16329 vector bool long long);
16330 vector bool int vec_and (vector bool int, vector bool int);
16331 vector signed int vec_and (vector bool int, vector signed int);
16332 vector signed int vec_and (vector signed int, vector bool int);
16333 vector signed int vec_and (vector signed int, vector signed int);
16334 vector unsigned int vec_and (vector bool int, vector unsigned int);
16335 vector unsigned int vec_and (vector unsigned int, vector bool int);
16336 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
16337 vector bool short vec_and (vector bool short, vector bool short);
16338 vector signed short vec_and (vector bool short, vector signed short);
16339 vector signed short vec_and (vector signed short, vector bool short);
16340 vector signed short vec_and (vector signed short, vector signed short);
16341 vector unsigned short vec_and (vector bool short,
16342 vector unsigned short);
16343 vector unsigned short vec_and (vector unsigned short,
16344 vector bool short);
16345 vector unsigned short vec_and (vector unsigned short,
16346 vector unsigned short);
16347 vector signed char vec_and (vector bool char, vector signed char);
16348 vector bool char vec_and (vector bool char, vector bool char);
16349 vector signed char vec_and (vector signed char, vector bool char);
16350 vector signed char vec_and (vector signed char, vector signed char);
16351 vector unsigned char vec_and (vector bool char, vector unsigned char);
16352 vector unsigned char vec_and (vector unsigned char, vector bool char);
16353 vector unsigned char vec_and (vector unsigned char,
16354 vector unsigned char);
16356 vector float vec_andc (vector float, vector float);
16357 vector float vec_andc (vector float, vector bool int);
16358 vector float vec_andc (vector bool int, vector float);
16359 vector bool int vec_andc (vector bool int, vector bool int);
16360 vector signed int vec_andc (vector bool int, vector signed int);
16361 vector signed int vec_andc (vector signed int, vector bool int);
16362 vector signed int vec_andc (vector signed int, vector signed int);
16363 vector unsigned int vec_andc (vector bool int, vector unsigned int);
16364 vector unsigned int vec_andc (vector unsigned int, vector bool int);
16365 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
16366 vector bool short vec_andc (vector bool short, vector bool short);
16367 vector signed short vec_andc (vector bool short, vector signed short);
16368 vector signed short vec_andc (vector signed short, vector bool short);
16369 vector signed short vec_andc (vector signed short, vector signed short);
16370 vector unsigned short vec_andc (vector bool short,
16371 vector unsigned short);
16372 vector unsigned short vec_andc (vector unsigned short,
16373 vector bool short);
16374 vector unsigned short vec_andc (vector unsigned short,
16375 vector unsigned short);
16376 vector signed char vec_andc (vector bool char, vector signed char);
16377 vector bool char vec_andc (vector bool char, vector bool char);
16378 vector signed char vec_andc (vector signed char, vector bool char);
16379 vector signed char vec_andc (vector signed char, vector signed char);
16380 vector unsigned char vec_andc (vector bool char, vector unsigned char);
16381 vector unsigned char vec_andc (vector unsigned char, vector bool char);
16382 vector unsigned char vec_andc (vector unsigned char,
16383 vector unsigned char);
16385 vector unsigned char vec_avg (vector unsigned char,
16386 vector unsigned char);
16387 vector signed char vec_avg (vector signed char, vector signed char);
16388 vector unsigned short vec_avg (vector unsigned short,
16389 vector unsigned short);
16390 vector signed short vec_avg (vector signed short, vector signed short);
16391 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
16392 vector signed int vec_avg (vector signed int, vector signed int);
16394 vector signed int vec_vavgsw (vector signed int, vector signed int);
16396 vector unsigned int vec_vavguw (vector unsigned int,
16397 vector unsigned int);
16399 vector signed short vec_vavgsh (vector signed short,
16400 vector signed short);
16402 vector unsigned short vec_vavguh (vector unsigned short,
16403 vector unsigned short);
16405 vector signed char vec_vavgsb (vector signed char, vector signed char);
16407 vector unsigned char vec_vavgub (vector unsigned char,
16408 vector unsigned char);
16410 vector float vec_ceil (vector float);
16412 vector signed int vec_cmpb (vector float, vector float);
16414 vector bool char vec_cmpeq (vector bool char, vector bool char);
16415 vector bool short vec_cmpeq (vector bool short, vector bool short);
16416 vector bool int vec_cmpeq (vector bool int, vector bool int);
16417 vector bool char vec_cmpeq (vector signed char, vector signed char);
16418 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
16419 vector bool short vec_cmpeq (vector signed short, vector signed short);
16420 vector bool short vec_cmpeq (vector unsigned short,
16421 vector unsigned short);
16422 vector bool int vec_cmpeq (vector signed int, vector signed int);
16423 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
16424 vector bool int vec_cmpeq (vector float, vector float);
16426 vector bool int vec_vcmpeqfp (vector float, vector float);
16428 vector bool int vec_vcmpequw (vector signed int, vector signed int);
16429 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
16431 vector bool short vec_vcmpequh (vector signed short,
16432 vector signed short);
16433 vector bool short vec_vcmpequh (vector unsigned short,
16434 vector unsigned short);
16436 vector bool char vec_vcmpequb (vector signed char, vector signed char);
16437 vector bool char vec_vcmpequb (vector unsigned char,
16438 vector unsigned char);
16440 vector bool int vec_cmpge (vector float, vector float);
16442 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
16443 vector bool char vec_cmpgt (vector signed char, vector signed char);
16444 vector bool short vec_cmpgt (vector unsigned short,
16445 vector unsigned short);
16446 vector bool short vec_cmpgt (vector signed short, vector signed short);
16447 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
16448 vector bool int vec_cmpgt (vector signed int, vector signed int);
16449 vector bool int vec_cmpgt (vector float, vector float);
16451 vector bool int vec_vcmpgtfp (vector float, vector float);
16453 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
16455 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
16457 vector bool short vec_vcmpgtsh (vector signed short,
16458 vector signed short);
16460 vector bool short vec_vcmpgtuh (vector unsigned short,
16461 vector unsigned short);
16463 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
16465 vector bool char vec_vcmpgtub (vector unsigned char,
16466 vector unsigned char);
16468 vector bool int vec_cmple (vector float, vector float);
16470 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
16471 vector bool char vec_cmplt (vector signed char, vector signed char);
16472 vector bool short vec_cmplt (vector unsigned short,
16473 vector unsigned short);
16474 vector bool short vec_cmplt (vector signed short, vector signed short);
16475 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
16476 vector bool int vec_cmplt (vector signed int, vector signed int);
16477 vector bool int vec_cmplt (vector float, vector float);
16479 vector float vec_cpsgn (vector float, vector float);
16481 vector float vec_ctf (vector unsigned int, const int);
16482 vector float vec_ctf (vector signed int, const int);
16483 vector double vec_ctf (vector unsigned long, const int);
16484 vector double vec_ctf (vector signed long, const int);
16486 vector float vec_vcfsx (vector signed int, const int);
16488 vector float vec_vcfux (vector unsigned int, const int);
16490 vector signed int vec_cts (vector float, const int);
16491 vector signed long vec_cts (vector double, const int);
16493 vector unsigned int vec_ctu (vector float, const int);
16494 vector unsigned long vec_ctu (vector double, const int);
16496 vector double vec_doublee (vector float);
16497 vector double vec_doublee (vector signed int);
16498 vector double vec_doublee (vector unsigned int);
16500 vector double vec_doubleo (vector float);
16501 vector double vec_doubleo (vector signed int);
16502 vector double vec_doubleo (vector unsigned int);
16504 vector double vec_doubleh (vector float);
16505 vector double vec_doubleh (vector signed int);
16506 vector double vec_doubleh (vector unsigned int);
16508 vector double vec_doublel (vector float);
16509 vector double vec_doublel (vector signed int);
16510 vector double vec_doublel (vector unsigned int);
16512 void vec_dss (const int);
16514 void vec_dssall (void);
16516 void vec_dst (const vector unsigned char *, int, const int);
16517 void vec_dst (const vector signed char *, int, const int);
16518 void vec_dst (const vector bool char *, int, const int);
16519 void vec_dst (const vector unsigned short *, int, const int);
16520 void vec_dst (const vector signed short *, int, const int);
16521 void vec_dst (const vector bool short *, int, const int);
16522 void vec_dst (const vector pixel *, int, const int);
16523 void vec_dst (const vector unsigned int *, int, const int);
16524 void vec_dst (const vector signed int *, int, const int);
16525 void vec_dst (const vector bool int *, int, const int);
16526 void vec_dst (const vector float *, int, const int);
16527 void vec_dst (const unsigned char *, int, const int);
16528 void vec_dst (const signed char *, int, const int);
16529 void vec_dst (const unsigned short *, int, const int);
16530 void vec_dst (const short *, int, const int);
16531 void vec_dst (const unsigned int *, int, const int);
16532 void vec_dst (const int *, int, const int);
16533 void vec_dst (const unsigned long *, int, const int);
16534 void vec_dst (const long *, int, const int);
16535 void vec_dst (const float *, int, const int);
16537 void vec_dstst (const vector unsigned char *, int, const int);
16538 void vec_dstst (const vector signed char *, int, const int);
16539 void vec_dstst (const vector bool char *, int, const int);
16540 void vec_dstst (const vector unsigned short *, int, const int);
16541 void vec_dstst (const vector signed short *, int, const int);
16542 void vec_dstst (const vector bool short *, int, const int);
16543 void vec_dstst (const vector pixel *, int, const int);
16544 void vec_dstst (const vector unsigned int *, int, const int);
16545 void vec_dstst (const vector signed int *, int, const int);
16546 void vec_dstst (const vector bool int *, int, const int);
16547 void vec_dstst (const vector float *, int, const int);
16548 void vec_dstst (const unsigned char *, int, const int);
16549 void vec_dstst (const signed char *, int, const int);
16550 void vec_dstst (const unsigned short *, int, const int);
16551 void vec_dstst (const short *, int, const int);
16552 void vec_dstst (const unsigned int *, int, const int);
16553 void vec_dstst (const int *, int, const int);
16554 void vec_dstst (const unsigned long *, int, const int);
16555 void vec_dstst (const long *, int, const int);
16556 void vec_dstst (const float *, int, const int);
16558 void vec_dststt (const vector unsigned char *, int, const int);
16559 void vec_dststt (const vector signed char *, int, const int);
16560 void vec_dststt (const vector bool char *, int, const int);
16561 void vec_dststt (const vector unsigned short *, int, const int);
16562 void vec_dststt (const vector signed short *, int, const int);
16563 void vec_dststt (const vector bool short *, int, const int);
16564 void vec_dststt (const vector pixel *, int, const int);
16565 void vec_dststt (const vector unsigned int *, int, const int);
16566 void vec_dststt (const vector signed int *, int, const int);
16567 void vec_dststt (const vector bool int *, int, const int);
16568 void vec_dststt (const vector float *, int, const int);
16569 void vec_dststt (const unsigned char *, int, const int);
16570 void vec_dststt (const signed char *, int, const int);
16571 void vec_dststt (const unsigned short *, int, const int);
16572 void vec_dststt (const short *, int, const int);
16573 void vec_dststt (const unsigned int *, int, const int);
16574 void vec_dststt (const int *, int, const int);
16575 void vec_dststt (const unsigned long *, int, const int);
16576 void vec_dststt (const long *, int, const int);
16577 void vec_dststt (const float *, int, const int);
16579 void vec_dstt (const vector unsigned char *, int, const int);
16580 void vec_dstt (const vector signed char *, int, const int);
16581 void vec_dstt (const vector bool char *, int, const int);
16582 void vec_dstt (const vector unsigned short *, int, const int);
16583 void vec_dstt (const vector signed short *, int, const int);
16584 void vec_dstt (const vector bool short *, int, const int);
16585 void vec_dstt (const vector pixel *, int, const int);
16586 void vec_dstt (const vector unsigned int *, int, const int);
16587 void vec_dstt (const vector signed int *, int, const int);
16588 void vec_dstt (const vector bool int *, int, const int);
16589 void vec_dstt (const vector float *, int, const int);
16590 void vec_dstt (const unsigned char *, int, const int);
16591 void vec_dstt (const signed char *, int, const int);
16592 void vec_dstt (const unsigned short *, int, const int);
16593 void vec_dstt (const short *, int, const int);
16594 void vec_dstt (const unsigned int *, int, const int);
16595 void vec_dstt (const int *, int, const int);
16596 void vec_dstt (const unsigned long *, int, const int);
16597 void vec_dstt (const long *, int, const int);
16598 void vec_dstt (const float *, int, const int);
16600 vector float vec_expte (vector float);
16602 vector float vec_floor (vector float);
16604 vector float vec_float (vector signed int);
16605 vector float vec_float (vector unsigned int);
16607 vector float vec_float2 (vector signed long long, vector signed long long);
16608 vector float vec_float2 (vector unsigned long long, vector signed long long);
16610 vector float vec_floate (vector double);
16611 vector float vec_floate (vector signed long long);
16612 vector float vec_floate (vector unsigned long long);
16614 vector float vec_floato (vector double);
16615 vector float vec_floato (vector signed long long);
16616 vector float vec_floato (vector unsigned long long);
16618 vector float vec_ld (int, const vector float *);
16619 vector float vec_ld (int, const float *);
16620 vector bool int vec_ld (int, const vector bool int *);
16621 vector signed int vec_ld (int, const vector signed int *);
16622 vector signed int vec_ld (int, const int *);
16623 vector unsigned int vec_ld (int, const vector unsigned int *);
16624 vector unsigned int vec_ld (int, const unsigned int *);
16625 vector bool short vec_ld (int, const vector bool short *);
16626 vector pixel vec_ld (int, const vector pixel *);
16627 vector signed short vec_ld (int, const vector signed short *);
16628 vector signed short vec_ld (int, const short *);
16629 vector unsigned short vec_ld (int, const vector unsigned short *);
16630 vector unsigned short vec_ld (int, const unsigned short *);
16631 vector bool char vec_ld (int, const vector bool char *);
16632 vector signed char vec_ld (int, const vector signed char *);
16633 vector signed char vec_ld (int, const signed char *);
16634 vector unsigned char vec_ld (int, const vector unsigned char *);
16635 vector unsigned char vec_ld (int, const unsigned char *);
16637 vector signed char vec_lde (int, const signed char *);
16638 vector unsigned char vec_lde (int, const unsigned char *);
16639 vector signed short vec_lde (int, const short *);
16640 vector unsigned short vec_lde (int, const unsigned short *);
16641 vector float vec_lde (int, const float *);
16642 vector signed int vec_lde (int, const int *);
16643 vector unsigned int vec_lde (int, const unsigned int *);
16645 vector float vec_lvewx (int, float *);
16646 vector signed int vec_lvewx (int, int *);
16647 vector unsigned int vec_lvewx (int, unsigned int *);
16649 vector signed short vec_lvehx (int, short *);
16650 vector unsigned short vec_lvehx (int, unsigned short *);
16652 vector signed char vec_lvebx (int, char *);
16653 vector unsigned char vec_lvebx (int, unsigned char *);
16655 vector float vec_ldl (int, const vector float *);
16656 vector float vec_ldl (int, const float *);
16657 vector bool int vec_ldl (int, const vector bool int *);
16658 vector signed int vec_ldl (int, const vector signed int *);
16659 vector signed int vec_ldl (int, const int *);
16660 vector unsigned int vec_ldl (int, const vector unsigned int *);
16661 vector unsigned int vec_ldl (int, const unsigned int *);
16662 vector bool short vec_ldl (int, const vector bool short *);
16663 vector pixel vec_ldl (int, const vector pixel *);
16664 vector signed short vec_ldl (int, const vector signed short *);
16665 vector signed short vec_ldl (int, const short *);
16666 vector unsigned short vec_ldl (int, const vector unsigned short *);
16667 vector unsigned short vec_ldl (int, const unsigned short *);
16668 vector bool char vec_ldl (int, const vector bool char *);
16669 vector signed char vec_ldl (int, const vector signed char *);
16670 vector signed char vec_ldl (int, const signed char *);
16671 vector unsigned char vec_ldl (int, const vector unsigned char *);
16672 vector unsigned char vec_ldl (int, const unsigned char *);
16674 vector float vec_loge (vector float);
16676 vector unsigned char vec_lvsl (int, const unsigned char *);
16677 vector unsigned char vec_lvsl (int, const signed char *);
16678 vector unsigned char vec_lvsl (int, const unsigned short *);
16679 vector unsigned char vec_lvsl (int, const short *);
16680 vector unsigned char vec_lvsl (int, const unsigned int *);
16681 vector unsigned char vec_lvsl (int, const int *);
16682 vector unsigned char vec_lvsl (int, const unsigned long *);
16683 vector unsigned char vec_lvsl (int, const long *);
16684 vector unsigned char vec_lvsl (int, const float *);
16686 vector unsigned char vec_lvsr (int, const unsigned char *);
16687 vector unsigned char vec_lvsr (int, const signed char *);
16688 vector unsigned char vec_lvsr (int, const unsigned short *);
16689 vector unsigned char vec_lvsr (int, const short *);
16690 vector unsigned char vec_lvsr (int, const unsigned int *);
16691 vector unsigned char vec_lvsr (int, const int *);
16692 vector unsigned char vec_lvsr (int, const unsigned long *);
16693 vector unsigned char vec_lvsr (int, const long *);
16694 vector unsigned char vec_lvsr (int, const float *);
16696 vector float vec_madd (vector float, vector float, vector float);
16698 vector signed short vec_madds (vector signed short,
16699 vector signed short,
16700 vector signed short);
16702 vector unsigned char vec_max (vector bool char, vector unsigned char);
16703 vector unsigned char vec_max (vector unsigned char, vector bool char);
16704 vector unsigned char vec_max (vector unsigned char,
16705 vector unsigned char);
16706 vector signed char vec_max (vector bool char, vector signed char);
16707 vector signed char vec_max (vector signed char, vector bool char);
16708 vector signed char vec_max (vector signed char, vector signed char);
16709 vector unsigned short vec_max (vector bool short,
16710 vector unsigned short);
16711 vector unsigned short vec_max (vector unsigned short,
16712 vector bool short);
16713 vector unsigned short vec_max (vector unsigned short,
16714 vector unsigned short);
16715 vector signed short vec_max (vector bool short, vector signed short);
16716 vector signed short vec_max (vector signed short, vector bool short);
16717 vector signed short vec_max (vector signed short, vector signed short);
16718 vector unsigned int vec_max (vector bool int, vector unsigned int);
16719 vector unsigned int vec_max (vector unsigned int, vector bool int);
16720 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
16721 vector signed int vec_max (vector bool int, vector signed int);
16722 vector signed int vec_max (vector signed int, vector bool int);
16723 vector signed int vec_max (vector signed int, vector signed int);
16724 vector float vec_max (vector float, vector float);
16726 vector float vec_vmaxfp (vector float, vector float);
16728 vector signed int vec_vmaxsw (vector bool int, vector signed int);
16729 vector signed int vec_vmaxsw (vector signed int, vector bool int);
16730 vector signed int vec_vmaxsw (vector signed int, vector signed int);
16732 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
16733 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
16734 vector unsigned int vec_vmaxuw (vector unsigned int,
16735 vector unsigned int);
16737 vector signed short vec_vmaxsh (vector bool short, vector signed short);
16738 vector signed short vec_vmaxsh (vector signed short, vector bool short);
16739 vector signed short vec_vmaxsh (vector signed short,
16740 vector signed short);
16742 vector unsigned short vec_vmaxuh (vector bool short,
16743 vector unsigned short);
16744 vector unsigned short vec_vmaxuh (vector unsigned short,
16745 vector bool short);
16746 vector unsigned short vec_vmaxuh (vector unsigned short,
16747 vector unsigned short);
16749 vector signed char vec_vmaxsb (vector bool char, vector signed char);
16750 vector signed char vec_vmaxsb (vector signed char, vector bool char);
16751 vector signed char vec_vmaxsb (vector signed char, vector signed char);
16753 vector unsigned char vec_vmaxub (vector bool char,
16754 vector unsigned char);
16755 vector unsigned char vec_vmaxub (vector unsigned char,
16757 vector unsigned char vec_vmaxub (vector unsigned char,
16758 vector unsigned char);
16760 vector bool char vec_mergeh (vector bool char, vector bool char);
16761 vector signed char vec_mergeh (vector signed char, vector signed char);
16762 vector unsigned char vec_mergeh (vector unsigned char,
16763 vector unsigned char);
16764 vector bool short vec_mergeh (vector bool short, vector bool short);
16765 vector pixel vec_mergeh (vector pixel, vector pixel);
16766 vector signed short vec_mergeh (vector signed short,
16767 vector signed short);
16768 vector unsigned short vec_mergeh (vector unsigned short,
16769 vector unsigned short);
16770 vector float vec_mergeh (vector float, vector float);
16771 vector bool int vec_mergeh (vector bool int, vector bool int);
16772 vector signed int vec_mergeh (vector signed int, vector signed int);
16773 vector unsigned int vec_mergeh (vector unsigned int,
16774 vector unsigned int);
16776 vector float vec_vmrghw (vector float, vector float);
16777 vector bool int vec_vmrghw (vector bool int, vector bool int);
16778 vector signed int vec_vmrghw (vector signed int, vector signed int);
16779 vector unsigned int vec_vmrghw (vector unsigned int,
16780 vector unsigned int);
16782 vector bool short vec_vmrghh (vector bool short, vector bool short);
16783 vector signed short vec_vmrghh (vector signed short,
16784 vector signed short);
16785 vector unsigned short vec_vmrghh (vector unsigned short,
16786 vector unsigned short);
16787 vector pixel vec_vmrghh (vector pixel, vector pixel);
16789 vector bool char vec_vmrghb (vector bool char, vector bool char);
16790 vector signed char vec_vmrghb (vector signed char, vector signed char);
16791 vector unsigned char vec_vmrghb (vector unsigned char,
16792 vector unsigned char);
16794 vector bool char vec_mergel (vector bool char, vector bool char);
16795 vector signed char vec_mergel (vector signed char, vector signed char);
16796 vector unsigned char vec_mergel (vector unsigned char,
16797 vector unsigned char);
16798 vector bool short vec_mergel (vector bool short, vector bool short);
16799 vector pixel vec_mergel (vector pixel, vector pixel);
16800 vector signed short vec_mergel (vector signed short,
16801 vector signed short);
16802 vector unsigned short vec_mergel (vector unsigned short,
16803 vector unsigned short);
16804 vector float vec_mergel (vector float, vector float);
16805 vector bool int vec_mergel (vector bool int, vector bool int);
16806 vector signed int vec_mergel (vector signed int, vector signed int);
16807 vector unsigned int vec_mergel (vector unsigned int,
16808 vector unsigned int);
16810 vector float vec_vmrglw (vector float, vector float);
16811 vector signed int vec_vmrglw (vector signed int, vector signed int);
16812 vector unsigned int vec_vmrglw (vector unsigned int,
16813 vector unsigned int);
16814 vector bool int vec_vmrglw (vector bool int, vector bool int);
16816 vector bool short vec_vmrglh (vector bool short, vector bool short);
16817 vector signed short vec_vmrglh (vector signed short,
16818 vector signed short);
16819 vector unsigned short vec_vmrglh (vector unsigned short,
16820 vector unsigned short);
16821 vector pixel vec_vmrglh (vector pixel, vector pixel);
16823 vector bool char vec_vmrglb (vector bool char, vector bool char);
16824 vector signed char vec_vmrglb (vector signed char, vector signed char);
16825 vector unsigned char vec_vmrglb (vector unsigned char,
16826 vector unsigned char);
16828 vector unsigned short vec_mfvscr (void);
16830 vector unsigned char vec_min (vector bool char, vector unsigned char);
16831 vector unsigned char vec_min (vector unsigned char, vector bool char);
16832 vector unsigned char vec_min (vector unsigned char,
16833 vector unsigned char);
16834 vector signed char vec_min (vector bool char, vector signed char);
16835 vector signed char vec_min (vector signed char, vector bool char);
16836 vector signed char vec_min (vector signed char, vector signed char);
16837 vector unsigned short vec_min (vector bool short,
16838 vector unsigned short);
16839 vector unsigned short vec_min (vector unsigned short,
16840 vector bool short);
16841 vector unsigned short vec_min (vector unsigned short,
16842 vector unsigned short);
16843 vector signed short vec_min (vector bool short, vector signed short);
16844 vector signed short vec_min (vector signed short, vector bool short);
16845 vector signed short vec_min (vector signed short, vector signed short);
16846 vector unsigned int vec_min (vector bool int, vector unsigned int);
16847 vector unsigned int vec_min (vector unsigned int, vector bool int);
16848 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
16849 vector signed int vec_min (vector bool int, vector signed int);
16850 vector signed int vec_min (vector signed int, vector bool int);
16851 vector signed int vec_min (vector signed int, vector signed int);
16852 vector float vec_min (vector float, vector float);
16854 vector float vec_vminfp (vector float, vector float);
16856 vector signed int vec_vminsw (vector bool int, vector signed int);
16857 vector signed int vec_vminsw (vector signed int, vector bool int);
16858 vector signed int vec_vminsw (vector signed int, vector signed int);
16860 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
16861 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
16862 vector unsigned int vec_vminuw (vector unsigned int,
16863 vector unsigned int);
16865 vector signed short vec_vminsh (vector bool short, vector signed short);
16866 vector signed short vec_vminsh (vector signed short, vector bool short);
16867 vector signed short vec_vminsh (vector signed short,
16868 vector signed short);
16870 vector unsigned short vec_vminuh (vector bool short,
16871 vector unsigned short);
16872 vector unsigned short vec_vminuh (vector unsigned short,
16873 vector bool short);
16874 vector unsigned short vec_vminuh (vector unsigned short,
16875 vector unsigned short);
16877 vector signed char vec_vminsb (vector bool char, vector signed char);
16878 vector signed char vec_vminsb (vector signed char, vector bool char);
16879 vector signed char vec_vminsb (vector signed char, vector signed char);
16881 vector unsigned char vec_vminub (vector bool char,
16882 vector unsigned char);
16883 vector unsigned char vec_vminub (vector unsigned char,
16885 vector unsigned char vec_vminub (vector unsigned char,
16886 vector unsigned char);
16888 vector signed short vec_mladd (vector signed short,
16889 vector signed short,
16890 vector signed short);
16891 vector signed short vec_mladd (vector signed short,
16892 vector unsigned short,
16893 vector unsigned short);
16894 vector signed short vec_mladd (vector unsigned short,
16895 vector signed short,
16896 vector signed short);
16897 vector unsigned short vec_mladd (vector unsigned short,
16898 vector unsigned short,
16899 vector unsigned short);
16901 vector signed short vec_mradds (vector signed short,
16902 vector signed short,
16903 vector signed short);
16905 vector unsigned int vec_msum (vector unsigned char,
16906 vector unsigned char,
16907 vector unsigned int);
16908 vector signed int vec_msum (vector signed char,
16909 vector unsigned char,
16910 vector signed int);
16911 vector unsigned int vec_msum (vector unsigned short,
16912 vector unsigned short,
16913 vector unsigned int);
16914 vector signed int vec_msum (vector signed short,
16915 vector signed short,
16916 vector signed int);
16918 vector signed int vec_vmsumshm (vector signed short,
16919 vector signed short,
16920 vector signed int);
16922 vector unsigned int vec_vmsumuhm (vector unsigned short,
16923 vector unsigned short,
16924 vector unsigned int);
16926 vector signed int vec_vmsummbm (vector signed char,
16927 vector unsigned char,
16928 vector signed int);
16930 vector unsigned int vec_vmsumubm (vector unsigned char,
16931 vector unsigned char,
16932 vector unsigned int);
16934 vector unsigned int vec_msums (vector unsigned short,
16935 vector unsigned short,
16936 vector unsigned int);
16937 vector signed int vec_msums (vector signed short,
16938 vector signed short,
16939 vector signed int);
16941 vector signed int vec_vmsumshs (vector signed short,
16942 vector signed short,
16943 vector signed int);
16945 vector unsigned int vec_vmsumuhs (vector unsigned short,
16946 vector unsigned short,
16947 vector unsigned int);
16949 void vec_mtvscr (vector signed int);
16950 void vec_mtvscr (vector unsigned int);
16951 void vec_mtvscr (vector bool int);
16952 void vec_mtvscr (vector signed short);
16953 void vec_mtvscr (vector unsigned short);
16954 void vec_mtvscr (vector bool short);
16955 void vec_mtvscr (vector pixel);
16956 void vec_mtvscr (vector signed char);
16957 void vec_mtvscr (vector unsigned char);
16958 void vec_mtvscr (vector bool char);
16960 vector unsigned short vec_mule (vector unsigned char,
16961 vector unsigned char);
16962 vector signed short vec_mule (vector signed char,
16963 vector signed char);
16964 vector unsigned int vec_mule (vector unsigned short,
16965 vector unsigned short);
16966 vector signed int vec_mule (vector signed short, vector signed short);
16967 vector unsigned long long vec_mule (vector unsigned int,
16968 vector unsigned int);
16969 vector signed long long vec_mule (vector signed int,
16970 vector signed int);
16972 vector signed int vec_vmulesh (vector signed short,
16973 vector signed short);
16975 vector unsigned int vec_vmuleuh (vector unsigned short,
16976 vector unsigned short);
16978 vector signed short vec_vmulesb (vector signed char,
16979 vector signed char);
16981 vector unsigned short vec_vmuleub (vector unsigned char,
16982 vector unsigned char);
16984 vector unsigned short vec_mulo (vector unsigned char,
16985 vector unsigned char);
16986 vector signed short vec_mulo (vector signed char, vector signed char);
16987 vector unsigned int vec_mulo (vector unsigned short,
16988 vector unsigned short);
16989 vector signed int vec_mulo (vector signed short, vector signed short);
16990 vector unsigned long long vec_mulo (vector unsigned int,
16991 vector unsigned int);
16992 vector signed long long vec_mulo (vector signed int,
16993 vector signed int);
16995 vector signed int vec_vmulosh (vector signed short,
16996 vector signed short);
16998 vector unsigned int vec_vmulouh (vector unsigned short,
16999 vector unsigned short);
17001 vector signed short vec_vmulosb (vector signed char,
17002 vector signed char);
17004 vector unsigned short vec_vmuloub (vector unsigned char,
17005 vector unsigned char);
17007 vector float vec_nmsub (vector float, vector float, vector float);
17009 vector signed char vec_nabs (vector signed char);
17010 vector signed short vec_nabs (vector signed short);
17011 vector signed int vec_nabs (vector signed int);
17012 vector float vec_nabs (vector float);
17013 vector double vec_nabs (vector double);
17015 vector signed char vec_neg (vector signed char);
17016 vector signed short vec_neg (vector signed short);
17017 vector signed int vec_neg (vector signed int);
17018 vector signed long long vec_neg (vector signed long long);
17019 vector float char vec_neg (vector float);
17020 vector double vec_neg (vector double);
17022 vector float vec_nor (vector float, vector float);
17023 vector signed int vec_nor (vector signed int, vector signed int);
17024 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
17025 vector bool int vec_nor (vector bool int, vector bool int);
17026 vector signed short vec_nor (vector signed short, vector signed short);
17027 vector unsigned short vec_nor (vector unsigned short,
17028 vector unsigned short);
17029 vector bool short vec_nor (vector bool short, vector bool short);
17030 vector signed char vec_nor (vector signed char, vector signed char);
17031 vector unsigned char vec_nor (vector unsigned char,
17032 vector unsigned char);
17033 vector bool char vec_nor (vector bool char, vector bool char);
17035 vector float vec_or (vector float, vector float);
17036 vector float vec_or (vector float, vector bool int);
17037 vector float vec_or (vector bool int, vector float);
17038 vector bool int vec_or (vector bool int, vector bool int);
17039 vector signed int vec_or (vector bool int, vector signed int);
17040 vector signed int vec_or (vector signed int, vector bool int);
17041 vector signed int vec_or (vector signed int, vector signed int);
17042 vector unsigned int vec_or (vector bool int, vector unsigned int);
17043 vector unsigned int vec_or (vector unsigned int, vector bool int);
17044 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
17045 vector bool short vec_or (vector bool short, vector bool short);
17046 vector signed short vec_or (vector bool short, vector signed short);
17047 vector signed short vec_or (vector signed short, vector bool short);
17048 vector signed short vec_or (vector signed short, vector signed short);
17049 vector unsigned short vec_or (vector bool short, vector unsigned short);
17050 vector unsigned short vec_or (vector unsigned short, vector bool short);
17051 vector unsigned short vec_or (vector unsigned short,
17052 vector unsigned short);
17053 vector signed char vec_or (vector bool char, vector signed char);
17054 vector bool char vec_or (vector bool char, vector bool char);
17055 vector signed char vec_or (vector signed char, vector bool char);
17056 vector signed char vec_or (vector signed char, vector signed char);
17057 vector unsigned char vec_or (vector bool char, vector unsigned char);
17058 vector unsigned char vec_or (vector unsigned char, vector bool char);
17059 vector unsigned char vec_or (vector unsigned char,
17060 vector unsigned char);
17062 vector signed char vec_pack (vector signed short, vector signed short);
17063 vector unsigned char vec_pack (vector unsigned short,
17064 vector unsigned short);
17065 vector bool char vec_pack (vector bool short, vector bool short);
17066 vector signed short vec_pack (vector signed int, vector signed int);
17067 vector unsigned short vec_pack (vector unsigned int,
17068 vector unsigned int);
17069 vector bool short vec_pack (vector bool int, vector bool int);
17071 vector bool short vec_vpkuwum (vector bool int, vector bool int);
17072 vector signed short vec_vpkuwum (vector signed int, vector signed int);
17073 vector unsigned short vec_vpkuwum (vector unsigned int,
17074 vector unsigned int);
17076 vector bool char vec_vpkuhum (vector bool short, vector bool short);
17077 vector signed char vec_vpkuhum (vector signed short,
17078 vector signed short);
17079 vector unsigned char vec_vpkuhum (vector unsigned short,
17080 vector unsigned short);
17082 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
17084 vector unsigned char vec_packs (vector unsigned short,
17085 vector unsigned short);
17086 vector signed char vec_packs (vector signed short, vector signed short);
17087 vector unsigned short vec_packs (vector unsigned int,
17088 vector unsigned int);
17089 vector signed short vec_packs (vector signed int, vector signed int);
17091 vector signed short vec_vpkswss (vector signed int, vector signed int);
17093 vector unsigned short vec_vpkuwus (vector unsigned int,
17094 vector unsigned int);
17096 vector signed char vec_vpkshss (vector signed short,
17097 vector signed short);
17099 vector unsigned char vec_vpkuhus (vector unsigned short,
17100 vector unsigned short);
17102 vector unsigned char vec_packsu (vector unsigned short,
17103 vector unsigned short);
17104 vector unsigned char vec_packsu (vector signed short,
17105 vector signed short);
17106 vector unsigned short vec_packsu (vector unsigned int,
17107 vector unsigned int);
17108 vector unsigned short vec_packsu (vector signed int, vector signed int);
17110 vector unsigned short vec_vpkswus (vector signed int,
17111 vector signed int);
17113 vector unsigned char vec_vpkshus (vector signed short,
17114 vector signed short);
17116 vector float vec_perm (vector float,
17118 vector unsigned char);
17119 vector signed int vec_perm (vector signed int,
17121 vector unsigned char);
17122 vector unsigned int vec_perm (vector unsigned int,
17123 vector unsigned int,
17124 vector unsigned char);
17125 vector bool int vec_perm (vector bool int,
17127 vector unsigned char);
17128 vector signed short vec_perm (vector signed short,
17129 vector signed short,
17130 vector unsigned char);
17131 vector unsigned short vec_perm (vector unsigned short,
17132 vector unsigned short,
17133 vector unsigned char);
17134 vector bool short vec_perm (vector bool short,
17136 vector unsigned char);
17137 vector pixel vec_perm (vector pixel,
17139 vector unsigned char);
17140 vector signed char vec_perm (vector signed char,
17141 vector signed char,
17142 vector unsigned char);
17143 vector unsigned char vec_perm (vector unsigned char,
17144 vector unsigned char,
17145 vector unsigned char);
17146 vector bool char vec_perm (vector bool char,
17148 vector unsigned char);
17150 vector float vec_re (vector float);
17152 vector bool char vec_reve (vector bool char);
17153 vector signed char vec_reve (vector signed char);
17154 vector unsigned char vec_reve (vector unsigned char);
17155 vector bool int vec_reve (vector bool int);
17156 vector signed int vec_reve (vector signed int);
17157 vector unsigned int vec_reve (vector unsigned int);
17158 vector bool long long vec_reve (vector bool long long);
17159 vector signed long long vec_reve (vector signed long long);
17160 vector unsigned long long vec_reve (vector unsigned long long);
17161 vector bool short vec_reve (vector bool short);
17162 vector signed short vec_reve (vector signed short);
17163 vector unsigned short vec_reve (vector unsigned short);
17165 vector signed char vec_rl (vector signed char,
17166 vector unsigned char);
17167 vector unsigned char vec_rl (vector unsigned char,
17168 vector unsigned char);
17169 vector signed short vec_rl (vector signed short, vector unsigned short);
17170 vector unsigned short vec_rl (vector unsigned short,
17171 vector unsigned short);
17172 vector signed int vec_rl (vector signed int, vector unsigned int);
17173 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
17175 vector signed int vec_vrlw (vector signed int, vector unsigned int);
17176 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
17178 vector signed short vec_vrlh (vector signed short,
17179 vector unsigned short);
17180 vector unsigned short vec_vrlh (vector unsigned short,
17181 vector unsigned short);
17183 vector signed char vec_vrlb (vector signed char, vector unsigned char);
17184 vector unsigned char vec_vrlb (vector unsigned char,
17185 vector unsigned char);
17187 vector float vec_round (vector float);
17189 vector float vec_rsqrt (vector float);
17191 vector float vec_rsqrte (vector float);
17193 vector float vec_sel (vector float, vector float, vector bool int);
17194 vector float vec_sel (vector float, vector float, vector unsigned int);
17195 vector signed int vec_sel (vector signed int,
17198 vector signed int vec_sel (vector signed int,
17200 vector unsigned int);
17201 vector unsigned int vec_sel (vector unsigned int,
17202 vector unsigned int,
17204 vector unsigned int vec_sel (vector unsigned int,
17205 vector unsigned int,
17206 vector unsigned int);
17207 vector bool int vec_sel (vector bool int,
17210 vector bool int vec_sel (vector bool int,
17212 vector unsigned int);
17213 vector signed short vec_sel (vector signed short,
17214 vector signed short,
17215 vector bool short);
17216 vector signed short vec_sel (vector signed short,
17217 vector signed short,
17218 vector unsigned short);
17219 vector unsigned short vec_sel (vector unsigned short,
17220 vector unsigned short,
17221 vector bool short);
17222 vector unsigned short vec_sel (vector unsigned short,
17223 vector unsigned short,
17224 vector unsigned short);
17225 vector bool short vec_sel (vector bool short,
17227 vector bool short);
17228 vector bool short vec_sel (vector bool short,
17230 vector unsigned short);
17231 vector signed char vec_sel (vector signed char,
17232 vector signed char,
17234 vector signed char vec_sel (vector signed char,
17235 vector signed char,
17236 vector unsigned char);
17237 vector unsigned char vec_sel (vector unsigned char,
17238 vector unsigned char,
17240 vector unsigned char vec_sel (vector unsigned char,
17241 vector unsigned char,
17242 vector unsigned char);
17243 vector bool char vec_sel (vector bool char,
17246 vector bool char vec_sel (vector bool char,
17248 vector unsigned char);
17250 vector signed long long vec_signed (vector double);
17251 vector signed int vec_signed (vector float);
17253 vector signed int vec_signede (vector double);
17254 vector signed int vec_signedo (vector double);
17255 vector signed int vec_signed2 (vector double, vector double);
17257 vector signed char vec_sl (vector signed char,
17258 vector unsigned char);
17259 vector unsigned char vec_sl (vector unsigned char,
17260 vector unsigned char);
17261 vector signed short vec_sl (vector signed short, vector unsigned short);
17262 vector unsigned short vec_sl (vector unsigned short,
17263 vector unsigned short);
17264 vector signed int vec_sl (vector signed int, vector unsigned int);
17265 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
17267 vector signed int vec_vslw (vector signed int, vector unsigned int);
17268 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
17270 vector signed short vec_vslh (vector signed short,
17271 vector unsigned short);
17272 vector unsigned short vec_vslh (vector unsigned short,
17273 vector unsigned short);
17275 vector signed char vec_vslb (vector signed char, vector unsigned char);
17276 vector unsigned char vec_vslb (vector unsigned char,
17277 vector unsigned char);
17279 vector float vec_sld (vector float, vector float, const int);
17280 vector double vec_sld (vector double, vector double, const int);
17282 vector signed int vec_sld (vector signed int,
17285 vector unsigned int vec_sld (vector unsigned int,
17286 vector unsigned int,
17288 vector bool int vec_sld (vector bool int,
17291 vector signed short vec_sld (vector signed short,
17292 vector signed short,
17294 vector unsigned short vec_sld (vector unsigned short,
17295 vector unsigned short,
17297 vector bool short vec_sld (vector bool short,
17300 vector pixel vec_sld (vector pixel,
17303 vector signed char vec_sld (vector signed char,
17304 vector signed char,
17306 vector unsigned char vec_sld (vector unsigned char,
17307 vector unsigned char,
17309 vector bool char vec_sld (vector bool char,
17312 vector bool long long int vec_sld (vector bool long long int,
17313 vector bool long long int, const int);
17314 vector long long int vec_sld (vector long long int,
17315 vector long long int, const int);
17316 vector unsigned long long int vec_sld (vector unsigned long long int,
17317 vector unsigned long long int,
17320 vector signed char vec_sldw (vector signed char,
17321 vector signed char,
17323 vector unsigned char vec_sldw (vector unsigned char,
17324 vector unsigned char,
17326 vector signed short vec_sldw (vector signed short,
17327 vector signed short,
17329 vector unsigned short vec_sldw (vector unsigned short,
17330 vector unsigned short,
17332 vector signed int vec_sldw (vector signed int,
17335 vector unsigned int vec_sldw (vector unsigned int,
17336 vector unsigned int,
17338 vector signed long long vec_sldw (vector signed long long,
17339 vector signed long long,
17341 vector unsigned long long vec_sldw (vector unsigned long long,
17342 vector unsigned long long,
17345 vector signed int vec_sll (vector signed int,
17346 vector unsigned int);
17347 vector signed int vec_sll (vector signed int,
17348 vector unsigned short);
17349 vector signed int vec_sll (vector signed int,
17350 vector unsigned char);
17351 vector unsigned int vec_sll (vector unsigned int,
17352 vector unsigned int);
17353 vector unsigned int vec_sll (vector unsigned int,
17354 vector unsigned short);
17355 vector unsigned int vec_sll (vector unsigned int,
17356 vector unsigned char);
17357 vector bool int vec_sll (vector bool int,
17358 vector unsigned int);
17359 vector bool int vec_sll (vector bool int,
17360 vector unsigned short);
17361 vector bool int vec_sll (vector bool int,
17362 vector unsigned char);
17363 vector signed short vec_sll (vector signed short,
17364 vector unsigned int);
17365 vector signed short vec_sll (vector signed short,
17366 vector unsigned short);
17367 vector signed short vec_sll (vector signed short,
17368 vector unsigned char);
17369 vector unsigned short vec_sll (vector unsigned short,
17370 vector unsigned int);
17371 vector unsigned short vec_sll (vector unsigned short,
17372 vector unsigned short);
17373 vector unsigned short vec_sll (vector unsigned short,
17374 vector unsigned char);
17375 vector long long int vec_sll (vector long long int,
17376 vector unsigned char);
17377 vector unsigned long long int vec_sll (vector unsigned long long int,
17378 vector unsigned char);
17379 vector bool short vec_sll (vector bool short, vector unsigned int);
17380 vector bool short vec_sll (vector bool short, vector unsigned short);
17381 vector bool short vec_sll (vector bool short, vector unsigned char);
17382 vector pixel vec_sll (vector pixel, vector unsigned int);
17383 vector pixel vec_sll (vector pixel, vector unsigned short);
17384 vector pixel vec_sll (vector pixel, vector unsigned char);
17385 vector signed char vec_sll (vector signed char, vector unsigned int);
17386 vector signed char vec_sll (vector signed char, vector unsigned short);
17387 vector signed char vec_sll (vector signed char, vector unsigned char);
17388 vector unsigned char vec_sll (vector unsigned char,
17389 vector unsigned int);
17390 vector unsigned char vec_sll (vector unsigned char,
17391 vector unsigned short);
17392 vector unsigned char vec_sll (vector unsigned char,
17393 vector unsigned char);
17394 vector bool char vec_sll (vector bool char, vector unsigned int);
17395 vector bool char vec_sll (vector bool char, vector unsigned short);
17396 vector bool char vec_sll (vector bool char, vector unsigned char);
17398 vector float vec_slo (vector float, vector signed char);
17399 vector float vec_slo (vector float, vector unsigned char);
17400 vector signed int vec_slo (vector signed int, vector signed char);
17401 vector signed int vec_slo (vector signed int, vector unsigned char);
17402 vector unsigned int vec_slo (vector unsigned int, vector signed char);
17403 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
17404 vector signed short vec_slo (vector signed short, vector signed char);
17405 vector signed short vec_slo (vector signed short, vector unsigned char);
17406 vector unsigned short vec_slo (vector unsigned short,
17407 vector signed char);
17408 vector unsigned short vec_slo (vector unsigned short,
17409 vector unsigned char);
17410 vector pixel vec_slo (vector pixel, vector signed char);
17411 vector pixel vec_slo (vector pixel, vector unsigned char);
17412 vector signed char vec_slo (vector signed char, vector signed char);
17413 vector signed char vec_slo (vector signed char, vector unsigned char);
17414 vector unsigned char vec_slo (vector unsigned char, vector signed char);
17415 vector unsigned char vec_slo (vector unsigned char,
17416 vector unsigned char);
17417 vector signed long long vec_slo (vector signed long long, vector signed char);
17418 vector signed long long vec_slo (vector signed long long, vector unsigned char);
17419 vector unsigned long long vec_slo (vector unsigned long long, vector signed char);
17420 vector unsigned long long vec_slo (vector unsigned long long, vector unsigned char);
17422 vector signed char vec_splat (vector signed char, const int);
17423 vector unsigned char vec_splat (vector unsigned char, const int);
17424 vector bool char vec_splat (vector bool char, const int);
17425 vector signed short vec_splat (vector signed short, const int);
17426 vector unsigned short vec_splat (vector unsigned short, const int);
17427 vector bool short vec_splat (vector bool short, const int);
17428 vector pixel vec_splat (vector pixel, const int);
17429 vector float vec_splat (vector float, const int);
17430 vector signed int vec_splat (vector signed int, const int);
17431 vector unsigned int vec_splat (vector unsigned int, const int);
17432 vector bool int vec_splat (vector bool int, const int);
17433 vector signed long vec_splat (vector signed long, const int);
17434 vector unsigned long vec_splat (vector unsigned long, const int);
17436 vector signed char vec_splats (signed char);
17437 vector unsigned char vec_splats (unsigned char);
17438 vector signed short vec_splats (signed short);
17439 vector unsigned short vec_splats (unsigned short);
17440 vector signed int vec_splats (signed int);
17441 vector unsigned int vec_splats (unsigned int);
17442 vector float vec_splats (float);
17444 vector float vec_vspltw (vector float, const int);
17445 vector signed int vec_vspltw (vector signed int, const int);
17446 vector unsigned int vec_vspltw (vector unsigned int, const int);
17447 vector bool int vec_vspltw (vector bool int, const int);
17449 vector bool short vec_vsplth (vector bool short, const int);
17450 vector signed short vec_vsplth (vector signed short, const int);
17451 vector unsigned short vec_vsplth (vector unsigned short, const int);
17452 vector pixel vec_vsplth (vector pixel, const int);
17454 vector signed char vec_vspltb (vector signed char, const int);
17455 vector unsigned char vec_vspltb (vector unsigned char, const int);
17456 vector bool char vec_vspltb (vector bool char, const int);
17458 vector signed char vec_splat_s8 (const int);
17460 vector signed short vec_splat_s16 (const int);
17462 vector signed int vec_splat_s32 (const int);
17464 vector unsigned char vec_splat_u8 (const int);
17466 vector unsigned short vec_splat_u16 (const int);
17468 vector unsigned int vec_splat_u32 (const int);
17470 vector signed char vec_sr (vector signed char, vector unsigned char);
17471 vector unsigned char vec_sr (vector unsigned char,
17472 vector unsigned char);
17473 vector signed short vec_sr (vector signed short,
17474 vector unsigned short);
17475 vector unsigned short vec_sr (vector unsigned short,
17476 vector unsigned short);
17477 vector signed int vec_sr (vector signed int, vector unsigned int);
17478 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
17480 vector signed int vec_vsrw (vector signed int, vector unsigned int);
17481 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
17483 vector signed short vec_vsrh (vector signed short,
17484 vector unsigned short);
17485 vector unsigned short vec_vsrh (vector unsigned short,
17486 vector unsigned short);
17488 vector signed char vec_vsrb (vector signed char, vector unsigned char);
17489 vector unsigned char vec_vsrb (vector unsigned char,
17490 vector unsigned char);
17492 vector signed char vec_sra (vector signed char, vector unsigned char);
17493 vector unsigned char vec_sra (vector unsigned char,
17494 vector unsigned char);
17495 vector signed short vec_sra (vector signed short,
17496 vector unsigned short);
17497 vector unsigned short vec_sra (vector unsigned short,
17498 vector unsigned short);
17499 vector signed int vec_sra (vector signed int, vector unsigned int);
17500 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
17502 vector signed int vec_vsraw (vector signed int, vector unsigned int);
17503 vector unsigned int vec_vsraw (vector unsigned int,
17504 vector unsigned int);
17506 vector signed short vec_vsrah (vector signed short,
17507 vector unsigned short);
17508 vector unsigned short vec_vsrah (vector unsigned short,
17509 vector unsigned short);
17511 vector signed char vec_vsrab (vector signed char, vector unsigned char);
17512 vector unsigned char vec_vsrab (vector unsigned char,
17513 vector unsigned char);
17515 vector signed int vec_srl (vector signed int, vector unsigned int);
17516 vector signed int vec_srl (vector signed int, vector unsigned short);
17517 vector signed int vec_srl (vector signed int, vector unsigned char);
17518 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
17519 vector unsigned int vec_srl (vector unsigned int,
17520 vector unsigned short);
17521 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
17522 vector bool int vec_srl (vector bool int, vector unsigned int);
17523 vector bool int vec_srl (vector bool int, vector unsigned short);
17524 vector bool int vec_srl (vector bool int, vector unsigned char);
17525 vector signed short vec_srl (vector signed short, vector unsigned int);
17526 vector signed short vec_srl (vector signed short,
17527 vector unsigned short);
17528 vector signed short vec_srl (vector signed short, vector unsigned char);
17529 vector unsigned short vec_srl (vector unsigned short,
17530 vector unsigned int);
17531 vector unsigned short vec_srl (vector unsigned short,
17532 vector unsigned short);
17533 vector unsigned short vec_srl (vector unsigned short,
17534 vector unsigned char);
17535 vector long long int vec_srl (vector long long int,
17536 vector unsigned char);
17537 vector unsigned long long int vec_srl (vector unsigned long long int,
17538 vector unsigned char);
17539 vector bool short vec_srl (vector bool short, vector unsigned int);
17540 vector bool short vec_srl (vector bool short, vector unsigned short);
17541 vector bool short vec_srl (vector bool short, vector unsigned char);
17542 vector pixel vec_srl (vector pixel, vector unsigned int);
17543 vector pixel vec_srl (vector pixel, vector unsigned short);
17544 vector pixel vec_srl (vector pixel, vector unsigned char);
17545 vector signed char vec_srl (vector signed char, vector unsigned int);
17546 vector signed char vec_srl (vector signed char, vector unsigned short);
17547 vector signed char vec_srl (vector signed char, vector unsigned char);
17548 vector unsigned char vec_srl (vector unsigned char,
17549 vector unsigned int);
17550 vector unsigned char vec_srl (vector unsigned char,
17551 vector unsigned short);
17552 vector unsigned char vec_srl (vector unsigned char,
17553 vector unsigned char);
17554 vector bool char vec_srl (vector bool char, vector unsigned int);
17555 vector bool char vec_srl (vector bool char, vector unsigned short);
17556 vector bool char vec_srl (vector bool char, vector unsigned char);
17558 vector float vec_sro (vector float, vector signed char);
17559 vector float vec_sro (vector float, vector unsigned char);
17560 vector signed int vec_sro (vector signed int, vector signed char);
17561 vector signed int vec_sro (vector signed int, vector unsigned char);
17562 vector unsigned int vec_sro (vector unsigned int, vector signed char);
17563 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
17564 vector signed short vec_sro (vector signed short, vector signed char);
17565 vector signed short vec_sro (vector signed short, vector unsigned char);
17566 vector unsigned short vec_sro (vector unsigned short,
17567 vector signed char);
17568 vector unsigned short vec_sro (vector unsigned short,
17569 vector unsigned char);
17570 vector long long int vec_sro (vector long long int,
17572 vector long long int vec_sro (vector long long int,
17573 vector unsigned char);
17574 vector unsigned long long int vec_sro (vector unsigned long long int,
17576 vector unsigned long long int vec_sro (vector unsigned long long int,
17577 vector unsigned char);
17578 vector pixel vec_sro (vector pixel, vector signed char);
17579 vector pixel vec_sro (vector pixel, vector unsigned char);
17580 vector signed char vec_sro (vector signed char, vector signed char);
17581 vector signed char vec_sro (vector signed char, vector unsigned char);
17582 vector unsigned char vec_sro (vector unsigned char, vector signed char);
17583 vector unsigned char vec_sro (vector unsigned char,
17584 vector unsigned char);
17586 void vec_st (vector float, int, vector float *);
17587 void vec_st (vector float, int, float *);
17588 void vec_st (vector signed int, int, vector signed int *);
17589 void vec_st (vector signed int, int, int *);
17590 void vec_st (vector unsigned int, int, vector unsigned int *);
17591 void vec_st (vector unsigned int, int, unsigned int *);
17592 void vec_st (vector bool int, int, vector bool int *);
17593 void vec_st (vector bool int, int, unsigned int *);
17594 void vec_st (vector bool int, int, int *);
17595 void vec_st (vector signed short, int, vector signed short *);
17596 void vec_st (vector signed short, int, short *);
17597 void vec_st (vector unsigned short, int, vector unsigned short *);
17598 void vec_st (vector unsigned short, int, unsigned short *);
17599 void vec_st (vector bool short, int, vector bool short *);
17600 void vec_st (vector bool short, int, unsigned short *);
17601 void vec_st (vector pixel, int, vector pixel *);
17602 void vec_st (vector bool short, int, short *);
17603 void vec_st (vector signed char, int, vector signed char *);
17604 void vec_st (vector signed char, int, signed char *);
17605 void vec_st (vector unsigned char, int, vector unsigned char *);
17606 void vec_st (vector unsigned char, int, unsigned char *);
17607 void vec_st (vector bool char, int, vector bool char *);
17608 void vec_st (vector bool char, int, unsigned char *);
17609 void vec_st (vector bool char, int, signed char *);
17611 void vec_ste (vector signed char, int, signed char *);
17612 void vec_ste (vector unsigned char, int, unsigned char *);
17613 void vec_ste (vector bool char, int, signed char *);
17614 void vec_ste (vector bool char, int, unsigned char *);
17615 void vec_ste (vector signed short, int, short *);
17616 void vec_ste (vector unsigned short, int, unsigned short *);
17617 void vec_ste (vector bool short, int, short *);
17618 void vec_ste (vector bool short, int, unsigned short *);
17619 void vec_ste (vector pixel, int, short *);
17620 void vec_ste (vector pixel, int, unsigned short *);
17621 void vec_ste (vector float, int, float *);
17622 void vec_ste (vector signed int, int, int *);
17623 void vec_ste (vector unsigned int, int, unsigned int *);
17624 void vec_ste (vector bool int, int, int *);
17625 void vec_ste (vector bool int, int, unsigned int *);
17627 void vec_stvewx (vector float, int, float *);
17628 void vec_stvewx (vector signed int, int, int *);
17629 void vec_stvewx (vector unsigned int, int, unsigned int *);
17630 void vec_stvewx (vector bool int, int, int *);
17631 void vec_stvewx (vector bool int, int, unsigned int *);
17633 void vec_stvehx (vector signed short, int, short *);
17634 void vec_stvehx (vector unsigned short, int, unsigned short *);
17635 void vec_stvehx (vector bool short, int, short *);
17636 void vec_stvehx (vector bool short, int, unsigned short *);
17638 void vec_stvebx (vector signed char, int, signed char *);
17639 void vec_stvebx (vector unsigned char, int, unsigned char *);
17640 void vec_stvebx (vector bool char, int, signed char *);
17641 void vec_stvebx (vector bool char, int, unsigned char *);
17643 void vec_stl (vector float, int, vector float *);
17644 void vec_stl (vector float, int, float *);
17645 void vec_stl (vector signed int, int, vector signed int *);
17646 void vec_stl (vector signed int, int, int *);
17647 void vec_stl (vector unsigned int, int, vector unsigned int *);
17648 void vec_stl (vector unsigned int, int, unsigned int *);
17649 void vec_stl (vector bool int, int, vector bool int *);
17650 void vec_stl (vector bool int, int, unsigned int *);
17651 void vec_stl (vector bool int, int, int *);
17652 void vec_stl (vector signed short, int, vector signed short *);
17653 void vec_stl (vector signed short, int, short *);
17654 void vec_stl (vector unsigned short, int, vector unsigned short *);
17655 void vec_stl (vector unsigned short, int, unsigned short *);
17656 void vec_stl (vector bool short, int, vector bool short *);
17657 void vec_stl (vector bool short, int, unsigned short *);
17658 void vec_stl (vector bool short, int, short *);
17659 void vec_stl (vector pixel, int, vector pixel *);
17660 void vec_stl (vector signed char, int, vector signed char *);
17661 void vec_stl (vector signed char, int, signed char *);
17662 void vec_stl (vector unsigned char, int, vector unsigned char *);
17663 void vec_stl (vector unsigned char, int, unsigned char *);
17664 void vec_stl (vector bool char, int, vector bool char *);
17665 void vec_stl (vector bool char, int, unsigned char *);
17666 void vec_stl (vector bool char, int, signed char *);
17668 vector signed char vec_sub (vector bool char, vector signed char);
17669 vector signed char vec_sub (vector signed char, vector bool char);
17670 vector signed char vec_sub (vector signed char, vector signed char);
17671 vector unsigned char vec_sub (vector bool char, vector unsigned char);
17672 vector unsigned char vec_sub (vector unsigned char, vector bool char);
17673 vector unsigned char vec_sub (vector unsigned char,
17674 vector unsigned char);
17675 vector signed short vec_sub (vector bool short, vector signed short);
17676 vector signed short vec_sub (vector signed short, vector bool short);
17677 vector signed short vec_sub (vector signed short, vector signed short);
17678 vector unsigned short vec_sub (vector bool short,
17679 vector unsigned short);
17680 vector unsigned short vec_sub (vector unsigned short,
17681 vector bool short);
17682 vector unsigned short vec_sub (vector unsigned short,
17683 vector unsigned short);
17684 vector signed int vec_sub (vector bool int, vector signed int);
17685 vector signed int vec_sub (vector signed int, vector bool int);
17686 vector signed int vec_sub (vector signed int, vector signed int);
17687 vector unsigned int vec_sub (vector bool int, vector unsigned int);
17688 vector unsigned int vec_sub (vector unsigned int, vector bool int);
17689 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
17690 vector float vec_sub (vector float, vector float);
17692 vector float vec_vsubfp (vector float, vector float);
17694 vector signed int vec_vsubuwm (vector bool int, vector signed int);
17695 vector signed int vec_vsubuwm (vector signed int, vector bool int);
17696 vector signed int vec_vsubuwm (vector signed int, vector signed int);
17697 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
17698 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
17699 vector unsigned int vec_vsubuwm (vector unsigned int,
17700 vector unsigned int);
17702 vector signed short vec_vsubuhm (vector bool short,
17703 vector signed short);
17704 vector signed short vec_vsubuhm (vector signed short,
17705 vector bool short);
17706 vector signed short vec_vsubuhm (vector signed short,
17707 vector signed short);
17708 vector unsigned short vec_vsubuhm (vector bool short,
17709 vector unsigned short);
17710 vector unsigned short vec_vsubuhm (vector unsigned short,
17711 vector bool short);
17712 vector unsigned short vec_vsubuhm (vector unsigned short,
17713 vector unsigned short);
17715 vector signed char vec_vsububm (vector bool char, vector signed char);
17716 vector signed char vec_vsububm (vector signed char, vector bool char);
17717 vector signed char vec_vsububm (vector signed char, vector signed char);
17718 vector unsigned char vec_vsububm (vector bool char,
17719 vector unsigned char);
17720 vector unsigned char vec_vsububm (vector unsigned char,
17722 vector unsigned char vec_vsububm (vector unsigned char,
17723 vector unsigned char);
17725 vector signed int vec_subc (vector signed int, vector signed int);
17726 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
17727 vector signed __int128 vec_subc (vector signed __int128,
17728 vector signed __int128);
17729 vector unsigned __int128 vec_subc (vector unsigned __int128,
17730 vector unsigned __int128);
17732 vector signed int vec_sube (vector signed int, vector signed int,
17733 vector signed int);
17734 vector unsigned int vec_sube (vector unsigned int, vector unsigned int,
17735 vector unsigned int);
17736 vector signed __int128 vec_sube (vector signed __int128,
17737 vector signed __int128,
17738 vector signed __int128);
17739 vector unsigned __int128 vec_sube (vector unsigned __int128,
17740 vector unsigned __int128,
17741 vector unsigned __int128);
17743 vector signed int vec_subec (vector signed int, vector signed int,
17744 vector signed int);
17745 vector unsigned int vec_subec (vector unsigned int, vector unsigned int,
17746 vector unsigned int);
17747 vector signed __int128 vec_subec (vector signed __int128,
17748 vector signed __int128,
17749 vector signed __int128);
17750 vector unsigned __int128 vec_subec (vector unsigned __int128,
17751 vector unsigned __int128,
17752 vector unsigned __int128);
17754 vector unsigned char vec_subs (vector bool char, vector unsigned char);
17755 vector unsigned char vec_subs (vector unsigned char, vector bool char);
17756 vector unsigned char vec_subs (vector unsigned char,
17757 vector unsigned char);
17758 vector signed char vec_subs (vector bool char, vector signed char);
17759 vector signed char vec_subs (vector signed char, vector bool char);
17760 vector signed char vec_subs (vector signed char, vector signed char);
17761 vector unsigned short vec_subs (vector bool short,
17762 vector unsigned short);
17763 vector unsigned short vec_subs (vector unsigned short,
17764 vector bool short);
17765 vector unsigned short vec_subs (vector unsigned short,
17766 vector unsigned short);
17767 vector signed short vec_subs (vector bool short, vector signed short);
17768 vector signed short vec_subs (vector signed short, vector bool short);
17769 vector signed short vec_subs (vector signed short, vector signed short);
17770 vector unsigned int vec_subs (vector bool int, vector unsigned int);
17771 vector unsigned int vec_subs (vector unsigned int, vector bool int);
17772 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
17773 vector signed int vec_subs (vector bool int, vector signed int);
17774 vector signed int vec_subs (vector signed int, vector bool int);
17775 vector signed int vec_subs (vector signed int, vector signed int);
17777 vector signed int vec_vsubsws (vector bool int, vector signed int);
17778 vector signed int vec_vsubsws (vector signed int, vector bool int);
17779 vector signed int vec_vsubsws (vector signed int, vector signed int);
17781 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
17782 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
17783 vector unsigned int vec_vsubuws (vector unsigned int,
17784 vector unsigned int);
17786 vector signed short vec_vsubshs (vector bool short,
17787 vector signed short);
17788 vector signed short vec_vsubshs (vector signed short,
17789 vector bool short);
17790 vector signed short vec_vsubshs (vector signed short,
17791 vector signed short);
17793 vector unsigned short vec_vsubuhs (vector bool short,
17794 vector unsigned short);
17795 vector unsigned short vec_vsubuhs (vector unsigned short,
17796 vector bool short);
17797 vector unsigned short vec_vsubuhs (vector unsigned short,
17798 vector unsigned short);
17800 vector signed char vec_vsubsbs (vector bool char, vector signed char);
17801 vector signed char vec_vsubsbs (vector signed char, vector bool char);
17802 vector signed char vec_vsubsbs (vector signed char, vector signed char);
17804 vector unsigned char vec_vsububs (vector bool char,
17805 vector unsigned char);
17806 vector unsigned char vec_vsububs (vector unsigned char,
17808 vector unsigned char vec_vsububs (vector unsigned char,
17809 vector unsigned char);
17811 vector unsigned int vec_sum4s (vector unsigned char,
17812 vector unsigned int);
17813 vector signed int vec_sum4s (vector signed char, vector signed int);
17814 vector signed int vec_sum4s (vector signed short, vector signed int);
17816 vector signed int vec_vsum4shs (vector signed short, vector signed int);
17818 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
17820 vector unsigned int vec_vsum4ubs (vector unsigned char,
17821 vector unsigned int);
17823 vector signed int vec_sum2s (vector signed int, vector signed int);
17825 vector signed int vec_sums (vector signed int, vector signed int);
17827 vector float vec_trunc (vector float);
17829 vector signed long long vec_unsigned (vector double);
17830 vector signed int vec_unsigned (vector float);
17832 vector signed int vec_unsignede (vector double);
17833 vector signed int vec_unsignedo (vector double);
17834 vector signed int vec_unsigned2 (vector double, vector double);
17836 vector signed short vec_unpackh (vector signed char);
17837 vector bool short vec_unpackh (vector bool char);
17838 vector signed int vec_unpackh (vector signed short);
17839 vector bool int vec_unpackh (vector bool short);
17840 vector unsigned int vec_unpackh (vector pixel);
17841 vector double vec_unpackh (vector float);
17843 vector bool int vec_vupkhsh (vector bool short);
17844 vector signed int vec_vupkhsh (vector signed short);
17846 vector unsigned int vec_vupkhpx (vector pixel);
17848 vector bool short vec_vupkhsb (vector bool char);
17849 vector signed short vec_vupkhsb (vector signed char);
17851 vector signed short vec_unpackl (vector signed char);
17852 vector bool short vec_unpackl (vector bool char);
17853 vector unsigned int vec_unpackl (vector pixel);
17854 vector signed int vec_unpackl (vector signed short);
17855 vector bool int vec_unpackl (vector bool short);
17856 vector double vec_unpackl (vector float);
17858 vector unsigned int vec_vupklpx (vector pixel);
17860 vector bool int vec_vupklsh (vector bool short);
17861 vector signed int vec_vupklsh (vector signed short);
17863 vector bool short vec_vupklsb (vector bool char);
17864 vector signed short vec_vupklsb (vector signed char);
17866 vector float vec_xor (vector float, vector float);
17867 vector float vec_xor (vector float, vector bool int);
17868 vector float vec_xor (vector bool int, vector float);
17869 vector bool int vec_xor (vector bool int, vector bool int);
17870 vector signed int vec_xor (vector bool int, vector signed int);
17871 vector signed int vec_xor (vector signed int, vector bool int);
17872 vector signed int vec_xor (vector signed int, vector signed int);
17873 vector unsigned int vec_xor (vector bool int, vector unsigned int);
17874 vector unsigned int vec_xor (vector unsigned int, vector bool int);
17875 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
17876 vector bool short vec_xor (vector bool short, vector bool short);
17877 vector signed short vec_xor (vector bool short, vector signed short);
17878 vector signed short vec_xor (vector signed short, vector bool short);
17879 vector signed short vec_xor (vector signed short, vector signed short);
17880 vector unsigned short vec_xor (vector bool short,
17881 vector unsigned short);
17882 vector unsigned short vec_xor (vector unsigned short,
17883 vector bool short);
17884 vector unsigned short vec_xor (vector unsigned short,
17885 vector unsigned short);
17886 vector signed char vec_xor (vector bool char, vector signed char);
17887 vector bool char vec_xor (vector bool char, vector bool char);
17888 vector signed char vec_xor (vector signed char, vector bool char);
17889 vector signed char vec_xor (vector signed char, vector signed char);
17890 vector unsigned char vec_xor (vector bool char, vector unsigned char);
17891 vector unsigned char vec_xor (vector unsigned char, vector bool char);
17892 vector unsigned char vec_xor (vector unsigned char,
17893 vector unsigned char);
17895 int vec_all_eq (vector signed char, vector bool char);
17896 int vec_all_eq (vector signed char, vector signed char);
17897 int vec_all_eq (vector unsigned char, vector bool char);
17898 int vec_all_eq (vector unsigned char, vector unsigned char);
17899 int vec_all_eq (vector bool char, vector bool char);
17900 int vec_all_eq (vector bool char, vector unsigned char);
17901 int vec_all_eq (vector bool char, vector signed char);
17902 int vec_all_eq (vector signed short, vector bool short);
17903 int vec_all_eq (vector signed short, vector signed short);
17904 int vec_all_eq (vector unsigned short, vector bool short);
17905 int vec_all_eq (vector unsigned short, vector unsigned short);
17906 int vec_all_eq (vector bool short, vector bool short);
17907 int vec_all_eq (vector bool short, vector unsigned short);
17908 int vec_all_eq (vector bool short, vector signed short);
17909 int vec_all_eq (vector pixel, vector pixel);
17910 int vec_all_eq (vector signed int, vector bool int);
17911 int vec_all_eq (vector signed int, vector signed int);
17912 int vec_all_eq (vector unsigned int, vector bool int);
17913 int vec_all_eq (vector unsigned int, vector unsigned int);
17914 int vec_all_eq (vector bool int, vector bool int);
17915 int vec_all_eq (vector bool int, vector unsigned int);
17916 int vec_all_eq (vector bool int, vector signed int);
17917 int vec_all_eq (vector float, vector float);
17919 int vec_all_ge (vector bool char, vector unsigned char);
17920 int vec_all_ge (vector unsigned char, vector bool char);
17921 int vec_all_ge (vector unsigned char, vector unsigned char);
17922 int vec_all_ge (vector bool char, vector signed char);
17923 int vec_all_ge (vector signed char, vector bool char);
17924 int vec_all_ge (vector signed char, vector signed char);
17925 int vec_all_ge (vector bool short, vector unsigned short);
17926 int vec_all_ge (vector unsigned short, vector bool short);
17927 int vec_all_ge (vector unsigned short, vector unsigned short);
17928 int vec_all_ge (vector signed short, vector signed short);
17929 int vec_all_ge (vector bool short, vector signed short);
17930 int vec_all_ge (vector signed short, vector bool short);
17931 int vec_all_ge (vector bool int, vector unsigned int);
17932 int vec_all_ge (vector unsigned int, vector bool int);
17933 int vec_all_ge (vector unsigned int, vector unsigned int);
17934 int vec_all_ge (vector bool int, vector signed int);
17935 int vec_all_ge (vector signed int, vector bool int);
17936 int vec_all_ge (vector signed int, vector signed int);
17937 int vec_all_ge (vector float, vector float);
17939 int vec_all_gt (vector bool char, vector unsigned char);
17940 int vec_all_gt (vector unsigned char, vector bool char);
17941 int vec_all_gt (vector unsigned char, vector unsigned char);
17942 int vec_all_gt (vector bool char, vector signed char);
17943 int vec_all_gt (vector signed char, vector bool char);
17944 int vec_all_gt (vector signed char, vector signed char);
17945 int vec_all_gt (vector bool short, vector unsigned short);
17946 int vec_all_gt (vector unsigned short, vector bool short);
17947 int vec_all_gt (vector unsigned short, vector unsigned short);
17948 int vec_all_gt (vector bool short, vector signed short);
17949 int vec_all_gt (vector signed short, vector bool short);
17950 int vec_all_gt (vector signed short, vector signed short);
17951 int vec_all_gt (vector bool int, vector unsigned int);
17952 int vec_all_gt (vector unsigned int, vector bool int);
17953 int vec_all_gt (vector unsigned int, vector unsigned int);
17954 int vec_all_gt (vector bool int, vector signed int);
17955 int vec_all_gt (vector signed int, vector bool int);
17956 int vec_all_gt (vector signed int, vector signed int);
17957 int vec_all_gt (vector float, vector float);
17959 int vec_all_in (vector float, vector float);
17961 int vec_all_le (vector bool char, vector unsigned char);
17962 int vec_all_le (vector unsigned char, vector bool char);
17963 int vec_all_le (vector unsigned char, vector unsigned char);
17964 int vec_all_le (vector bool char, vector signed char);
17965 int vec_all_le (vector signed char, vector bool char);
17966 int vec_all_le (vector signed char, vector signed char);
17967 int vec_all_le (vector bool short, vector unsigned short);
17968 int vec_all_le (vector unsigned short, vector bool short);
17969 int vec_all_le (vector unsigned short, vector unsigned short);
17970 int vec_all_le (vector bool short, vector signed short);
17971 int vec_all_le (vector signed short, vector bool short);
17972 int vec_all_le (vector signed short, vector signed short);
17973 int vec_all_le (vector bool int, vector unsigned int);
17974 int vec_all_le (vector unsigned int, vector bool int);
17975 int vec_all_le (vector unsigned int, vector unsigned int);
17976 int vec_all_le (vector bool int, vector signed int);
17977 int vec_all_le (vector signed int, vector bool int);
17978 int vec_all_le (vector signed int, vector signed int);
17979 int vec_all_le (vector float, vector float);
17981 int vec_all_lt (vector bool char, vector unsigned char);
17982 int vec_all_lt (vector unsigned char, vector bool char);
17983 int vec_all_lt (vector unsigned char, vector unsigned char);
17984 int vec_all_lt (vector bool char, vector signed char);
17985 int vec_all_lt (vector signed char, vector bool char);
17986 int vec_all_lt (vector signed char, vector signed char);
17987 int vec_all_lt (vector bool short, vector unsigned short);
17988 int vec_all_lt (vector unsigned short, vector bool short);
17989 int vec_all_lt (vector unsigned short, vector unsigned short);
17990 int vec_all_lt (vector bool short, vector signed short);
17991 int vec_all_lt (vector signed short, vector bool short);
17992 int vec_all_lt (vector signed short, vector signed short);
17993 int vec_all_lt (vector bool int, vector unsigned int);
17994 int vec_all_lt (vector unsigned int, vector bool int);
17995 int vec_all_lt (vector unsigned int, vector unsigned int);
17996 int vec_all_lt (vector bool int, vector signed int);
17997 int vec_all_lt (vector signed int, vector bool int);
17998 int vec_all_lt (vector signed int, vector signed int);
17999 int vec_all_lt (vector float, vector float);
18001 int vec_all_nan (vector float);
18003 int vec_all_ne (vector signed char, vector bool char);
18004 int vec_all_ne (vector signed char, vector signed char);
18005 int vec_all_ne (vector unsigned char, vector bool char);
18006 int vec_all_ne (vector unsigned char, vector unsigned char);
18007 int vec_all_ne (vector bool char, vector bool char);
18008 int vec_all_ne (vector bool char, vector unsigned char);
18009 int vec_all_ne (vector bool char, vector signed char);
18010 int vec_all_ne (vector signed short, vector bool short);
18011 int vec_all_ne (vector signed short, vector signed short);
18012 int vec_all_ne (vector unsigned short, vector bool short);
18013 int vec_all_ne (vector unsigned short, vector unsigned short);
18014 int vec_all_ne (vector bool short, vector bool short);
18015 int vec_all_ne (vector bool short, vector unsigned short);
18016 int vec_all_ne (vector bool short, vector signed short);
18017 int vec_all_ne (vector pixel, vector pixel);
18018 int vec_all_ne (vector signed int, vector bool int);
18019 int vec_all_ne (vector signed int, vector signed int);
18020 int vec_all_ne (vector unsigned int, vector bool int);
18021 int vec_all_ne (vector unsigned int, vector unsigned int);
18022 int vec_all_ne (vector bool int, vector bool int);
18023 int vec_all_ne (vector bool int, vector unsigned int);
18024 int vec_all_ne (vector bool int, vector signed int);
18025 int vec_all_ne (vector float, vector float);
18027 int vec_all_nge (vector float, vector float);
18029 int vec_all_ngt (vector float, vector float);
18031 int vec_all_nle (vector float, vector float);
18033 int vec_all_nlt (vector float, vector float);
18035 int vec_all_numeric (vector float);
18037 int vec_any_eq (vector signed char, vector bool char);
18038 int vec_any_eq (vector signed char, vector signed char);
18039 int vec_any_eq (vector unsigned char, vector bool char);
18040 int vec_any_eq (vector unsigned char, vector unsigned char);
18041 int vec_any_eq (vector bool char, vector bool char);
18042 int vec_any_eq (vector bool char, vector unsigned char);
18043 int vec_any_eq (vector bool char, vector signed char);
18044 int vec_any_eq (vector signed short, vector bool short);
18045 int vec_any_eq (vector signed short, vector signed short);
18046 int vec_any_eq (vector unsigned short, vector bool short);
18047 int vec_any_eq (vector unsigned short, vector unsigned short);
18048 int vec_any_eq (vector bool short, vector bool short);
18049 int vec_any_eq (vector bool short, vector unsigned short);
18050 int vec_any_eq (vector bool short, vector signed short);
18051 int vec_any_eq (vector pixel, vector pixel);
18052 int vec_any_eq (vector signed int, vector bool int);
18053 int vec_any_eq (vector signed int, vector signed int);
18054 int vec_any_eq (vector unsigned int, vector bool int);
18055 int vec_any_eq (vector unsigned int, vector unsigned int);
18056 int vec_any_eq (vector bool int, vector bool int);
18057 int vec_any_eq (vector bool int, vector unsigned int);
18058 int vec_any_eq (vector bool int, vector signed int);
18059 int vec_any_eq (vector float, vector float);
18061 int vec_any_ge (vector signed char, vector bool char);
18062 int vec_any_ge (vector unsigned char, vector bool char);
18063 int vec_any_ge (vector unsigned char, vector unsigned char);
18064 int vec_any_ge (vector signed char, vector signed char);
18065 int vec_any_ge (vector bool char, vector unsigned char);
18066 int vec_any_ge (vector bool char, vector signed char);
18067 int vec_any_ge (vector unsigned short, vector bool short);
18068 int vec_any_ge (vector unsigned short, vector unsigned short);
18069 int vec_any_ge (vector signed short, vector signed short);
18070 int vec_any_ge (vector signed short, vector bool short);
18071 int vec_any_ge (vector bool short, vector unsigned short);
18072 int vec_any_ge (vector bool short, vector signed short);
18073 int vec_any_ge (vector signed int, vector bool int);
18074 int vec_any_ge (vector unsigned int, vector bool int);
18075 int vec_any_ge (vector unsigned int, vector unsigned int);
18076 int vec_any_ge (vector signed int, vector signed int);
18077 int vec_any_ge (vector bool int, vector unsigned int);
18078 int vec_any_ge (vector bool int, vector signed int);
18079 int vec_any_ge (vector float, vector float);
18081 int vec_any_gt (vector bool char, vector unsigned char);
18082 int vec_any_gt (vector unsigned char, vector bool char);
18083 int vec_any_gt (vector unsigned char, vector unsigned char);
18084 int vec_any_gt (vector bool char, vector signed char);
18085 int vec_any_gt (vector signed char, vector bool char);
18086 int vec_any_gt (vector signed char, vector signed char);
18087 int vec_any_gt (vector bool short, vector unsigned short);
18088 int vec_any_gt (vector unsigned short, vector bool short);
18089 int vec_any_gt (vector unsigned short, vector unsigned short);
18090 int vec_any_gt (vector bool short, vector signed short);
18091 int vec_any_gt (vector signed short, vector bool short);
18092 int vec_any_gt (vector signed short, vector signed short);
18093 int vec_any_gt (vector bool int, vector unsigned int);
18094 int vec_any_gt (vector unsigned int, vector bool int);
18095 int vec_any_gt (vector unsigned int, vector unsigned int);
18096 int vec_any_gt (vector bool int, vector signed int);
18097 int vec_any_gt (vector signed int, vector bool int);
18098 int vec_any_gt (vector signed int, vector signed int);
18099 int vec_any_gt (vector float, vector float);
18101 int vec_any_le (vector bool char, vector unsigned char);
18102 int vec_any_le (vector unsigned char, vector bool char);
18103 int vec_any_le (vector unsigned char, vector unsigned char);
18104 int vec_any_le (vector bool char, vector signed char);
18105 int vec_any_le (vector signed char, vector bool char);
18106 int vec_any_le (vector signed char, vector signed char);
18107 int vec_any_le (vector bool short, vector unsigned short);
18108 int vec_any_le (vector unsigned short, vector bool short);
18109 int vec_any_le (vector unsigned short, vector unsigned short);
18110 int vec_any_le (vector bool short, vector signed short);
18111 int vec_any_le (vector signed short, vector bool short);
18112 int vec_any_le (vector signed short, vector signed short);
18113 int vec_any_le (vector bool int, vector unsigned int);
18114 int vec_any_le (vector unsigned int, vector bool int);
18115 int vec_any_le (vector unsigned int, vector unsigned int);
18116 int vec_any_le (vector bool int, vector signed int);
18117 int vec_any_le (vector signed int, vector bool int);
18118 int vec_any_le (vector signed int, vector signed int);
18119 int vec_any_le (vector float, vector float);
18121 int vec_any_lt (vector bool char, vector unsigned char);
18122 int vec_any_lt (vector unsigned char, vector bool char);
18123 int vec_any_lt (vector unsigned char, vector unsigned char);
18124 int vec_any_lt (vector bool char, vector signed char);
18125 int vec_any_lt (vector signed char, vector bool char);
18126 int vec_any_lt (vector signed char, vector signed char);
18127 int vec_any_lt (vector bool short, vector unsigned short);
18128 int vec_any_lt (vector unsigned short, vector bool short);
18129 int vec_any_lt (vector unsigned short, vector unsigned short);
18130 int vec_any_lt (vector bool short, vector signed short);
18131 int vec_any_lt (vector signed short, vector bool short);
18132 int vec_any_lt (vector signed short, vector signed short);
18133 int vec_any_lt (vector bool int, vector unsigned int);
18134 int vec_any_lt (vector unsigned int, vector bool int);
18135 int vec_any_lt (vector unsigned int, vector unsigned int);
18136 int vec_any_lt (vector bool int, vector signed int);
18137 int vec_any_lt (vector signed int, vector bool int);
18138 int vec_any_lt (vector signed int, vector signed int);
18139 int vec_any_lt (vector float, vector float);
18141 int vec_any_nan (vector float);
18143 int vec_any_ne (vector signed char, vector bool char);
18144 int vec_any_ne (vector signed char, vector signed char);
18145 int vec_any_ne (vector unsigned char, vector bool char);
18146 int vec_any_ne (vector unsigned char, vector unsigned char);
18147 int vec_any_ne (vector bool char, vector bool char);
18148 int vec_any_ne (vector bool char, vector unsigned char);
18149 int vec_any_ne (vector bool char, vector signed char);
18150 int vec_any_ne (vector signed short, vector bool short);
18151 int vec_any_ne (vector signed short, vector signed short);
18152 int vec_any_ne (vector unsigned short, vector bool short);
18153 int vec_any_ne (vector unsigned short, vector unsigned short);
18154 int vec_any_ne (vector bool short, vector bool short);
18155 int vec_any_ne (vector bool short, vector unsigned short);
18156 int vec_any_ne (vector bool short, vector signed short);
18157 int vec_any_ne (vector pixel, vector pixel);
18158 int vec_any_ne (vector signed int, vector bool int);
18159 int vec_any_ne (vector signed int, vector signed int);
18160 int vec_any_ne (vector unsigned int, vector bool int);
18161 int vec_any_ne (vector unsigned int, vector unsigned int);
18162 int vec_any_ne (vector bool int, vector bool int);
18163 int vec_any_ne (vector bool int, vector unsigned int);
18164 int vec_any_ne (vector bool int, vector signed int);
18165 int vec_any_ne (vector float, vector float);
18167 int vec_any_nge (vector float, vector float);
18169 int vec_any_ngt (vector float, vector float);
18171 int vec_any_nle (vector float, vector float);
18173 int vec_any_nlt (vector float, vector float);
18175 int vec_any_numeric (vector float);
18177 int vec_any_out (vector float, vector float);
18180 If the vector/scalar (VSX) instruction set is available, the following
18181 additional functions are available:
18184 vector double vec_abs (vector double);
18185 vector double vec_add (vector double, vector double);
18186 vector double vec_and (vector double, vector double);
18187 vector double vec_and (vector double, vector bool long);
18188 vector double vec_and (vector bool long, vector double);
18189 vector long vec_and (vector long, vector long);
18190 vector long vec_and (vector long, vector bool long);
18191 vector long vec_and (vector bool long, vector long);
18192 vector unsigned long vec_and (vector unsigned long, vector unsigned long);
18193 vector unsigned long vec_and (vector unsigned long, vector bool long);
18194 vector unsigned long vec_and (vector bool long, vector unsigned long);
18195 vector double vec_andc (vector double, vector double);
18196 vector double vec_andc (vector double, vector bool long);
18197 vector double vec_andc (vector bool long, vector double);
18198 vector long vec_andc (vector long, vector long);
18199 vector long vec_andc (vector long, vector bool long);
18200 vector long vec_andc (vector bool long, vector long);
18201 vector unsigned long vec_andc (vector unsigned long, vector unsigned long);
18202 vector unsigned long vec_andc (vector unsigned long, vector bool long);
18203 vector unsigned long vec_andc (vector bool long, vector unsigned long);
18204 vector double vec_ceil (vector double);
18205 vector bool long vec_cmpeq (vector double, vector double);
18206 vector bool long vec_cmpge (vector double, vector double);
18207 vector bool long vec_cmpgt (vector double, vector double);
18208 vector bool long vec_cmple (vector double, vector double);
18209 vector bool long vec_cmplt (vector double, vector double);
18210 vector double vec_cpsgn (vector double, vector double);
18211 vector float vec_div (vector float, vector float);
18212 vector double vec_div (vector double, vector double);
18213 vector long vec_div (vector long, vector long);
18214 vector unsigned long vec_div (vector unsigned long, vector unsigned long);
18215 vector double vec_floor (vector double);
18216 vector __int128 vec_ld (int, const vector __int128 *);
18217 vector unsigned __int128 vec_ld (int, const vector unsigned __int128 *);
18218 vector __int128 vec_ld (int, const __int128 *);
18219 vector unsigned __int128 vec_ld (int, const unsigned __int128 *);
18220 vector double vec_ld (int, const vector double *);
18221 vector double vec_ld (int, const double *);
18222 vector double vec_ldl (int, const vector double *);
18223 vector double vec_ldl (int, const double *);
18224 vector unsigned char vec_lvsl (int, const double *);
18225 vector unsigned char vec_lvsr (int, const double *);
18226 vector double vec_madd (vector double, vector double, vector double);
18227 vector double vec_max (vector double, vector double);
18228 vector signed long vec_mergeh (vector signed long, vector signed long);
18229 vector signed long vec_mergeh (vector signed long, vector bool long);
18230 vector signed long vec_mergeh (vector bool long, vector signed long);
18231 vector unsigned long vec_mergeh (vector unsigned long, vector unsigned long);
18232 vector unsigned long vec_mergeh (vector unsigned long, vector bool long);
18233 vector unsigned long vec_mergeh (vector bool long, vector unsigned long);
18234 vector signed long vec_mergel (vector signed long, vector signed long);
18235 vector signed long vec_mergel (vector signed long, vector bool long);
18236 vector signed long vec_mergel (vector bool long, vector signed long);
18237 vector unsigned long vec_mergel (vector unsigned long, vector unsigned long);
18238 vector unsigned long vec_mergel (vector unsigned long, vector bool long);
18239 vector unsigned long vec_mergel (vector bool long, vector unsigned long);
18240 vector double vec_min (vector double, vector double);
18241 vector float vec_msub (vector float, vector float, vector float);
18242 vector double vec_msub (vector double, vector double, vector double);
18243 vector float vec_mul (vector float, vector float);
18244 vector double vec_mul (vector double, vector double);
18245 vector long vec_mul (vector long, vector long);
18246 vector unsigned long vec_mul (vector unsigned long, vector unsigned long);
18247 vector float vec_nearbyint (vector float);
18248 vector double vec_nearbyint (vector double);
18249 vector float vec_nmadd (vector float, vector float, vector float);
18250 vector double vec_nmadd (vector double, vector double, vector double);
18251 vector double vec_nmsub (vector double, vector double, vector double);
18252 vector double vec_nor (vector double, vector double);
18253 vector long vec_nor (vector long, vector long);
18254 vector long vec_nor (vector long, vector bool long);
18255 vector long vec_nor (vector bool long, vector long);
18256 vector unsigned long vec_nor (vector unsigned long, vector unsigned long);
18257 vector unsigned long vec_nor (vector unsigned long, vector bool long);
18258 vector unsigned long vec_nor (vector bool long, vector unsigned long);
18259 vector double vec_or (vector double, vector double);
18260 vector double vec_or (vector double, vector bool long);
18261 vector double vec_or (vector bool long, vector double);
18262 vector long vec_or (vector long, vector long);
18263 vector long vec_or (vector long, vector bool long);
18264 vector long vec_or (vector bool long, vector long);
18265 vector unsigned long vec_or (vector unsigned long, vector unsigned long);
18266 vector unsigned long vec_or (vector unsigned long, vector bool long);
18267 vector unsigned long vec_or (vector bool long, vector unsigned long);
18268 vector double vec_perm (vector double, vector double, vector unsigned char);
18269 vector long vec_perm (vector long, vector long, vector unsigned char);
18270 vector unsigned long vec_perm (vector unsigned long, vector unsigned long,
18271 vector unsigned char);
18272 vector bool char vec_permxor (vector bool char, vector bool char,
18274 vector unsigned char vec_permxor (vector signed char, vector signed char,
18275 vector signed char);
18276 vector unsigned char vec_permxor (vector unsigned char, vector unsigned char,
18277 vector unsigned char);
18278 vector double vec_rint (vector double);
18279 vector double vec_recip (vector double, vector double);
18280 vector double vec_rsqrt (vector double);
18281 vector double vec_rsqrte (vector double);
18282 vector double vec_sel (vector double, vector double, vector bool long);
18283 vector double vec_sel (vector double, vector double, vector unsigned long);
18284 vector long vec_sel (vector long, vector long, vector long);
18285 vector long vec_sel (vector long, vector long, vector unsigned long);
18286 vector long vec_sel (vector long, vector long, vector bool long);
18287 vector unsigned long vec_sel (vector unsigned long, vector unsigned long,
18289 vector unsigned long vec_sel (vector unsigned long, vector unsigned long,
18290 vector unsigned long);
18291 vector unsigned long vec_sel (vector unsigned long, vector unsigned long,
18293 vector double vec_splats (double);
18294 vector signed long vec_splats (signed long);
18295 vector unsigned long vec_splats (unsigned long);
18296 vector float vec_sqrt (vector float);
18297 vector double vec_sqrt (vector double);
18298 void vec_st (vector double, int, vector double *);
18299 void vec_st (vector double, int, double *);
18300 vector double vec_sub (vector double, vector double);
18301 vector double vec_trunc (vector double);
18302 vector double vec_xl (int, vector double *);
18303 vector double vec_xl (int, double *);
18304 vector long long vec_xl (int, vector long long *);
18305 vector long long vec_xl (int, long long *);
18306 vector unsigned long long vec_xl (int, vector unsigned long long *);
18307 vector unsigned long long vec_xl (int, unsigned long long *);
18308 vector float vec_xl (int, vector float *);
18309 vector float vec_xl (int, float *);
18310 vector int vec_xl (int, vector int *);
18311 vector int vec_xl (int, int *);
18312 vector unsigned int vec_xl (int, vector unsigned int *);
18313 vector unsigned int vec_xl (int, unsigned int *);
18314 vector double vec_xor (vector double, vector double);
18315 vector double vec_xor (vector double, vector bool long);
18316 vector double vec_xor (vector bool long, vector double);
18317 vector long vec_xor (vector long, vector long);
18318 vector long vec_xor (vector long, vector bool long);
18319 vector long vec_xor (vector bool long, vector long);
18320 vector unsigned long vec_xor (vector unsigned long, vector unsigned long);
18321 vector unsigned long vec_xor (vector unsigned long, vector bool long);
18322 vector unsigned long vec_xor (vector bool long, vector unsigned long);
18323 void vec_xst (vector double, int, vector double *);
18324 void vec_xst (vector double, int, double *);
18325 void vec_xst (vector long long, int, vector long long *);
18326 void vec_xst (vector long long, int, long long *);
18327 void vec_xst (vector unsigned long long, int, vector unsigned long long *);
18328 void vec_xst (vector unsigned long long, int, unsigned long long *);
18329 void vec_xst (vector float, int, vector float *);
18330 void vec_xst (vector float, int, float *);
18331 void vec_xst (vector int, int, vector int *);
18332 void vec_xst (vector int, int, int *);
18333 void vec_xst (vector unsigned int, int, vector unsigned int *);
18334 void vec_xst (vector unsigned int, int, unsigned int *);
18335 int vec_all_eq (vector double, vector double);
18336 int vec_all_ge (vector double, vector double);
18337 int vec_all_gt (vector double, vector double);
18338 int vec_all_le (vector double, vector double);
18339 int vec_all_lt (vector double, vector double);
18340 int vec_all_nan (vector double);
18341 int vec_all_ne (vector double, vector double);
18342 int vec_all_nge (vector double, vector double);
18343 int vec_all_ngt (vector double, vector double);
18344 int vec_all_nle (vector double, vector double);
18345 int vec_all_nlt (vector double, vector double);
18346 int vec_all_numeric (vector double);
18347 int vec_any_eq (vector double, vector double);
18348 int vec_any_ge (vector double, vector double);
18349 int vec_any_gt (vector double, vector double);
18350 int vec_any_le (vector double, vector double);
18351 int vec_any_lt (vector double, vector double);
18352 int vec_any_nan (vector double);
18353 int vec_any_ne (vector double, vector double);
18354 int vec_any_nge (vector double, vector double);
18355 int vec_any_ngt (vector double, vector double);
18356 int vec_any_nle (vector double, vector double);
18357 int vec_any_nlt (vector double, vector double);
18358 int vec_any_numeric (vector double);
18360 vector double vec_vsx_ld (int, const vector double *);
18361 vector double vec_vsx_ld (int, const double *);
18362 vector float vec_vsx_ld (int, const vector float *);
18363 vector float vec_vsx_ld (int, const float *);
18364 vector bool int vec_vsx_ld (int, const vector bool int *);
18365 vector signed int vec_vsx_ld (int, const vector signed int *);
18366 vector signed int vec_vsx_ld (int, const int *);
18367 vector signed int vec_vsx_ld (int, const long *);
18368 vector unsigned int vec_vsx_ld (int, const vector unsigned int *);
18369 vector unsigned int vec_vsx_ld (int, const unsigned int *);
18370 vector unsigned int vec_vsx_ld (int, const unsigned long *);
18371 vector bool short vec_vsx_ld (int, const vector bool short *);
18372 vector pixel vec_vsx_ld (int, const vector pixel *);
18373 vector signed short vec_vsx_ld (int, const vector signed short *);
18374 vector signed short vec_vsx_ld (int, const short *);
18375 vector unsigned short vec_vsx_ld (int, const vector unsigned short *);
18376 vector unsigned short vec_vsx_ld (int, const unsigned short *);
18377 vector bool char vec_vsx_ld (int, const vector bool char *);
18378 vector signed char vec_vsx_ld (int, const vector signed char *);
18379 vector signed char vec_vsx_ld (int, const signed char *);
18380 vector unsigned char vec_vsx_ld (int, const vector unsigned char *);
18381 vector unsigned char vec_vsx_ld (int, const unsigned char *);
18383 void vec_vsx_st (vector double, int, vector double *);
18384 void vec_vsx_st (vector double, int, double *);
18385 void vec_vsx_st (vector float, int, vector float *);
18386 void vec_vsx_st (vector float, int, float *);
18387 void vec_vsx_st (vector signed int, int, vector signed int *);
18388 void vec_vsx_st (vector signed int, int, int *);
18389 void vec_vsx_st (vector unsigned int, int, vector unsigned int *);
18390 void vec_vsx_st (vector unsigned int, int, unsigned int *);
18391 void vec_vsx_st (vector bool int, int, vector bool int *);
18392 void vec_vsx_st (vector bool int, int, unsigned int *);
18393 void vec_vsx_st (vector bool int, int, int *);
18394 void vec_vsx_st (vector signed short, int, vector signed short *);
18395 void vec_vsx_st (vector signed short, int, short *);
18396 void vec_vsx_st (vector unsigned short, int, vector unsigned short *);
18397 void vec_vsx_st (vector unsigned short, int, unsigned short *);
18398 void vec_vsx_st (vector bool short, int, vector bool short *);
18399 void vec_vsx_st (vector bool short, int, unsigned short *);
18400 void vec_vsx_st (vector pixel, int, vector pixel *);
18401 void vec_vsx_st (vector pixel, int, unsigned short *);
18402 void vec_vsx_st (vector pixel, int, short *);
18403 void vec_vsx_st (vector bool short, int, short *);
18404 void vec_vsx_st (vector signed char, int, vector signed char *);
18405 void vec_vsx_st (vector signed char, int, signed char *);
18406 void vec_vsx_st (vector unsigned char, int, vector unsigned char *);
18407 void vec_vsx_st (vector unsigned char, int, unsigned char *);
18408 void vec_vsx_st (vector bool char, int, vector bool char *);
18409 void vec_vsx_st (vector bool char, int, unsigned char *);
18410 void vec_vsx_st (vector bool char, int, signed char *);
18412 vector double vec_xxpermdi (vector double, vector double, const int);
18413 vector float vec_xxpermdi (vector float, vector float, const int);
18414 vector long long vec_xxpermdi (vector long long, vector long long, const int);
18415 vector unsigned long long vec_xxpermdi (vector unsigned long long,
18416 vector unsigned long long, const int);
18417 vector int vec_xxpermdi (vector int, vector int, const int);
18418 vector unsigned int vec_xxpermdi (vector unsigned int,
18419 vector unsigned int, const int);
18420 vector short vec_xxpermdi (vector short, vector short, const int);
18421 vector unsigned short vec_xxpermdi (vector unsigned short,
18422 vector unsigned short, const int);
18423 vector signed char vec_xxpermdi (vector signed char, vector signed char,
18425 vector unsigned char vec_xxpermdi (vector unsigned char,
18426 vector unsigned char, const int);
18428 vector double vec_xxsldi (vector double, vector double, int);
18429 vector float vec_xxsldi (vector float, vector float, int);
18430 vector long long vec_xxsldi (vector long long, vector long long, int);
18431 vector unsigned long long vec_xxsldi (vector unsigned long long,
18432 vector unsigned long long, int);
18433 vector int vec_xxsldi (vector int, vector int, int);
18434 vector unsigned int vec_xxsldi (vector unsigned int, vector unsigned int, int);
18435 vector short vec_xxsldi (vector short, vector short, int);
18436 vector unsigned short vec_xxsldi (vector unsigned short,
18437 vector unsigned short, int);
18438 vector signed char vec_xxsldi (vector signed char, vector signed char, int);
18439 vector unsigned char vec_xxsldi (vector unsigned char,
18440 vector unsigned char, int);
18443 Note that the @samp{vec_ld} and @samp{vec_st} built-in functions always
18444 generate the AltiVec @samp{LVX} and @samp{STVX} instructions even
18445 if the VSX instruction set is available. The @samp{vec_vsx_ld} and
18446 @samp{vec_vsx_st} built-in functions always generate the VSX @samp{LXVD2X},
18447 @samp{LXVW4X}, @samp{STXVD2X}, and @samp{STXVW4X} instructions.
18449 If the ISA 2.07 additions to the vector/scalar (power8-vector)
18450 instruction set are available, the following additional functions are
18451 available for both 32-bit and 64-bit targets. For 64-bit targets, you
18452 can use @var{vector long} instead of @var{vector long long},
18453 @var{vector bool long} instead of @var{vector bool long long}, and
18454 @var{vector unsigned long} instead of @var{vector unsigned long long}.
18457 vector long long vec_abs (vector long long);
18459 vector long long vec_add (vector long long, vector long long);
18460 vector unsigned long long vec_add (vector unsigned long long,
18461 vector unsigned long long);
18463 int vec_all_eq (vector long long, vector long long);
18464 int vec_all_eq (vector unsigned long long, vector unsigned long long);
18465 int vec_all_ge (vector long long, vector long long);
18466 int vec_all_ge (vector unsigned long long, vector unsigned long long);
18467 int vec_all_gt (vector long long, vector long long);
18468 int vec_all_gt (vector unsigned long long, vector unsigned long long);
18469 int vec_all_le (vector long long, vector long long);
18470 int vec_all_le (vector unsigned long long, vector unsigned long long);
18471 int vec_all_lt (vector long long, vector long long);
18472 int vec_all_lt (vector unsigned long long, vector unsigned long long);
18473 int vec_all_ne (vector long long, vector long long);
18474 int vec_all_ne (vector unsigned long long, vector unsigned long long);
18476 int vec_any_eq (vector long long, vector long long);
18477 int vec_any_eq (vector unsigned long long, vector unsigned long long);
18478 int vec_any_ge (vector long long, vector long long);
18479 int vec_any_ge (vector unsigned long long, vector unsigned long long);
18480 int vec_any_gt (vector long long, vector long long);
18481 int vec_any_gt (vector unsigned long long, vector unsigned long long);
18482 int vec_any_le (vector long long, vector long long);
18483 int vec_any_le (vector unsigned long long, vector unsigned long long);
18484 int vec_any_lt (vector long long, vector long long);
18485 int vec_any_lt (vector unsigned long long, vector unsigned long long);
18486 int vec_any_ne (vector long long, vector long long);
18487 int vec_any_ne (vector unsigned long long, vector unsigned long long);
18489 vector bool long long vec_cmpeq (vector bool long long, vector bool long long);
18491 vector long long vec_eqv (vector long long, vector long long);
18492 vector long long vec_eqv (vector bool long long, vector long long);
18493 vector long long vec_eqv (vector long long, vector bool long long);
18494 vector unsigned long long vec_eqv (vector unsigned long long,
18495 vector unsigned long long);
18496 vector unsigned long long vec_eqv (vector bool long long,
18497 vector unsigned long long);
18498 vector unsigned long long vec_eqv (vector unsigned long long,
18499 vector bool long long);
18500 vector int vec_eqv (vector int, vector int);
18501 vector int vec_eqv (vector bool int, vector int);
18502 vector int vec_eqv (vector int, vector bool int);
18503 vector unsigned int vec_eqv (vector unsigned int, vector unsigned int);
18504 vector unsigned int vec_eqv (vector bool unsigned int,
18505 vector unsigned int);
18506 vector unsigned int vec_eqv (vector unsigned int,
18507 vector bool unsigned int);
18508 vector short vec_eqv (vector short, vector short);
18509 vector short vec_eqv (vector bool short, vector short);
18510 vector short vec_eqv (vector short, vector bool short);
18511 vector unsigned short vec_eqv (vector unsigned short, vector unsigned short);
18512 vector unsigned short vec_eqv (vector bool unsigned short,
18513 vector unsigned short);
18514 vector unsigned short vec_eqv (vector unsigned short,
18515 vector bool unsigned short);
18516 vector signed char vec_eqv (vector signed char, vector signed char);
18517 vector signed char vec_eqv (vector bool signed char, vector signed char);
18518 vector signed char vec_eqv (vector signed char, vector bool signed char);
18519 vector unsigned char vec_eqv (vector unsigned char, vector unsigned char);
18520 vector unsigned char vec_eqv (vector bool unsigned char, vector unsigned char);
18521 vector unsigned char vec_eqv (vector unsigned char, vector bool unsigned char);
18523 vector long long vec_max (vector long long, vector long long);
18524 vector unsigned long long vec_max (vector unsigned long long,
18525 vector unsigned long long);
18527 vector signed int vec_mergee (vector signed int, vector signed int);
18528 vector unsigned int vec_mergee (vector unsigned int, vector unsigned int);
18529 vector bool int vec_mergee (vector bool int, vector bool int);
18531 vector signed int vec_mergeo (vector signed int, vector signed int);
18532 vector unsigned int vec_mergeo (vector unsigned int, vector unsigned int);
18533 vector bool int vec_mergeo (vector bool int, vector bool int);
18535 vector long long vec_min (vector long long, vector long long);
18536 vector unsigned long long vec_min (vector unsigned long long,
18537 vector unsigned long long);
18539 vector signed long long vec_nabs (vector signed long long);
18541 vector long long vec_nand (vector long long, vector long long);
18542 vector long long vec_nand (vector bool long long, vector long long);
18543 vector long long vec_nand (vector long long, vector bool long long);
18544 vector unsigned long long vec_nand (vector unsigned long long,
18545 vector unsigned long long);
18546 vector unsigned long long vec_nand (vector bool long long,
18547 vector unsigned long long);
18548 vector unsigned long long vec_nand (vector unsigned long long,
18549 vector bool long long);
18550 vector int vec_nand (vector int, vector int);
18551 vector int vec_nand (vector bool int, vector int);
18552 vector int vec_nand (vector int, vector bool int);
18553 vector unsigned int vec_nand (vector unsigned int, vector unsigned int);
18554 vector unsigned int vec_nand (vector bool unsigned int,
18555 vector unsigned int);
18556 vector unsigned int vec_nand (vector unsigned int,
18557 vector bool unsigned int);
18558 vector short vec_nand (vector short, vector short);
18559 vector short vec_nand (vector bool short, vector short);
18560 vector short vec_nand (vector short, vector bool short);
18561 vector unsigned short vec_nand (vector unsigned short, vector unsigned short);
18562 vector unsigned short vec_nand (vector bool unsigned short,
18563 vector unsigned short);
18564 vector unsigned short vec_nand (vector unsigned short,
18565 vector bool unsigned short);
18566 vector signed char vec_nand (vector signed char, vector signed char);
18567 vector signed char vec_nand (vector bool signed char, vector signed char);
18568 vector signed char vec_nand (vector signed char, vector bool signed char);
18569 vector unsigned char vec_nand (vector unsigned char, vector unsigned char);
18570 vector unsigned char vec_nand (vector bool unsigned char, vector unsigned char);
18571 vector unsigned char vec_nand (vector unsigned char, vector bool unsigned char);
18573 vector long long vec_orc (vector long long, vector long long);
18574 vector long long vec_orc (vector bool long long, vector long long);
18575 vector long long vec_orc (vector long long, vector bool long long);
18576 vector unsigned long long vec_orc (vector unsigned long long,
18577 vector unsigned long long);
18578 vector unsigned long long vec_orc (vector bool long long,
18579 vector unsigned long long);
18580 vector unsigned long long vec_orc (vector unsigned long long,
18581 vector bool long long);
18582 vector int vec_orc (vector int, vector int);
18583 vector int vec_orc (vector bool int, vector int);
18584 vector int vec_orc (vector int, vector bool int);
18585 vector unsigned int vec_orc (vector unsigned int, vector unsigned int);
18586 vector unsigned int vec_orc (vector bool unsigned int,
18587 vector unsigned int);
18588 vector unsigned int vec_orc (vector unsigned int,
18589 vector bool unsigned int);
18590 vector short vec_orc (vector short, vector short);
18591 vector short vec_orc (vector bool short, vector short);
18592 vector short vec_orc (vector short, vector bool short);
18593 vector unsigned short vec_orc (vector unsigned short, vector unsigned short);
18594 vector unsigned short vec_orc (vector bool unsigned short,
18595 vector unsigned short);
18596 vector unsigned short vec_orc (vector unsigned short,
18597 vector bool unsigned short);
18598 vector signed char vec_orc (vector signed char, vector signed char);
18599 vector signed char vec_orc (vector bool signed char, vector signed char);
18600 vector signed char vec_orc (vector signed char, vector bool signed char);
18601 vector unsigned char vec_orc (vector unsigned char, vector unsigned char);
18602 vector unsigned char vec_orc (vector bool unsigned char, vector unsigned char);
18603 vector unsigned char vec_orc (vector unsigned char, vector bool unsigned char);
18605 vector int vec_pack (vector long long, vector long long);
18606 vector unsigned int vec_pack (vector unsigned long long,
18607 vector unsigned long long);
18608 vector bool int vec_pack (vector bool long long, vector bool long long);
18609 vector float vec_pack (vector double, vector double);
18611 vector int vec_packs (vector long long, vector long long);
18612 vector unsigned int vec_packs (vector unsigned long long,
18613 vector unsigned long long);
18615 vector unsigned char vec_packsu (vector signed short, vector signed short )
18616 vector unsigned char vec_packsu (vector unsigned short, vector unsigned short )
18617 vector unsigned short int vec_packsu (vector signed int, vector signed int);
18618 vector unsigned short int vec_packsu (vector unsigned int,
18619 vector unsigned int);
18620 vector unsigned int vec_packsu (vector long long, vector long long);
18621 vector unsigned int vec_packsu (vector unsigned long long,
18622 vector unsigned long long);
18623 vector unsigned int vec_packsu (vector signed long long,
18624 vector signed long long);
18626 vector unsigned char vec_popcnt (vector signed char);
18627 vector unsigned char vec_popcnt (vector unsigned char);
18628 vector unsigned short vec_popcnt (vector signed short);
18629 vector unsigned short vec_popcnt (vector unsigned short);
18630 vector unsigned int vec_popcnt (vector signed int);
18631 vector unsigned int vec_popcnt (vector unsigned int);
18632 vector unsigned long long vec_popcnt (vector signed long long);
18633 vector unsigned long long vec_popcnt (vector unsigned long long);
18635 vector long long vec_rl (vector long long,
18636 vector unsigned long long);
18637 vector long long vec_rl (vector unsigned long long,
18638 vector unsigned long long);
18640 vector long long vec_sl (vector long long, vector unsigned long long);
18641 vector long long vec_sl (vector unsigned long long,
18642 vector unsigned long long);
18644 vector long long vec_sr (vector long long, vector unsigned long long);
18645 vector unsigned long long char vec_sr (vector unsigned long long,
18646 vector unsigned long long);
18648 vector long long vec_sra (vector long long, vector unsigned long long);
18649 vector unsigned long long vec_sra (vector unsigned long long,
18650 vector unsigned long long);
18652 vector long long vec_sub (vector long long, vector long long);
18653 vector unsigned long long vec_sub (vector unsigned long long,
18654 vector unsigned long long);
18656 vector long long vec_unpackh (vector int);
18657 vector unsigned long long vec_unpackh (vector unsigned int);
18659 vector long long vec_unpackl (vector int);
18660 vector unsigned long long vec_unpackl (vector unsigned int);
18662 vector long long vec_vaddudm (vector long long, vector long long);
18663 vector long long vec_vaddudm (vector bool long long, vector long long);
18664 vector long long vec_vaddudm (vector long long, vector bool long long);
18665 vector unsigned long long vec_vaddudm (vector unsigned long long,
18666 vector unsigned long long);
18667 vector unsigned long long vec_vaddudm (vector bool unsigned long long,
18668 vector unsigned long long);
18669 vector unsigned long long vec_vaddudm (vector unsigned long long,
18670 vector bool unsigned long long);
18672 vector long long vec_vbpermq (vector signed char, vector signed char);
18673 vector long long vec_vbpermq (vector unsigned char, vector unsigned char);
18675 vector unsigned char vec_bperm (vector unsigned char, vector unsigned char);
18676 vector unsigned char vec_bperm (vector unsigned long long,
18677 vector unsigned char);
18678 vector unsigned long long vec_bperm (vector unsigned __int128,
18679 vector unsigned char);
18681 vector long long vec_cntlz (vector long long);
18682 vector unsigned long long vec_cntlz (vector unsigned long long);
18683 vector int vec_cntlz (vector int);
18684 vector unsigned int vec_cntlz (vector int);
18685 vector short vec_cntlz (vector short);
18686 vector unsigned short vec_cntlz (vector unsigned short);
18687 vector signed char vec_cntlz (vector signed char);
18688 vector unsigned char vec_cntlz (vector unsigned char);
18690 vector long long vec_vclz (vector long long);
18691 vector unsigned long long vec_vclz (vector unsigned long long);
18692 vector int vec_vclz (vector int);
18693 vector unsigned int vec_vclz (vector int);
18694 vector short vec_vclz (vector short);
18695 vector unsigned short vec_vclz (vector unsigned short);
18696 vector signed char vec_vclz (vector signed char);
18697 vector unsigned char vec_vclz (vector unsigned char);
18699 vector signed char vec_vclzb (vector signed char);
18700 vector unsigned char vec_vclzb (vector unsigned char);
18702 vector long long vec_vclzd (vector long long);
18703 vector unsigned long long vec_vclzd (vector unsigned long long);
18705 vector short vec_vclzh (vector short);
18706 vector unsigned short vec_vclzh (vector unsigned short);
18708 vector int vec_vclzw (vector int);
18709 vector unsigned int vec_vclzw (vector int);
18711 vector signed char vec_vgbbd (vector signed char);
18712 vector unsigned char vec_vgbbd (vector unsigned char);
18714 vector long long vec_vmaxsd (vector long long, vector long long);
18716 vector unsigned long long vec_vmaxud (vector unsigned long long,
18717 unsigned vector long long);
18719 vector long long vec_vminsd (vector long long, vector long long);
18721 vector unsigned long long vec_vminud (vector long long,
18724 vector int vec_vpksdss (vector long long, vector long long);
18725 vector unsigned int vec_vpksdss (vector long long, vector long long);
18727 vector unsigned int vec_vpkudus (vector unsigned long long,
18728 vector unsigned long long);
18730 vector int vec_vpkudum (vector long long, vector long long);
18731 vector unsigned int vec_vpkudum (vector unsigned long long,
18732 vector unsigned long long);
18733 vector bool int vec_vpkudum (vector bool long long, vector bool long long);
18735 vector long long vec_vpopcnt (vector long long);
18736 vector unsigned long long vec_vpopcnt (vector unsigned long long);
18737 vector int vec_vpopcnt (vector int);
18738 vector unsigned int vec_vpopcnt (vector int);
18739 vector short vec_vpopcnt (vector short);
18740 vector unsigned short vec_vpopcnt (vector unsigned short);
18741 vector signed char vec_vpopcnt (vector signed char);
18742 vector unsigned char vec_vpopcnt (vector unsigned char);
18744 vector signed char vec_vpopcntb (vector signed char);
18745 vector unsigned char vec_vpopcntb (vector unsigned char);
18747 vector long long vec_vpopcntd (vector long long);
18748 vector unsigned long long vec_vpopcntd (vector unsigned long long);
18750 vector short vec_vpopcnth (vector short);
18751 vector unsigned short vec_vpopcnth (vector unsigned short);
18753 vector int vec_vpopcntw (vector int);
18754 vector unsigned int vec_vpopcntw (vector int);
18756 vector long long vec_vrld (vector long long, vector unsigned long long);
18757 vector unsigned long long vec_vrld (vector unsigned long long,
18758 vector unsigned long long);
18760 vector long long vec_vsld (vector long long, vector unsigned long long);
18761 vector long long vec_vsld (vector unsigned long long,
18762 vector unsigned long long);
18764 vector long long vec_vsrad (vector long long, vector unsigned long long);
18765 vector unsigned long long vec_vsrad (vector unsigned long long,
18766 vector unsigned long long);
18768 vector long long vec_vsrd (vector long long, vector unsigned long long);
18769 vector unsigned long long char vec_vsrd (vector unsigned long long,
18770 vector unsigned long long);
18772 vector long long vec_vsubudm (vector long long, vector long long);
18773 vector long long vec_vsubudm (vector bool long long, vector long long);
18774 vector long long vec_vsubudm (vector long long, vector bool long long);
18775 vector unsigned long long vec_vsubudm (vector unsigned long long,
18776 vector unsigned long long);
18777 vector unsigned long long vec_vsubudm (vector bool long long,
18778 vector unsigned long long);
18779 vector unsigned long long vec_vsubudm (vector unsigned long long,
18780 vector bool long long);
18782 vector long long vec_vupkhsw (vector int);
18783 vector unsigned long long vec_vupkhsw (vector unsigned int);
18785 vector long long vec_vupklsw (vector int);
18786 vector unsigned long long vec_vupklsw (vector int);
18789 If the ISA 2.07 additions to the vector/scalar (power8-vector)
18790 instruction set are available, the following additional functions are
18791 available for 64-bit targets. New vector types
18792 (@var{vector __int128_t} and @var{vector __uint128_t}) are available
18793 to hold the @var{__int128_t} and @var{__uint128_t} types to use these
18796 The normal vector extract, and set operations work on
18797 @var{vector __int128_t} and @var{vector __uint128_t} types,
18798 but the index value must be 0.
18801 vector __int128_t vec_vaddcuq (vector __int128_t, vector __int128_t);
18802 vector __uint128_t vec_vaddcuq (vector __uint128_t, vector __uint128_t);
18804 vector __int128_t vec_vadduqm (vector __int128_t, vector __int128_t);
18805 vector __uint128_t vec_vadduqm (vector __uint128_t, vector __uint128_t);
18807 vector __int128_t vec_vaddecuq (vector __int128_t, vector __int128_t,
18808 vector __int128_t);
18809 vector __uint128_t vec_vaddecuq (vector __uint128_t, vector __uint128_t,
18810 vector __uint128_t);
18812 vector __int128_t vec_vaddeuqm (vector __int128_t, vector __int128_t,
18813 vector __int128_t);
18814 vector __uint128_t vec_vaddeuqm (vector __uint128_t, vector __uint128_t,
18815 vector __uint128_t);
18817 vector __int128_t vec_vsubecuq (vector __int128_t, vector __int128_t,
18818 vector __int128_t);
18819 vector __uint128_t vec_vsubecuq (vector __uint128_t, vector __uint128_t,
18820 vector __uint128_t);
18822 vector __int128_t vec_vsubeuqm (vector __int128_t, vector __int128_t,
18823 vector __int128_t);
18824 vector __uint128_t vec_vsubeuqm (vector __uint128_t, vector __uint128_t,
18825 vector __uint128_t);
18827 vector __int128_t vec_vsubcuq (vector __int128_t, vector __int128_t);
18828 vector __uint128_t vec_vsubcuq (vector __uint128_t, vector __uint128_t);
18830 __int128_t vec_vsubuqm (__int128_t, __int128_t);
18831 __uint128_t vec_vsubuqm (__uint128_t, __uint128_t);
18833 vector __int128_t __builtin_bcdadd (vector __int128_t, vector __int128_t);
18834 int __builtin_bcdadd_lt (vector __int128_t, vector __int128_t);
18835 int __builtin_bcdadd_eq (vector __int128_t, vector __int128_t);
18836 int __builtin_bcdadd_gt (vector __int128_t, vector __int128_t);
18837 int __builtin_bcdadd_ov (vector __int128_t, vector __int128_t);
18838 vector __int128_t bcdsub (vector __int128_t, vector __int128_t);
18839 int __builtin_bcdsub_lt (vector __int128_t, vector __int128_t);
18840 int __builtin_bcdsub_eq (vector __int128_t, vector __int128_t);
18841 int __builtin_bcdsub_gt (vector __int128_t, vector __int128_t);
18842 int __builtin_bcdsub_ov (vector __int128_t, vector __int128_t);
18845 The following additional built-in functions are also available for the
18846 PowerPC family of processors, starting with ISA 3.0
18847 (@option{-mcpu=power9}) or later:
18849 unsigned int scalar_extract_exp (double source);
18850 unsigned long long int scalar_extract_exp (__ieee128 source);
18852 unsigned long long int scalar_extract_sig (double source);
18853 unsigned __int128 scalar_extract_sig (__ieee128 source);
18856 scalar_insert_exp (unsigned long long int significand, unsigned long long int exponent);
18858 scalar_insert_exp (double significand, unsigned long long int exponent);
18861 scalar_insert_exp (unsigned __int128 significand, unsigned long long int exponent);
18863 scalar_insert_exp (ieee_128 significand, unsigned long long int exponent);
18865 int scalar_cmp_exp_gt (double arg1, double arg2);
18866 int scalar_cmp_exp_lt (double arg1, double arg2);
18867 int scalar_cmp_exp_eq (double arg1, double arg2);
18868 int scalar_cmp_exp_unordered (double arg1, double arg2);
18870 bool scalar_test_data_class (float source, const int condition);
18871 bool scalar_test_data_class (double source, const int condition);
18872 bool scalar_test_data_class (__ieee128 source, const int condition);
18874 bool scalar_test_neg (float source);
18875 bool scalar_test_neg (double source);
18876 bool scalar_test_neg (__ieee128 source);
18879 The @code{scalar_extract_exp} and @code{scalar_extract_sig}
18880 functions require a 64-bit environment supporting ISA 3.0 or later.
18881 The @code{scalar_extract_exp} and @code{scalar_extract_sig} built-in
18882 functions return the significand and the biased exponent value
18883 respectively of their @code{source} arguments.
18884 When supplied with a 64-bit @code{source} argument, the
18885 result returned by @code{scalar_extract_sig} has
18886 the @code{0x0010000000000000} bit set if the
18887 function's @code{source} argument is in normalized form.
18888 Otherwise, this bit is set to 0.
18889 When supplied with a 128-bit @code{source} argument, the
18890 @code{0x00010000000000000000000000000000} bit of the result is
18892 Note that the sign of the significand is not represented in the result
18893 returned from the @code{scalar_extract_sig} function. Use the
18894 @code{scalar_test_neg} function to test the sign of its @code{double}
18897 The @code{scalar_insert_exp}
18898 functions require a 64-bit environment supporting ISA 3.0 or later.
18899 When supplied with a 64-bit first argument, the
18900 @code{scalar_insert_exp} built-in function returns a double-precision
18901 floating point value that is constructed by assembling the values of its
18902 @code{significand} and @code{exponent} arguments. The sign of the
18903 result is copied from the most significant bit of the
18904 @code{significand} argument. The significand and exponent components
18905 of the result are composed of the least significant 11 bits of the
18906 @code{exponent} argument and the least significant 52 bits of the
18907 @code{significand} argument respectively.
18909 When supplied with a 128-bit first argument, the
18910 @code{scalar_insert_exp} built-in function returns a quad-precision
18911 ieee floating point value. The sign bit of the result is copied from
18912 the most significant bit of the @code{significand} argument.
18913 The significand and exponent components of the result are composed of
18914 the least significant 15 bits of the @code{exponent} argument and the
18915 least significant 112 bits of the @code{significand} argument respectively.
18917 The @code{scalar_cmp_exp_gt}, @code{scalar_cmp_exp_lt},
18918 @code{scalar_cmp_exp_eq}, and @code{scalar_cmp_exp_unordered} built-in
18919 functions return a non-zero value if @code{arg1} is greater than, less
18920 than, equal to, or not comparable to @code{arg2} respectively. The
18921 arguments are not comparable if one or the other equals NaN (not a
18924 The @code{scalar_test_data_class} built-in function returns 1
18925 if any of the condition tests enabled by the value of the
18926 @code{condition} variable are true, and 0 otherwise. The
18927 @code{condition} argument must be a compile-time constant integer with
18928 value not exceeding 127. The
18929 @code{condition} argument is encoded as a bitmask with each bit
18930 enabling the testing of a different condition, as characterized by the
18934 0x20 Test for +Infinity
18935 0x10 Test for -Infinity
18936 0x08 Test for +Zero
18937 0x04 Test for -Zero
18938 0x02 Test for +Denormal
18939 0x01 Test for -Denormal
18942 The @code{scalar_test_neg} built-in function returns 1 if its
18943 @code{source} argument holds a negative value, 0 otherwise.
18945 The following built-in functions are also available for the PowerPC family
18946 of processors, starting with ISA 3.0 or later
18947 (@option{-mcpu=power9}). These string functions are described
18948 separately in order to group the descriptions closer to the function
18951 int vec_all_nez (vector signed char, vector signed char);
18952 int vec_all_nez (vector unsigned char, vector unsigned char);
18953 int vec_all_nez (vector signed short, vector signed short);
18954 int vec_all_nez (vector unsigned short, vector unsigned short);
18955 int vec_all_nez (vector signed int, vector signed int);
18956 int vec_all_nez (vector unsigned int, vector unsigned int);
18958 int vec_any_eqz (vector signed char, vector signed char);
18959 int vec_any_eqz (vector unsigned char, vector unsigned char);
18960 int vec_any_eqz (vector signed short, vector signed short);
18961 int vec_any_eqz (vector unsigned short, vector unsigned short);
18962 int vec_any_eqz (vector signed int, vector signed int);
18963 int vec_any_eqz (vector unsigned int, vector unsigned int);
18965 vector bool char vec_cmpnez (vector signed char arg1, vector signed char arg2);
18966 vector bool char vec_cmpnez (vector unsigned char arg1, vector unsigned char arg2);
18967 vector bool short vec_cmpnez (vector signed short arg1, vector signed short arg2);
18968 vector bool short vec_cmpnez (vector unsigned short arg1, vector unsigned short arg2);
18969 vector bool int vec_cmpnez (vector signed int arg1, vector signed int arg2);
18970 vector bool int vec_cmpnez (vector unsigned int, vector unsigned int);
18972 vector signed char vec_cnttz (vector signed char);
18973 vector unsigned char vec_cnttz (vector unsigned char);
18974 vector signed short vec_cnttz (vector signed short);
18975 vector unsigned short vec_cnttz (vector unsigned short);
18976 vector signed int vec_cnttz (vector signed int);
18977 vector unsigned int vec_cnttz (vector unsigned int);
18978 vector signed long long vec_cnttz (vector signed long long);
18979 vector unsigned long long vec_cnttz (vector unsigned long long);
18981 signed int vec_cntlz_lsbb (vector signed char);
18982 signed int vec_cntlz_lsbb (vector unsigned char);
18984 signed int vec_cnttz_lsbb (vector signed char);
18985 signed int vec_cnttz_lsbb (vector unsigned char);
18987 unsigned int vec_first_match_index (vector signed char, vector signed char);
18988 unsigned int vec_first_match_index (vector unsigned char,
18989 vector unsigned char);
18990 unsigned int vec_first_match_index (vector signed int, vector signed int);
18991 unsigned int vec_first_match_index (vector unsigned int, vector unsigned int);
18992 unsigned int vec_first_match_index (vector signed short, vector signed short);
18993 unsigned int vec_first_match_index (vector unsigned short,
18994 vector unsigned short);
18995 unsigned int vec_first_match_or_eos_index (vector signed char,
18996 vector signed char);
18997 unsigned int vec_first_match_or_eos_index (vector unsigned char,
18998 vector unsigned char);
18999 unsigned int vec_first_match_or_eos_index (vector signed int,
19000 vector signed int);
19001 unsigned int vec_first_match_or_eos_index (vector unsigned int,
19002 vector unsigned int);
19003 unsigned int vec_first_match_or_eos_index (vector signed short,
19004 vector signed short);
19005 unsigned int vec_first_match_or_eos_index (vector unsigned short,
19006 vector unsigned short);
19007 unsigned int vec_first_mismatch_index (vector signed char,
19008 vector signed char);
19009 unsigned int vec_first_mismatch_index (vector unsigned char,
19010 vector unsigned char);
19011 unsigned int vec_first_mismatch_index (vector signed int,
19012 vector signed int);
19013 unsigned int vec_first_mismatch_index (vector unsigned int,
19014 vector unsigned int);
19015 unsigned int vec_first_mismatch_index (vector signed short,
19016 vector signed short);
19017 unsigned int vec_first_mismatch_index (vector unsigned short,
19018 vector unsigned short);
19019 unsigned int vec_first_mismatch_or_eos_index (vector signed char,
19020 vector signed char);
19021 unsigned int vec_first_mismatch_or_eos_index (vector unsigned char,
19022 vector unsigned char);
19023 unsigned int vec_first_mismatch_or_eos_index (vector signed int,
19024 vector signed int);
19025 unsigned int vec_first_mismatch_or_eos_index (vector unsigned int,
19026 vector unsigned int);
19027 unsigned int vec_first_mismatch_or_eos_index (vector signed short,
19028 vector signed short);
19029 unsigned int vec_first_mismatch_or_eos_index (vector unsigned short,
19030 vector unsigned short);
19032 vector unsigned short vec_pack_to_short_fp32 (vector float, vector float);
19034 vector signed char vec_xl_be (signed long long, signed char *);
19035 vector unsigned char vec_xl_be (signed long long, unsigned char *);
19036 vector signed int vec_xl_be (signed long long, signed int *);
19037 vector unsigned int vec_xl_be (signed long long, unsigned int *);
19038 vector signed __int128 vec_xl_be (signed long long, signed __int128 *);
19039 vector unsigned __int128 vec_xl_be (signed long long, unsigned __int128 *);
19040 vector signed long long vec_xl_be (signed long long, signed long long *);
19041 vector unsigned long long vec_xl_be (signed long long, unsigned long long *);
19042 vector signed short vec_xl_be (signed long long, signed short *);
19043 vector unsigned short vec_xl_be (signed long long, unsigned short *);
19044 vector double vec_xl_be (signed long long, double *);
19045 vector float vec_xl_be (signed long long, float *);
19047 vector signed char vec_xl_len (signed char *addr, size_t len);
19048 vector unsigned char vec_xl_len (unsigned char *addr, size_t len);
19049 vector signed int vec_xl_len (signed int *addr, size_t len);
19050 vector unsigned int vec_xl_len (unsigned int *addr, size_t len);
19051 vector signed __int128 vec_xl_len (signed __int128 *addr, size_t len);
19052 vector unsigned __int128 vec_xl_len (unsigned __int128 *addr, size_t len);
19053 vector signed long long vec_xl_len (signed long long *addr, size_t len);
19054 vector unsigned long long vec_xl_len (unsigned long long *addr, size_t len);
19055 vector signed short vec_xl_len (signed short *addr, size_t len);
19056 vector unsigned short vec_xl_len (unsigned short *addr, size_t len);
19057 vector double vec_xl_len (double *addr, size_t len);
19058 vector float vec_xl_len (float *addr, size_t len);
19060 vector unsigned char vec_xl_len_r (unsigned char *addr, size_t len);
19062 void vec_xst_len (vector signed char data, signed char *addr, size_t len);
19063 void vec_xst_len (vector unsigned char data, unsigned char *addr, size_t len);
19064 void vec_xst_len (vector signed int data, signed int *addr, size_t len);
19065 void vec_xst_len (vector unsigned int data, unsigned int *addr, size_t len);
19066 void vec_xst_len (vector unsigned __int128 data, unsigned __int128 *addr, size_t len);
19067 void vec_xst_len (vector signed long long data, signed long long *addr, size_t len);
19068 void vec_xst_len (vector unsigned long long data, unsigned long long *addr, size_t len);
19069 void vec_xst_len (vector signed short data, signed short *addr, size_t len);
19070 void vec_xst_len (vector unsigned short data, unsigned short *addr, size_t len);
19071 void vec_xst_len (vector signed __int128 data, signed __int128 *addr, size_t len);
19072 void vec_xst_len (vector double data, double *addr, size_t len);
19073 void vec_xst_len (vector float data, float *addr, size_t len);
19075 void vec_xst_len_r (vector unsigned char data, unsigned char *addr, size_t len);
19077 signed char vec_xlx (unsigned int index, vector signed char data);
19078 unsigned char vec_xlx (unsigned int index, vector unsigned char data);
19079 signed short vec_xlx (unsigned int index, vector signed short data);
19080 unsigned short vec_xlx (unsigned int index, vector unsigned short data);
19081 signed int vec_xlx (unsigned int index, vector signed int data);
19082 unsigned int vec_xlx (unsigned int index, vector unsigned int data);
19083 float vec_xlx (unsigned int index, vector float data);
19085 signed char vec_xrx (unsigned int index, vector signed char data);
19086 unsigned char vec_xrx (unsigned int index, vector unsigned char data);
19087 signed short vec_xrx (unsigned int index, vector signed short data);
19088 unsigned short vec_xrx (unsigned int index, vector unsigned short data);
19089 signed int vec_xrx (unsigned int index, vector signed int data);
19090 unsigned int vec_xrx (unsigned int index, vector unsigned int data);
19091 float vec_xrx (unsigned int index, vector float data);
19094 The @code{vec_all_nez}, @code{vec_any_eqz}, and @code{vec_cmpnez}
19095 perform pairwise comparisons between the elements at the same
19096 positions within their two vector arguments.
19097 The @code{vec_all_nez} function returns a
19098 non-zero value if and only if all pairwise comparisons are not
19099 equal and no element of either vector argument contains a zero.
19100 The @code{vec_any_eqz} function returns a
19101 non-zero value if and only if at least one pairwise comparison is equal
19102 or if at least one element of either vector argument contains a zero.
19103 The @code{vec_cmpnez} function returns a vector of the same type as
19104 its two arguments, within which each element consists of all ones to
19105 denote that either the corresponding elements of the incoming arguments are
19106 not equal or that at least one of the corresponding elements contains
19107 zero. Otherwise, the element of the returned vector contains all zeros.
19109 The @code{vec_cntlz_lsbb} function returns the count of the number of
19110 consecutive leading byte elements (starting from position 0 within the
19111 supplied vector argument) for which the least-significant bit
19112 equals zero. The @code{vec_cnttz_lsbb} function returns the count of
19113 the number of consecutive trailing byte elements (starting from
19114 position 15 and counting backwards within the supplied vector
19115 argument) for which the least-significant bit equals zero.
19117 The @code{vec_xl_len} and @code{vec_xst_len} functions require a
19118 64-bit environment supporting ISA 3.0 or later. The @code{vec_xl_len}
19119 function loads a variable length vector from memory. The
19120 @code{vec_xst_len} function stores a variable length vector to memory.
19121 With both the @code{vec_xl_len} and @code{vec_xst_len} functions, the
19122 @code{addr} argument represents the memory address to or from which
19123 data will be transferred, and the
19124 @code{len} argument represents the number of bytes to be
19125 transferred, as computed by the C expression @code{min((len & 0xff), 16)}.
19126 If this expression's value is not a multiple of the vector element's
19127 size, the behavior of this function is undefined.
19128 In the case that the underlying computer is configured to run in
19129 big-endian mode, the data transfer moves bytes 0 to @code{(len - 1)} of
19130 the corresponding vector. In little-endian mode, the data transfer
19131 moves bytes @code{(16 - len)} to @code{15} of the corresponding
19132 vector. For the load function, any bytes of the result vector that
19133 are not loaded from memory are set to zero.
19134 The value of the @code{addr} argument need not be aligned on a
19135 multiple of the vector's element size.
19137 The @code{vec_xlx} and @code{vec_xrx} functions extract the single
19138 element selected by the @code{index} argument from the vector
19139 represented by the @code{data} argument. The @code{index} argument
19140 always specifies a byte offset, regardless of the size of the vector
19141 element. With @code{vec_xlx}, @code{index} is the offset of the first
19142 byte of the element to be extracted. With @code{vec_xrx}, @code{index}
19143 represents the last byte of the element to be extracted, measured
19144 from the right end of the vector. In other words, the last byte of
19145 the element to be extracted is found at position @code{(15 - index)}.
19146 There is no requirement that @code{index} be a multiple of the vector
19147 element size. However, if the size of the vector element added to
19148 @code{index} is greater than 15, the content of the returned value is
19151 If the ISA 3.0 instruction set additions (@option{-mcpu=power9})
19155 vector unsigned long long vec_bperm (vector unsigned long long,
19156 vector unsigned char);
19158 vector bool char vec_cmpne (vector bool char, vector bool char);
19159 vector bool char vec_cmpne (vector signed char, vector signed char);
19160 vector bool char vec_cmpne (vector unsigned char, vector unsigned char);
19161 vector bool int vec_cmpne (vector bool int, vector bool int);
19162 vector bool int vec_cmpne (vector signed int, vector signed int);
19163 vector bool int vec_cmpne (vector unsigned int, vector unsigned int);
19164 vector bool long long vec_cmpne (vector bool long long, vector bool long long);
19165 vector bool long long vec_cmpne (vector signed long long,
19166 vector signed long long);
19167 vector bool long long vec_cmpne (vector unsigned long long,
19168 vector unsigned long long);
19169 vector bool short vec_cmpne (vector bool short, vector bool short);
19170 vector bool short vec_cmpne (vector signed short, vector signed short);
19171 vector bool short vec_cmpne (vector unsigned short, vector unsigned short);
19172 vector bool long long vec_cmpne (vector double, vector double);
19173 vector bool int vec_cmpne (vector float, vector float);
19175 vector float vec_extract_fp32_from_shorth (vector unsigned short);
19176 vector float vec_extract_fp32_from_shortl (vector unsigned short);
19178 vector long long vec_vctz (vector long long);
19179 vector unsigned long long vec_vctz (vector unsigned long long);
19180 vector int vec_vctz (vector int);
19181 vector unsigned int vec_vctz (vector int);
19182 vector short vec_vctz (vector short);
19183 vector unsigned short vec_vctz (vector unsigned short);
19184 vector signed char vec_vctz (vector signed char);
19185 vector unsigned char vec_vctz (vector unsigned char);
19187 vector signed char vec_vctzb (vector signed char);
19188 vector unsigned char vec_vctzb (vector unsigned char);
19190 vector long long vec_vctzd (vector long long);
19191 vector unsigned long long vec_vctzd (vector unsigned long long);
19193 vector short vec_vctzh (vector short);
19194 vector unsigned short vec_vctzh (vector unsigned short);
19196 vector int vec_vctzw (vector int);
19197 vector unsigned int vec_vctzw (vector int);
19199 vector unsigned long long vec_extract4b (vector unsigned char, const int);
19201 vector unsigned char vec_insert4b (vector signed int, vector unsigned char,
19203 vector unsigned char vec_insert4b (vector unsigned int, vector unsigned char,
19206 vector unsigned int vec_parity_lsbb (vector signed int);
19207 vector unsigned int vec_parity_lsbb (vector unsigned int);
19208 vector unsigned __int128 vec_parity_lsbb (vector signed __int128);
19209 vector unsigned __int128 vec_parity_lsbb (vector unsigned __int128);
19210 vector unsigned long long vec_parity_lsbb (vector signed long long);
19211 vector unsigned long long vec_parity_lsbb (vector unsigned long long);
19213 vector int vec_vprtyb (vector int);
19214 vector unsigned int vec_vprtyb (vector unsigned int);
19215 vector long long vec_vprtyb (vector long long);
19216 vector unsigned long long vec_vprtyb (vector unsigned long long);
19218 vector int vec_vprtybw (vector int);
19219 vector unsigned int vec_vprtybw (vector unsigned int);
19221 vector long long vec_vprtybd (vector long long);
19222 vector unsigned long long vec_vprtybd (vector unsigned long long);
19225 On 64-bit targets, if the ISA 3.0 additions (@option{-mcpu=power9})
19229 vector long vec_vprtyb (vector long);
19230 vector unsigned long vec_vprtyb (vector unsigned long);
19231 vector __int128_t vec_vprtyb (vector __int128_t);
19232 vector __uint128_t vec_vprtyb (vector __uint128_t);
19234 vector long vec_vprtybd (vector long);
19235 vector unsigned long vec_vprtybd (vector unsigned long);
19237 vector __int128_t vec_vprtybq (vector __int128_t);
19238 vector __uint128_t vec_vprtybd (vector __uint128_t);
19241 The following built-in vector functions are available for the PowerPC family
19242 of processors, starting with ISA 3.0 or later (@option{-mcpu=power9}):
19244 __vector unsigned char
19245 vec_slv (__vector unsigned char src, __vector unsigned char shift_distance);
19246 __vector unsigned char
19247 vec_srv (__vector unsigned char src, __vector unsigned char shift_distance);
19250 The @code{vec_slv} and @code{vec_srv} functions operate on
19251 all of the bytes of their @code{src} and @code{shift_distance}
19252 arguments in parallel. The behavior of the @code{vec_slv} is as if
19253 there existed a temporary array of 17 unsigned characters
19254 @code{slv_array} within which elements 0 through 15 are the same as
19255 the entries in the @code{src} array and element 16 equals 0. The
19256 result returned from the @code{vec_slv} function is a
19257 @code{__vector} of 16 unsigned characters within which element
19258 @code{i} is computed using the C expression
19259 @code{0xff & (*((unsigned short *)(slv_array + i)) << (0x07 &
19260 shift_distance[i]))},
19261 with this resulting value coerced to the @code{unsigned char} type.
19262 The behavior of the @code{vec_srv} is as if
19263 there existed a temporary array of 17 unsigned characters
19264 @code{srv_array} within which element 0 equals zero and
19265 elements 1 through 16 equal the elements 0 through 15 of
19266 the @code{src} array. The
19267 result returned from the @code{vec_srv} function is a
19268 @code{__vector} of 16 unsigned characters within which element
19269 @code{i} is computed using the C expression
19270 @code{0xff & (*((unsigned short *)(srv_array + i)) >>
19271 (0x07 & shift_distance[i]))},
19272 with this resulting value coerced to the @code{unsigned char} type.
19274 The following built-in functions are available for the PowerPC family
19275 of processors, starting with ISA 3.0 or later (@option{-mcpu=power9}):
19277 __vector unsigned char
19278 vec_absd (__vector unsigned char arg1, __vector unsigned char arg2);
19279 __vector unsigned short
19280 vec_absd (__vector unsigned short arg1, __vector unsigned short arg2);
19281 __vector unsigned int
19282 vec_absd (__vector unsigned int arg1, __vector unsigned int arg2);
19284 __vector unsigned char
19285 vec_absdb (__vector unsigned char arg1, __vector unsigned char arg2);
19286 __vector unsigned short
19287 vec_absdh (__vector unsigned short arg1, __vector unsigned short arg2);
19288 __vector unsigned int
19289 vec_absdw (__vector unsigned int arg1, __vector unsigned int arg2);
19292 The @code{vec_absd}, @code{vec_absdb}, @code{vec_absdh}, and
19293 @code{vec_absdw} built-in functions each computes the absolute
19294 differences of the pairs of vector elements supplied in its two vector
19295 arguments, placing the absolute differences into the corresponding
19296 elements of the vector result.
19298 The following built-in functions are available for the PowerPC family
19299 of processors, starting with ISA 3.0 or later (@option{-mcpu=power9}):
19301 __vector unsigned int
19302 vec_extract_exp (__vector float source);
19303 __vector unsigned long long int
19304 vec_extract_exp (__vector double source);
19306 __vector unsigned int
19307 vec_extract_sig (__vector float source);
19308 __vector unsigned long long int
19309 vec_extract_sig (__vector double source);
19312 vec_insert_exp (__vector unsigned int significands,
19313 __vector unsigned int exponents);
19315 vec_insert_exp (__vector unsigned float significands,
19316 __vector unsigned int exponents);
19318 vec_insert_exp (__vector unsigned long long int significands,
19319 __vector unsigned long long int exponents);
19321 vec_insert_exp (__vector unsigned double significands,
19322 __vector unsigned long long int exponents);
19324 __vector bool int vec_test_data_class (__vector float source,
19325 const int condition);
19326 __vector bool long long int vec_test_data_class (__vector double source,
19327 const int condition);
19330 The @code{vec_extract_sig} and @code{vec_extract_exp} built-in
19331 functions return vectors representing the significands and biased
19332 exponent values of their @code{source} arguments respectively.
19333 Within the result vector returned by @code{vec_extract_sig}, the
19334 @code{0x800000} bit of each vector element returned when the
19335 function's @code{source} argument is of type @code{float} is set to 1
19336 if the corresponding floating point value is in normalized form.
19337 Otherwise, this bit is set to 0. When the @code{source} argument is
19338 of type @code{double}, the @code{0x10000000000000} bit within each of
19339 the result vector's elements is set according to the same rules.
19340 Note that the sign of the significand is not represented in the result
19341 returned from the @code{vec_extract_sig} function. To extract the
19343 @code{vec_cpsgn} function, which returns a new vector within which all
19344 of the sign bits of its second argument vector are overwritten with the
19345 sign bits copied from the coresponding elements of its first argument
19346 vector, and all other (non-sign) bits of the second argument vector
19347 are copied unchanged into the result vector.
19349 The @code{vec_insert_exp} built-in functions return a vector of
19350 single- or double-precision floating
19351 point values constructed by assembling the values of their
19352 @code{significands} and @code{exponents} arguments into the
19353 corresponding elements of the returned vector.
19355 element of the result is copied from the most significant bit of the
19356 corresponding entry within the @code{significands} argument.
19357 Note that the relevant
19358 bits of the @code{significands} argument are the same, for both integer
19359 and floating point types.
19361 significand and exponent components of each element of the result are
19362 composed of the least significant bits of the corresponding
19363 @code{significands} element and the least significant bits of the
19364 corresponding @code{exponents} element.
19366 The @code{vec_test_data_class} built-in function returns a vector
19367 representing the results of testing the @code{source} vector for the
19368 condition selected by the @code{condition} argument. The
19369 @code{condition} argument must be a compile-time constant integer with
19370 value not exceeding 127. The
19371 @code{condition} argument is encoded as a bitmask with each bit
19372 enabling the testing of a different condition, as characterized by the
19376 0x20 Test for +Infinity
19377 0x10 Test for -Infinity
19378 0x08 Test for +Zero
19379 0x04 Test for -Zero
19380 0x02 Test for +Denormal
19381 0x01 Test for -Denormal
19384 If any of the enabled test conditions is true, the corresponding entry
19385 in the result vector is -1. Otherwise (all of the enabled test
19386 conditions are false), the corresponding entry of the result vector is 0.
19388 The following built-in functions are available for the PowerPC family
19389 of processors, starting with ISA 3.0 or later (@option{-mcpu=power9}):
19391 vector unsigned int vec_rlmi (vector unsigned int, vector unsigned int,
19392 vector unsigned int);
19393 vector unsigned long long vec_rlmi (vector unsigned long long,
19394 vector unsigned long long,
19395 vector unsigned long long);
19396 vector unsigned int vec_rlnm (vector unsigned int, vector unsigned int,
19397 vector unsigned int);
19398 vector unsigned long long vec_rlnm (vector unsigned long long,
19399 vector unsigned long long,
19400 vector unsigned long long);
19401 vector unsigned int vec_vrlnm (vector unsigned int, vector unsigned int);
19402 vector unsigned long long vec_vrlnm (vector unsigned long long,
19403 vector unsigned long long);
19406 The result of @code{vec_rlmi} is obtained by rotating each element of
19407 the first argument vector left and inserting it under mask into the
19408 second argument vector. The third argument vector contains the mask
19409 beginning in bits 11:15, the mask end in bits 19:23, and the shift
19410 count in bits 27:31, of each element.
19412 The result of @code{vec_rlnm} is obtained by rotating each element of
19413 the first argument vector left and ANDing it with a mask specified by
19414 the second and third argument vectors. The second argument vector
19415 contains the shift count for each element in the low-order byte. The
19416 third argument vector contains the mask end for each element in the
19417 low-order byte, with the mask begin in the next higher byte.
19419 The result of @code{vec_vrlnm} is obtained by rotating each element
19420 of the first argument vector left and ANDing it with a mask. The
19421 second argument vector contains the mask beginning in bits 11:15,
19422 the mask end in bits 19:23, and the shift count in bits 27:31,
19425 If the ISA 3.0 instruction set additions (@option{-mcpu=power9})
19428 vector signed bool char vec_revb (vector signed char);
19429 vector signed char vec_revb (vector signed char);
19430 vector unsigned char vec_revb (vector unsigned char);
19431 vector bool short vec_revb (vector bool short);
19432 vector short vec_revb (vector short);
19433 vector unsigned short vec_revb (vector unsigned short);
19434 vector bool int vec_revb (vector bool int);
19435 vector int vec_revb (vector int);
19436 vector unsigned int vec_revb (vector unsigned int);
19437 vector float vec_revb (vector float);
19438 vector bool long long vec_revb (vector bool long long);
19439 vector long long vec_revb (vector long long);
19440 vector unsigned long long vec_revb (vector unsigned long long);
19441 vector double vec_revb (vector double);
19444 On 64-bit targets, if the ISA 3.0 additions (@option{-mcpu=power9})
19447 vector long vec_revb (vector long);
19448 vector unsigned long vec_revb (vector unsigned long);
19449 vector __int128_t vec_revb (vector __int128_t);
19450 vector __uint128_t vec_revb (vector __uint128_t);
19453 The @code{vec_revb} built-in function reverses the bytes on an element
19454 by element basis. A vector of @code{vector unsigned char} or
19455 @code{vector signed char} reverses the bytes in the whole word.
19457 If the cryptographic instructions are enabled (@option{-mcrypto} or
19458 @option{-mcpu=power8}), the following builtins are enabled.
19461 vector unsigned long long __builtin_crypto_vsbox (vector unsigned long long);
19463 vector unsigned long long __builtin_crypto_vcipher (vector unsigned long long,
19464 vector unsigned long long);
19466 vector unsigned long long __builtin_crypto_vcipherlast
19467 (vector unsigned long long,
19468 vector unsigned long long);
19470 vector unsigned long long __builtin_crypto_vncipher (vector unsigned long long,
19471 vector unsigned long long);
19473 vector unsigned long long __builtin_crypto_vncipherlast
19474 (vector unsigned long long,
19475 vector unsigned long long);
19477 vector unsigned char __builtin_crypto_vpermxor (vector unsigned char,
19478 vector unsigned char,
19479 vector unsigned char);
19481 vector unsigned short __builtin_crypto_vpermxor (vector unsigned short,
19482 vector unsigned short,
19483 vector unsigned short);
19485 vector unsigned int __builtin_crypto_vpermxor (vector unsigned int,
19486 vector unsigned int,
19487 vector unsigned int);
19489 vector unsigned long long __builtin_crypto_vpermxor (vector unsigned long long,
19490 vector unsigned long long,
19491 vector unsigned long long);
19493 vector unsigned char __builtin_crypto_vpmsumb (vector unsigned char,
19494 vector unsigned char);
19496 vector unsigned short __builtin_crypto_vpmsumb (vector unsigned short,
19497 vector unsigned short);
19499 vector unsigned int __builtin_crypto_vpmsumb (vector unsigned int,
19500 vector unsigned int);
19502 vector unsigned long long __builtin_crypto_vpmsumb (vector unsigned long long,
19503 vector unsigned long long);
19505 vector unsigned long long __builtin_crypto_vshasigmad
19506 (vector unsigned long long, int, int);
19508 vector unsigned int __builtin_crypto_vshasigmaw (vector unsigned int,
19512 The second argument to @var{__builtin_crypto_vshasigmad} and
19513 @var{__builtin_crypto_vshasigmaw} must be a constant
19514 integer that is 0 or 1. The third argument to these built-in functions
19515 must be a constant integer in the range of 0 to 15.
19517 If the ISA 3.0 instruction set additions
19518 are enabled (@option{-mcpu=power9}), the following additional
19519 functions are available for both 32-bit and 64-bit targets.
19521 vector short vec_xl (int, vector short *);
19522 vector short vec_xl (int, short *);
19523 vector unsigned short vec_xl (int, vector unsigned short *);
19524 vector unsigned short vec_xl (int, unsigned short *);
19525 vector char vec_xl (int, vector char *);
19526 vector char vec_xl (int, char *);
19527 vector unsigned char vec_xl (int, vector unsigned char *);
19528 vector unsigned char vec_xl (int, unsigned char *);
19530 void vec_xst (vector short, int, vector short *);
19531 void vec_xst (vector short, int, short *);
19532 void vec_xst (vector unsigned short, int, vector unsigned short *);
19533 void vec_xst (vector unsigned short, int, unsigned short *);
19534 void vec_xst (vector char, int, vector char *);
19535 void vec_xst (vector char, int, char *);
19536 void vec_xst (vector unsigned char, int, vector unsigned char *);
19537 void vec_xst (vector unsigned char, int, unsigned char *);
19539 @node PowerPC Hardware Transactional Memory Built-in Functions
19540 @subsection PowerPC Hardware Transactional Memory Built-in Functions
19541 GCC provides two interfaces for accessing the Hardware Transactional
19542 Memory (HTM) instructions available on some of the PowerPC family
19543 of processors (eg, POWER8). The two interfaces come in a low level
19544 interface, consisting of built-in functions specific to PowerPC and a
19545 higher level interface consisting of inline functions that are common
19546 between PowerPC and S/390.
19548 @subsubsection PowerPC HTM Low Level Built-in Functions
19550 The following low level built-in functions are available with
19551 @option{-mhtm} or @option{-mcpu=CPU} where CPU is `power8' or later.
19552 They all generate the machine instruction that is part of the name.
19554 The HTM builtins (with the exception of @code{__builtin_tbegin}) return
19555 the full 4-bit condition register value set by their associated hardware
19556 instruction. The header file @code{htmintrin.h} defines some macros that can
19557 be used to decipher the return value. The @code{__builtin_tbegin} builtin
19558 returns a simple true or false value depending on whether a transaction was
19559 successfully started or not. The arguments of the builtins match exactly the
19560 type and order of the associated hardware instruction's operands, except for
19561 the @code{__builtin_tcheck} builtin, which does not take any input arguments.
19562 Refer to the ISA manual for a description of each instruction's operands.
19565 unsigned int __builtin_tbegin (unsigned int)
19566 unsigned int __builtin_tend (unsigned int)
19568 unsigned int __builtin_tabort (unsigned int)
19569 unsigned int __builtin_tabortdc (unsigned int, unsigned int, unsigned int)
19570 unsigned int __builtin_tabortdci (unsigned int, unsigned int, int)
19571 unsigned int __builtin_tabortwc (unsigned int, unsigned int, unsigned int)
19572 unsigned int __builtin_tabortwci (unsigned int, unsigned int, int)
19574 unsigned int __builtin_tcheck (void)
19575 unsigned int __builtin_treclaim (unsigned int)
19576 unsigned int __builtin_trechkpt (void)
19577 unsigned int __builtin_tsr (unsigned int)
19580 In addition to the above HTM built-ins, we have added built-ins for
19581 some common extended mnemonics of the HTM instructions:
19584 unsigned int __builtin_tendall (void)
19585 unsigned int __builtin_tresume (void)
19586 unsigned int __builtin_tsuspend (void)
19589 Note that the semantics of the above HTM builtins are required to mimic
19590 the locking semantics used for critical sections. Builtins that are used
19591 to create a new transaction or restart a suspended transaction must have
19592 lock acquisition like semantics while those builtins that end or suspend a
19593 transaction must have lock release like semantics. Specifically, this must
19594 mimic lock semantics as specified by C++11, for example: Lock acquisition is
19595 as-if an execution of __atomic_exchange_n(&globallock,1,__ATOMIC_ACQUIRE)
19596 that returns 0, and lock release is as-if an execution of
19597 __atomic_store(&globallock,0,__ATOMIC_RELEASE), with globallock being an
19598 implicit implementation-defined lock used for all transactions. The HTM
19599 instructions associated with with the builtins inherently provide the
19600 correct acquisition and release hardware barriers required. However,
19601 the compiler must also be prohibited from moving loads and stores across
19602 the builtins in a way that would violate their semantics. This has been
19603 accomplished by adding memory barriers to the associated HTM instructions
19604 (which is a conservative approach to provide acquire and release semantics).
19605 Earlier versions of the compiler did not treat the HTM instructions as
19606 memory barriers. A @code{__TM_FENCE__} macro has been added, which can
19607 be used to determine whether the current compiler treats HTM instructions
19608 as memory barriers or not. This allows the user to explicitly add memory
19609 barriers to their code when using an older version of the compiler.
19611 The following set of built-in functions are available to gain access
19612 to the HTM specific special purpose registers.
19615 unsigned long __builtin_get_texasr (void)
19616 unsigned long __builtin_get_texasru (void)
19617 unsigned long __builtin_get_tfhar (void)
19618 unsigned long __builtin_get_tfiar (void)
19620 void __builtin_set_texasr (unsigned long);
19621 void __builtin_set_texasru (unsigned long);
19622 void __builtin_set_tfhar (unsigned long);
19623 void __builtin_set_tfiar (unsigned long);
19626 Example usage of these low level built-in functions may look like:
19629 #include <htmintrin.h>
19631 int num_retries = 10;
19635 if (__builtin_tbegin (0))
19637 /* Transaction State Initiated. */
19638 if (is_locked (lock))
19639 __builtin_tabort (0);
19640 ... transaction code...
19641 __builtin_tend (0);
19646 /* Transaction State Failed. Use locks if the transaction
19647 failure is "persistent" or we've tried too many times. */
19648 if (num_retries-- <= 0
19649 || _TEXASRU_FAILURE_PERSISTENT (__builtin_get_texasru ()))
19651 acquire_lock (lock);
19652 ... non transactional fallback path...
19653 release_lock (lock);
19660 One final built-in function has been added that returns the value of
19661 the 2-bit Transaction State field of the Machine Status Register (MSR)
19662 as stored in @code{CR0}.
19665 unsigned long __builtin_ttest (void)
19668 This built-in can be used to determine the current transaction state
19669 using the following code example:
19672 #include <htmintrin.h>
19674 unsigned char tx_state = _HTM_STATE (__builtin_ttest ());
19676 if (tx_state == _HTM_TRANSACTIONAL)
19678 /* Code to use in transactional state. */
19680 else if (tx_state == _HTM_NONTRANSACTIONAL)
19682 /* Code to use in non-transactional state. */
19684 else if (tx_state == _HTM_SUSPENDED)
19686 /* Code to use in transaction suspended state. */
19690 @subsubsection PowerPC HTM High Level Inline Functions
19692 The following high level HTM interface is made available by including
19693 @code{<htmxlintrin.h>} and using @option{-mhtm} or @option{-mcpu=CPU}
19694 where CPU is `power8' or later. This interface is common between PowerPC
19695 and S/390, allowing users to write one HTM source implementation that
19696 can be compiled and executed on either system.
19699 long __TM_simple_begin (void)
19700 long __TM_begin (void* const TM_buff)
19701 long __TM_end (void)
19702 void __TM_abort (void)
19703 void __TM_named_abort (unsigned char const code)
19704 void __TM_resume (void)
19705 void __TM_suspend (void)
19707 long __TM_is_user_abort (void* const TM_buff)
19708 long __TM_is_named_user_abort (void* const TM_buff, unsigned char *code)
19709 long __TM_is_illegal (void* const TM_buff)
19710 long __TM_is_footprint_exceeded (void* const TM_buff)
19711 long __TM_nesting_depth (void* const TM_buff)
19712 long __TM_is_nested_too_deep(void* const TM_buff)
19713 long __TM_is_conflict(void* const TM_buff)
19714 long __TM_is_failure_persistent(void* const TM_buff)
19715 long __TM_failure_address(void* const TM_buff)
19716 long long __TM_failure_code(void* const TM_buff)
19719 Using these common set of HTM inline functions, we can create
19720 a more portable version of the HTM example in the previous
19721 section that will work on either PowerPC or S/390:
19724 #include <htmxlintrin.h>
19726 int num_retries = 10;
19727 TM_buff_type TM_buff;
19731 if (__TM_begin (TM_buff) == _HTM_TBEGIN_STARTED)
19733 /* Transaction State Initiated. */
19734 if (is_locked (lock))
19736 ... transaction code...
19742 /* Transaction State Failed. Use locks if the transaction
19743 failure is "persistent" or we've tried too many times. */
19744 if (num_retries-- <= 0
19745 || __TM_is_failure_persistent (TM_buff))
19747 acquire_lock (lock);
19748 ... non transactional fallback path...
19749 release_lock (lock);
19756 @node PowerPC Atomic Memory Operation Functions
19757 @subsection PowerPC Atomic Memory Operation Functions
19758 ISA 3.0 of the PowerPC added new atomic memory operation (amo)
19759 instructions. GCC provides support for these instructions in 64-bit
19760 environments. All of the functions are declared in the include file
19763 The functions supported are:
19768 uint32_t amo_lwat_add (uint32_t *, uint32_t);
19769 uint32_t amo_lwat_xor (uint32_t *, uint32_t);
19770 uint32_t amo_lwat_ior (uint32_t *, uint32_t);
19771 uint32_t amo_lwat_and (uint32_t *, uint32_t);
19772 uint32_t amo_lwat_umax (uint32_t *, uint32_t);
19773 uint32_t amo_lwat_umin (uint32_t *, uint32_t);
19774 uint32_t amo_lwat_swap (uint32_t *, uint32_t);
19776 int32_t amo_lwat_sadd (int32_t *, int32_t);
19777 int32_t amo_lwat_smax (int32_t *, int32_t);
19778 int32_t amo_lwat_smin (int32_t *, int32_t);
19779 int32_t amo_lwat_sswap (int32_t *, int32_t);
19781 uint64_t amo_ldat_add (uint64_t *, uint64_t);
19782 uint64_t amo_ldat_xor (uint64_t *, uint64_t);
19783 uint64_t amo_ldat_ior (uint64_t *, uint64_t);
19784 uint64_t amo_ldat_and (uint64_t *, uint64_t);
19785 uint64_t amo_ldat_umax (uint64_t *, uint64_t);
19786 uint64_t amo_ldat_umin (uint64_t *, uint64_t);
19787 uint64_t amo_ldat_swap (uint64_t *, uint64_t);
19789 int64_t amo_ldat_sadd (int64_t *, int64_t);
19790 int64_t amo_ldat_smax (int64_t *, int64_t);
19791 int64_t amo_ldat_smin (int64_t *, int64_t);
19792 int64_t amo_ldat_sswap (int64_t *, int64_t);
19794 void amo_stwat_add (uint32_t *, uint32_t);
19795 void amo_stwat_xor (uint32_t *, uint32_t);
19796 void amo_stwat_ior (uint32_t *, uint32_t);
19797 void amo_stwat_and (uint32_t *, uint32_t);
19798 void amo_stwat_umax (uint32_t *, uint32_t);
19799 void amo_stwat_umin (uint32_t *, uint32_t);
19801 void amo_stwat_sadd (int32_t *, int32_t);
19802 void amo_stwat_smax (int32_t *, int32_t);
19803 void amo_stwat_smin (int32_t *, int32_t);
19805 void amo_stdat_add (uint64_t *, uint64_t);
19806 void amo_stdat_xor (uint64_t *, uint64_t);
19807 void amo_stdat_ior (uint64_t *, uint64_t);
19808 void amo_stdat_and (uint64_t *, uint64_t);
19809 void amo_stdat_umax (uint64_t *, uint64_t);
19810 void amo_stdat_umin (uint64_t *, uint64_t);
19812 void amo_stdat_sadd (int64_t *, int64_t);
19813 void amo_stdat_smax (int64_t *, int64_t);
19814 void amo_stdat_smin (int64_t *, int64_t);
19817 @node RX Built-in Functions
19818 @subsection RX Built-in Functions
19819 GCC supports some of the RX instructions which cannot be expressed in
19820 the C programming language via the use of built-in functions. The
19821 following functions are supported:
19823 @deftypefn {Built-in Function} void __builtin_rx_brk (void)
19824 Generates the @code{brk} machine instruction.
19827 @deftypefn {Built-in Function} void __builtin_rx_clrpsw (int)
19828 Generates the @code{clrpsw} machine instruction to clear the specified
19829 bit in the processor status word.
19832 @deftypefn {Built-in Function} void __builtin_rx_int (int)
19833 Generates the @code{int} machine instruction to generate an interrupt
19834 with the specified value.
19837 @deftypefn {Built-in Function} void __builtin_rx_machi (int, int)
19838 Generates the @code{machi} machine instruction to add the result of
19839 multiplying the top 16 bits of the two arguments into the
19843 @deftypefn {Built-in Function} void __builtin_rx_maclo (int, int)
19844 Generates the @code{maclo} machine instruction to add the result of
19845 multiplying the bottom 16 bits of the two arguments into the
19849 @deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int)
19850 Generates the @code{mulhi} machine instruction to place the result of
19851 multiplying the top 16 bits of the two arguments into the
19855 @deftypefn {Built-in Function} void __builtin_rx_mullo (int, int)
19856 Generates the @code{mullo} machine instruction to place the result of
19857 multiplying the bottom 16 bits of the two arguments into the
19861 @deftypefn {Built-in Function} int __builtin_rx_mvfachi (void)
19862 Generates the @code{mvfachi} machine instruction to read the top
19863 32 bits of the accumulator.
19866 @deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void)
19867 Generates the @code{mvfacmi} machine instruction to read the middle
19868 32 bits of the accumulator.
19871 @deftypefn {Built-in Function} int __builtin_rx_mvfc (int)
19872 Generates the @code{mvfc} machine instruction which reads the control
19873 register specified in its argument and returns its value.
19876 @deftypefn {Built-in Function} void __builtin_rx_mvtachi (int)
19877 Generates the @code{mvtachi} machine instruction to set the top
19878 32 bits of the accumulator.
19881 @deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int)
19882 Generates the @code{mvtaclo} machine instruction to set the bottom
19883 32 bits of the accumulator.
19886 @deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val)
19887 Generates the @code{mvtc} machine instruction which sets control
19888 register number @code{reg} to @code{val}.
19891 @deftypefn {Built-in Function} void __builtin_rx_mvtipl (int)
19892 Generates the @code{mvtipl} machine instruction set the interrupt
19896 @deftypefn {Built-in Function} void __builtin_rx_racw (int)
19897 Generates the @code{racw} machine instruction to round the accumulator
19898 according to the specified mode.
19901 @deftypefn {Built-in Function} int __builtin_rx_revw (int)
19902 Generates the @code{revw} machine instruction which swaps the bytes in
19903 the argument so that bits 0--7 now occupy bits 8--15 and vice versa,
19904 and also bits 16--23 occupy bits 24--31 and vice versa.
19907 @deftypefn {Built-in Function} void __builtin_rx_rmpa (void)
19908 Generates the @code{rmpa} machine instruction which initiates a
19909 repeated multiply and accumulate sequence.
19912 @deftypefn {Built-in Function} void __builtin_rx_round (float)
19913 Generates the @code{round} machine instruction which returns the
19914 floating-point argument rounded according to the current rounding mode
19915 set in the floating-point status word register.
19918 @deftypefn {Built-in Function} int __builtin_rx_sat (int)
19919 Generates the @code{sat} machine instruction which returns the
19920 saturated value of the argument.
19923 @deftypefn {Built-in Function} void __builtin_rx_setpsw (int)
19924 Generates the @code{setpsw} machine instruction to set the specified
19925 bit in the processor status word.
19928 @deftypefn {Built-in Function} void __builtin_rx_wait (void)
19929 Generates the @code{wait} machine instruction.
19932 @node S/390 System z Built-in Functions
19933 @subsection S/390 System z Built-in Functions
19934 @deftypefn {Built-in Function} int __builtin_tbegin (void*)
19935 Generates the @code{tbegin} machine instruction starting a
19936 non-constrained hardware transaction. If the parameter is non-NULL the
19937 memory area is used to store the transaction diagnostic buffer and
19938 will be passed as first operand to @code{tbegin}. This buffer can be
19939 defined using the @code{struct __htm_tdb} C struct defined in
19940 @code{htmintrin.h} and must reside on a double-word boundary. The
19941 second tbegin operand is set to @code{0xff0c}. This enables
19942 save/restore of all GPRs and disables aborts for FPR and AR
19943 manipulations inside the transaction body. The condition code set by
19944 the tbegin instruction is returned as integer value. The tbegin
19945 instruction by definition overwrites the content of all FPRs. The
19946 compiler will generate code which saves and restores the FPRs. For
19947 soft-float code it is recommended to used the @code{*_nofloat}
19948 variant. In order to prevent a TDB from being written it is required
19949 to pass a constant zero value as parameter. Passing a zero value
19950 through a variable is not sufficient. Although modifications of
19951 access registers inside the transaction will not trigger an
19952 transaction abort it is not supported to actually modify them. Access
19953 registers do not get saved when entering a transaction. They will have
19954 undefined state when reaching the abort code.
19957 Macros for the possible return codes of tbegin are defined in the
19958 @code{htmintrin.h} header file:
19961 @item _HTM_TBEGIN_STARTED
19962 @code{tbegin} has been executed as part of normal processing. The
19963 transaction body is supposed to be executed.
19964 @item _HTM_TBEGIN_INDETERMINATE
19965 The transaction was aborted due to an indeterminate condition which
19966 might be persistent.
19967 @item _HTM_TBEGIN_TRANSIENT
19968 The transaction aborted due to a transient failure. The transaction
19969 should be re-executed in that case.
19970 @item _HTM_TBEGIN_PERSISTENT
19971 The transaction aborted due to a persistent failure. Re-execution
19972 under same circumstances will not be productive.
19975 @defmac _HTM_FIRST_USER_ABORT_CODE
19976 The @code{_HTM_FIRST_USER_ABORT_CODE} defined in @code{htmintrin.h}
19977 specifies the first abort code which can be used for
19978 @code{__builtin_tabort}. Values below this threshold are reserved for
19982 @deftp {Data type} {struct __htm_tdb}
19983 The @code{struct __htm_tdb} defined in @code{htmintrin.h} describes
19984 the structure of the transaction diagnostic block as specified in the
19985 Principles of Operation manual chapter 5-91.
19988 @deftypefn {Built-in Function} int __builtin_tbegin_nofloat (void*)
19989 Same as @code{__builtin_tbegin} but without FPR saves and restores.
19990 Using this variant in code making use of FPRs will leave the FPRs in
19991 undefined state when entering the transaction abort handler code.
19994 @deftypefn {Built-in Function} int __builtin_tbegin_retry (void*, int)
19995 In addition to @code{__builtin_tbegin} a loop for transient failures
19996 is generated. If tbegin returns a condition code of 2 the transaction
19997 will be retried as often as specified in the second argument. The
19998 perform processor assist instruction is used to tell the CPU about the
19999 number of fails so far.
20002 @deftypefn {Built-in Function} int __builtin_tbegin_retry_nofloat (void*, int)
20003 Same as @code{__builtin_tbegin_retry} but without FPR saves and
20004 restores. Using this variant in code making use of FPRs will leave
20005 the FPRs in undefined state when entering the transaction abort
20009 @deftypefn {Built-in Function} void __builtin_tbeginc (void)
20010 Generates the @code{tbeginc} machine instruction starting a constrained
20011 hardware transaction. The second operand is set to @code{0xff08}.
20014 @deftypefn {Built-in Function} int __builtin_tend (void)
20015 Generates the @code{tend} machine instruction finishing a transaction
20016 and making the changes visible to other threads. The condition code
20017 generated by tend is returned as integer value.
20020 @deftypefn {Built-in Function} void __builtin_tabort (int)
20021 Generates the @code{tabort} machine instruction with the specified
20022 abort code. Abort codes from 0 through 255 are reserved and will
20023 result in an error message.
20026 @deftypefn {Built-in Function} void __builtin_tx_assist (int)
20027 Generates the @code{ppa rX,rY,1} machine instruction. Where the
20028 integer parameter is loaded into rX and a value of zero is loaded into
20029 rY. The integer parameter specifies the number of times the
20030 transaction repeatedly aborted.
20033 @deftypefn {Built-in Function} int __builtin_tx_nesting_depth (void)
20034 Generates the @code{etnd} machine instruction. The current nesting
20035 depth is returned as integer value. For a nesting depth of 0 the code
20036 is not executed as part of an transaction.
20039 @deftypefn {Built-in Function} void __builtin_non_tx_store (uint64_t *, uint64_t)
20041 Generates the @code{ntstg} machine instruction. The second argument
20042 is written to the first arguments location. The store operation will
20043 not be rolled-back in case of an transaction abort.
20046 @node SH Built-in Functions
20047 @subsection SH Built-in Functions
20048 The following built-in functions are supported on the SH1, SH2, SH3 and SH4
20049 families of processors:
20051 @deftypefn {Built-in Function} {void} __builtin_set_thread_pointer (void *@var{ptr})
20052 Sets the @samp{GBR} register to the specified value @var{ptr}. This is usually
20053 used by system code that manages threads and execution contexts. The compiler
20054 normally does not generate code that modifies the contents of @samp{GBR} and
20055 thus the value is preserved across function calls. Changing the @samp{GBR}
20056 value in user code must be done with caution, since the compiler might use
20057 @samp{GBR} in order to access thread local variables.
20061 @deftypefn {Built-in Function} {void *} __builtin_thread_pointer (void)
20062 Returns the value that is currently set in the @samp{GBR} register.
20063 Memory loads and stores that use the thread pointer as a base address are
20064 turned into @samp{GBR} based displacement loads and stores, if possible.
20072 int get_tcb_value (void)
20074 // Generate @samp{mov.l @@(8,gbr),r0} instruction
20075 return ((my_tcb*)__builtin_thread_pointer ())->c;
20081 @deftypefn {Built-in Function} {unsigned int} __builtin_sh_get_fpscr (void)
20082 Returns the value that is currently set in the @samp{FPSCR} register.
20085 @deftypefn {Built-in Function} {void} __builtin_sh_set_fpscr (unsigned int @var{val})
20086 Sets the @samp{FPSCR} register to the specified value @var{val}, while
20087 preserving the current values of the FR, SZ and PR bits.
20090 @node SPARC VIS Built-in Functions
20091 @subsection SPARC VIS Built-in Functions
20093 GCC supports SIMD operations on the SPARC using both the generic vector
20094 extensions (@pxref{Vector Extensions}) as well as built-in functions for
20095 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
20096 switch, the VIS extension is exposed as the following built-in functions:
20099 typedef int v1si __attribute__ ((vector_size (4)));
20100 typedef int v2si __attribute__ ((vector_size (8)));
20101 typedef short v4hi __attribute__ ((vector_size (8)));
20102 typedef short v2hi __attribute__ ((vector_size (4)));
20103 typedef unsigned char v8qi __attribute__ ((vector_size (8)));
20104 typedef unsigned char v4qi __attribute__ ((vector_size (4)));
20106 void __builtin_vis_write_gsr (int64_t);
20107 int64_t __builtin_vis_read_gsr (void);
20109 void * __builtin_vis_alignaddr (void *, long);
20110 void * __builtin_vis_alignaddrl (void *, long);
20111 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
20112 v2si __builtin_vis_faligndatav2si (v2si, v2si);
20113 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
20114 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
20116 v4hi __builtin_vis_fexpand (v4qi);
20118 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
20119 v4hi __builtin_vis_fmul8x16au (v4qi, v2hi);
20120 v4hi __builtin_vis_fmul8x16al (v4qi, v2hi);
20121 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
20122 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
20123 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
20124 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
20126 v4qi __builtin_vis_fpack16 (v4hi);
20127 v8qi __builtin_vis_fpack32 (v2si, v8qi);
20128 v2hi __builtin_vis_fpackfix (v2si);
20129 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
20131 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
20133 long __builtin_vis_edge8 (void *, void *);
20134 long __builtin_vis_edge8l (void *, void *);
20135 long __builtin_vis_edge16 (void *, void *);
20136 long __builtin_vis_edge16l (void *, void *);
20137 long __builtin_vis_edge32 (void *, void *);
20138 long __builtin_vis_edge32l (void *, void *);
20140 long __builtin_vis_fcmple16 (v4hi, v4hi);
20141 long __builtin_vis_fcmple32 (v2si, v2si);
20142 long __builtin_vis_fcmpne16 (v4hi, v4hi);
20143 long __builtin_vis_fcmpne32 (v2si, v2si);
20144 long __builtin_vis_fcmpgt16 (v4hi, v4hi);
20145 long __builtin_vis_fcmpgt32 (v2si, v2si);
20146 long __builtin_vis_fcmpeq16 (v4hi, v4hi);
20147 long __builtin_vis_fcmpeq32 (v2si, v2si);
20149 v4hi __builtin_vis_fpadd16 (v4hi, v4hi);
20150 v2hi __builtin_vis_fpadd16s (v2hi, v2hi);
20151 v2si __builtin_vis_fpadd32 (v2si, v2si);
20152 v1si __builtin_vis_fpadd32s (v1si, v1si);
20153 v4hi __builtin_vis_fpsub16 (v4hi, v4hi);
20154 v2hi __builtin_vis_fpsub16s (v2hi, v2hi);
20155 v2si __builtin_vis_fpsub32 (v2si, v2si);
20156 v1si __builtin_vis_fpsub32s (v1si, v1si);
20158 long __builtin_vis_array8 (long, long);
20159 long __builtin_vis_array16 (long, long);
20160 long __builtin_vis_array32 (long, long);
20163 When you use the @option{-mvis2} switch, the VIS version 2.0 built-in
20164 functions also become available:
20167 long __builtin_vis_bmask (long, long);
20168 int64_t __builtin_vis_bshuffledi (int64_t, int64_t);
20169 v2si __builtin_vis_bshufflev2si (v2si, v2si);
20170 v4hi __builtin_vis_bshufflev2si (v4hi, v4hi);
20171 v8qi __builtin_vis_bshufflev2si (v8qi, v8qi);
20173 long __builtin_vis_edge8n (void *, void *);
20174 long __builtin_vis_edge8ln (void *, void *);
20175 long __builtin_vis_edge16n (void *, void *);
20176 long __builtin_vis_edge16ln (void *, void *);
20177 long __builtin_vis_edge32n (void *, void *);
20178 long __builtin_vis_edge32ln (void *, void *);
20181 When you use the @option{-mvis3} switch, the VIS version 3.0 built-in
20182 functions also become available:
20185 void __builtin_vis_cmask8 (long);
20186 void __builtin_vis_cmask16 (long);
20187 void __builtin_vis_cmask32 (long);
20189 v4hi __builtin_vis_fchksm16 (v4hi, v4hi);
20191 v4hi __builtin_vis_fsll16 (v4hi, v4hi);
20192 v4hi __builtin_vis_fslas16 (v4hi, v4hi);
20193 v4hi __builtin_vis_fsrl16 (v4hi, v4hi);
20194 v4hi __builtin_vis_fsra16 (v4hi, v4hi);
20195 v2si __builtin_vis_fsll16 (v2si, v2si);
20196 v2si __builtin_vis_fslas16 (v2si, v2si);
20197 v2si __builtin_vis_fsrl16 (v2si, v2si);
20198 v2si __builtin_vis_fsra16 (v2si, v2si);
20200 long __builtin_vis_pdistn (v8qi, v8qi);
20202 v4hi __builtin_vis_fmean16 (v4hi, v4hi);
20204 int64_t __builtin_vis_fpadd64 (int64_t, int64_t);
20205 int64_t __builtin_vis_fpsub64 (int64_t, int64_t);
20207 v4hi __builtin_vis_fpadds16 (v4hi, v4hi);
20208 v2hi __builtin_vis_fpadds16s (v2hi, v2hi);
20209 v4hi __builtin_vis_fpsubs16 (v4hi, v4hi);
20210 v2hi __builtin_vis_fpsubs16s (v2hi, v2hi);
20211 v2si __builtin_vis_fpadds32 (v2si, v2si);
20212 v1si __builtin_vis_fpadds32s (v1si, v1si);
20213 v2si __builtin_vis_fpsubs32 (v2si, v2si);
20214 v1si __builtin_vis_fpsubs32s (v1si, v1si);
20216 long __builtin_vis_fucmple8 (v8qi, v8qi);
20217 long __builtin_vis_fucmpne8 (v8qi, v8qi);
20218 long __builtin_vis_fucmpgt8 (v8qi, v8qi);
20219 long __builtin_vis_fucmpeq8 (v8qi, v8qi);
20221 float __builtin_vis_fhadds (float, float);
20222 double __builtin_vis_fhaddd (double, double);
20223 float __builtin_vis_fhsubs (float, float);
20224 double __builtin_vis_fhsubd (double, double);
20225 float __builtin_vis_fnhadds (float, float);
20226 double __builtin_vis_fnhaddd (double, double);
20228 int64_t __builtin_vis_umulxhi (int64_t, int64_t);
20229 int64_t __builtin_vis_xmulx (int64_t, int64_t);
20230 int64_t __builtin_vis_xmulxhi (int64_t, int64_t);
20233 When you use the @option{-mvis4} switch, the VIS version 4.0 built-in
20234 functions also become available:
20237 v8qi __builtin_vis_fpadd8 (v8qi, v8qi);
20238 v8qi __builtin_vis_fpadds8 (v8qi, v8qi);
20239 v8qi __builtin_vis_fpaddus8 (v8qi, v8qi);
20240 v4hi __builtin_vis_fpaddus16 (v4hi, v4hi);
20242 v8qi __builtin_vis_fpsub8 (v8qi, v8qi);
20243 v8qi __builtin_vis_fpsubs8 (v8qi, v8qi);
20244 v8qi __builtin_vis_fpsubus8 (v8qi, v8qi);
20245 v4hi __builtin_vis_fpsubus16 (v4hi, v4hi);
20247 long __builtin_vis_fpcmple8 (v8qi, v8qi);
20248 long __builtin_vis_fpcmpgt8 (v8qi, v8qi);
20249 long __builtin_vis_fpcmpule16 (v4hi, v4hi);
20250 long __builtin_vis_fpcmpugt16 (v4hi, v4hi);
20251 long __builtin_vis_fpcmpule32 (v2si, v2si);
20252 long __builtin_vis_fpcmpugt32 (v2si, v2si);
20254 v8qi __builtin_vis_fpmax8 (v8qi, v8qi);
20255 v4hi __builtin_vis_fpmax16 (v4hi, v4hi);
20256 v2si __builtin_vis_fpmax32 (v2si, v2si);
20258 v8qi __builtin_vis_fpmaxu8 (v8qi, v8qi);
20259 v4hi __builtin_vis_fpmaxu16 (v4hi, v4hi);
20260 v2si __builtin_vis_fpmaxu32 (v2si, v2si);
20263 v8qi __builtin_vis_fpmin8 (v8qi, v8qi);
20264 v4hi __builtin_vis_fpmin16 (v4hi, v4hi);
20265 v2si __builtin_vis_fpmin32 (v2si, v2si);
20267 v8qi __builtin_vis_fpminu8 (v8qi, v8qi);
20268 v4hi __builtin_vis_fpminu16 (v4hi, v4hi);
20269 v2si __builtin_vis_fpminu32 (v2si, v2si);
20272 When you use the @option{-mvis4b} switch, the VIS version 4.0B
20273 built-in functions also become available:
20276 v8qi __builtin_vis_dictunpack8 (double, int);
20277 v4hi __builtin_vis_dictunpack16 (double, int);
20278 v2si __builtin_vis_dictunpack32 (double, int);
20280 long __builtin_vis_fpcmple8shl (v8qi, v8qi, int);
20281 long __builtin_vis_fpcmpgt8shl (v8qi, v8qi, int);
20282 long __builtin_vis_fpcmpeq8shl (v8qi, v8qi, int);
20283 long __builtin_vis_fpcmpne8shl (v8qi, v8qi, int);
20285 long __builtin_vis_fpcmple16shl (v4hi, v4hi, int);
20286 long __builtin_vis_fpcmpgt16shl (v4hi, v4hi, int);
20287 long __builtin_vis_fpcmpeq16shl (v4hi, v4hi, int);
20288 long __builtin_vis_fpcmpne16shl (v4hi, v4hi, int);
20290 long __builtin_vis_fpcmple32shl (v2si, v2si, int);
20291 long __builtin_vis_fpcmpgt32shl (v2si, v2si, int);
20292 long __builtin_vis_fpcmpeq32shl (v2si, v2si, int);
20293 long __builtin_vis_fpcmpne32shl (v2si, v2si, int);
20295 long __builtin_vis_fpcmpule8shl (v8qi, v8qi, int);
20296 long __builtin_vis_fpcmpugt8shl (v8qi, v8qi, int);
20297 long __builtin_vis_fpcmpule16shl (v4hi, v4hi, int);
20298 long __builtin_vis_fpcmpugt16shl (v4hi, v4hi, int);
20299 long __builtin_vis_fpcmpule32shl (v2si, v2si, int);
20300 long __builtin_vis_fpcmpugt32shl (v2si, v2si, int);
20302 long __builtin_vis_fpcmpde8shl (v8qi, v8qi, int);
20303 long __builtin_vis_fpcmpde16shl (v4hi, v4hi, int);
20304 long __builtin_vis_fpcmpde32shl (v2si, v2si, int);
20306 long __builtin_vis_fpcmpur8shl (v8qi, v8qi, int);
20307 long __builtin_vis_fpcmpur16shl (v4hi, v4hi, int);
20308 long __builtin_vis_fpcmpur32shl (v2si, v2si, int);
20311 @node SPU Built-in Functions
20312 @subsection SPU Built-in Functions
20314 GCC provides extensions for the SPU processor as described in the
20315 Sony/Toshiba/IBM SPU Language Extensions Specification. GCC's
20316 implementation differs in several ways.
20321 The optional extension of specifying vector constants in parentheses is
20325 A vector initializer requires no cast if the vector constant is of the
20326 same type as the variable it is initializing.
20329 If @code{signed} or @code{unsigned} is omitted, the signedness of the
20330 vector type is the default signedness of the base type. The default
20331 varies depending on the operating system, so a portable program should
20332 always specify the signedness.
20335 By default, the keyword @code{__vector} is added. The macro
20336 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
20340 GCC allows using a @code{typedef} name as the type specifier for a
20344 For C, overloaded functions are implemented with macros so the following
20348 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
20352 Since @code{spu_add} is a macro, the vector constant in the example
20353 is treated as four separate arguments. Wrap the entire argument in
20354 parentheses for this to work.
20357 The extended version of @code{__builtin_expect} is not supported.
20361 @emph{Note:} Only the interface described in the aforementioned
20362 specification is supported. Internally, GCC uses built-in functions to
20363 implement the required functionality, but these are not supported and
20364 are subject to change without notice.
20366 @node TI C6X Built-in Functions
20367 @subsection TI C6X Built-in Functions
20369 GCC provides intrinsics to access certain instructions of the TI C6X
20370 processors. These intrinsics, listed below, are available after
20371 inclusion of the @code{c6x_intrinsics.h} header file. They map directly
20372 to C6X instructions.
20376 int _sadd (int, int)
20377 int _ssub (int, int)
20378 int _sadd2 (int, int)
20379 int _ssub2 (int, int)
20380 long long _mpy2 (int, int)
20381 long long _smpy2 (int, int)
20382 int _add4 (int, int)
20383 int _sub4 (int, int)
20384 int _saddu4 (int, int)
20386 int _smpy (int, int)
20387 int _smpyh (int, int)
20388 int _smpyhl (int, int)
20389 int _smpylh (int, int)
20391 int _sshl (int, int)
20392 int _subc (int, int)
20394 int _avg2 (int, int)
20395 int _avgu4 (int, int)
20397 int _clrr (int, int)
20398 int _extr (int, int)
20399 int _extru (int, int)
20405 @node TILE-Gx Built-in Functions
20406 @subsection TILE-Gx Built-in Functions
20408 GCC provides intrinsics to access every instruction of the TILE-Gx
20409 processor. The intrinsics are of the form:
20413 unsigned long long __insn_@var{op} (...)
20417 Where @var{op} is the name of the instruction. Refer to the ISA manual
20418 for the complete list of instructions.
20420 GCC also provides intrinsics to directly access the network registers.
20421 The intrinsics are:
20425 unsigned long long __tile_idn0_receive (void)
20426 unsigned long long __tile_idn1_receive (void)
20427 unsigned long long __tile_udn0_receive (void)
20428 unsigned long long __tile_udn1_receive (void)
20429 unsigned long long __tile_udn2_receive (void)
20430 unsigned long long __tile_udn3_receive (void)
20431 void __tile_idn_send (unsigned long long)
20432 void __tile_udn_send (unsigned long long)
20436 The intrinsic @code{void __tile_network_barrier (void)} is used to
20437 guarantee that no network operations before it are reordered with
20440 @node TILEPro Built-in Functions
20441 @subsection TILEPro Built-in Functions
20443 GCC provides intrinsics to access every instruction of the TILEPro
20444 processor. The intrinsics are of the form:
20448 unsigned __insn_@var{op} (...)
20453 where @var{op} is the name of the instruction. Refer to the ISA manual
20454 for the complete list of instructions.
20456 GCC also provides intrinsics to directly access the network registers.
20457 The intrinsics are:
20461 unsigned __tile_idn0_receive (void)
20462 unsigned __tile_idn1_receive (void)
20463 unsigned __tile_sn_receive (void)
20464 unsigned __tile_udn0_receive (void)
20465 unsigned __tile_udn1_receive (void)
20466 unsigned __tile_udn2_receive (void)
20467 unsigned __tile_udn3_receive (void)
20468 void __tile_idn_send (unsigned)
20469 void __tile_sn_send (unsigned)
20470 void __tile_udn_send (unsigned)
20474 The intrinsic @code{void __tile_network_barrier (void)} is used to
20475 guarantee that no network operations before it are reordered with
20478 @node x86 Built-in Functions
20479 @subsection x86 Built-in Functions
20481 These built-in functions are available for the x86-32 and x86-64 family
20482 of computers, depending on the command-line switches used.
20484 If you specify command-line switches such as @option{-msse},
20485 the compiler could use the extended instruction sets even if the built-ins
20486 are not used explicitly in the program. For this reason, applications
20487 that perform run-time CPU detection must compile separate files for each
20488 supported architecture, using the appropriate flags. In particular,
20489 the file containing the CPU detection code should be compiled without
20492 The following machine modes are available for use with MMX built-in functions
20493 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
20494 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
20495 vector of eight 8-bit integers. Some of the built-in functions operate on
20496 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
20498 If 3DNow!@: extensions are enabled, @code{V2SF} is used as a mode for a vector
20499 of two 32-bit floating-point values.
20501 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
20502 floating-point values. Some instructions use a vector of four 32-bit
20503 integers, these use @code{V4SI}. Finally, some instructions operate on an
20504 entire vector register, interpreting it as a 128-bit integer, these use mode
20507 The x86-32 and x86-64 family of processors use additional built-in
20508 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
20509 floating point and @code{TC} 128-bit complex floating-point values.
20511 The following floating-point built-in functions are always available. All
20512 of them implement the function that is part of the name.
20515 __float128 __builtin_fabsq (__float128)
20516 __float128 __builtin_copysignq (__float128, __float128)
20519 The following built-in functions are always available.
20522 @item __float128 __builtin_infq (void)
20523 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
20524 @findex __builtin_infq
20526 @item __float128 __builtin_huge_valq (void)
20527 Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}.
20528 @findex __builtin_huge_valq
20530 @item __float128 __builtin_nanq (void)
20531 Similar to @code{__builtin_nan}, except the return type is @code{__float128}.
20532 @findex __builtin_nanq
20534 @item __float128 __builtin_nansq (void)
20535 Similar to @code{__builtin_nans}, except the return type is @code{__float128}.
20536 @findex __builtin_nansq
20539 The following built-in function is always available.
20542 @item void __builtin_ia32_pause (void)
20543 Generates the @code{pause} machine instruction with a compiler memory
20547 The following built-in functions are always available and can be used to
20548 check the target platform type.
20550 @deftypefn {Built-in Function} void __builtin_cpu_init (void)
20551 This function runs the CPU detection code to check the type of CPU and the
20552 features supported. This built-in function needs to be invoked along with the built-in functions
20553 to check CPU type and features, @code{__builtin_cpu_is} and
20554 @code{__builtin_cpu_supports}, only when used in a function that is
20555 executed before any constructors are called. The CPU detection code is
20556 automatically executed in a very high priority constructor.
20558 For example, this function has to be used in @code{ifunc} resolvers that
20559 check for CPU type using the built-in functions @code{__builtin_cpu_is}
20560 and @code{__builtin_cpu_supports}, or in constructors on targets that
20561 don't support constructor priority.
20564 static void (*resolve_memcpy (void)) (void)
20566 // ifunc resolvers fire before constructors, explicitly call the init
20568 __builtin_cpu_init ();
20569 if (__builtin_cpu_supports ("ssse3"))
20570 return ssse3_memcpy; // super fast memcpy with ssse3 instructions.
20572 return default_memcpy;
20575 void *memcpy (void *, const void *, size_t)
20576 __attribute__ ((ifunc ("resolve_memcpy")));
20581 @deftypefn {Built-in Function} int __builtin_cpu_is (const char *@var{cpuname})
20582 This function returns a positive integer if the run-time CPU
20583 is of type @var{cpuname}
20584 and returns @code{0} otherwise. The following CPU names can be detected:
20600 Intel Core i7 Nehalem CPU.
20603 Intel Core i7 Westmere CPU.
20606 Intel Core i7 Sandy Bridge CPU.
20612 AMD Family 10h CPU.
20615 AMD Family 10h Barcelona CPU.
20618 AMD Family 10h Shanghai CPU.
20621 AMD Family 10h Istanbul CPU.
20624 AMD Family 14h CPU.
20627 AMD Family 15h CPU.
20630 AMD Family 15h Bulldozer version 1.
20633 AMD Family 15h Bulldozer version 2.
20636 AMD Family 15h Bulldozer version 3.
20639 AMD Family 15h Bulldozer version 4.
20642 AMD Family 16h CPU.
20645 AMD Family 17h CPU.
20648 AMD Family 17h Zen version 1.
20651 Here is an example:
20653 if (__builtin_cpu_is ("corei7"))
20655 do_corei7 (); // Core i7 specific implementation.
20659 do_generic (); // Generic implementation.
20664 @deftypefn {Built-in Function} int __builtin_cpu_supports (const char *@var{feature})
20665 This function returns a positive integer if the run-time CPU
20666 supports @var{feature}
20667 and returns @code{0} otherwise. The following features can be detected:
20675 POPCNT instruction.
20683 SSSE3 instructions.
20685 SSE4.1 instructions.
20687 SSE4.2 instructions.
20693 AVX512F instructions.
20696 Here is an example:
20698 if (__builtin_cpu_supports ("popcnt"))
20700 asm("popcnt %1,%0" : "=r"(count) : "rm"(n) : "cc");
20704 count = generic_countbits (n); //generic implementation.
20710 The following built-in functions are made available by @option{-mmmx}.
20711 All of them generate the machine instruction that is part of the name.
20714 v8qi __builtin_ia32_paddb (v8qi, v8qi)
20715 v4hi __builtin_ia32_paddw (v4hi, v4hi)
20716 v2si __builtin_ia32_paddd (v2si, v2si)
20717 v8qi __builtin_ia32_psubb (v8qi, v8qi)
20718 v4hi __builtin_ia32_psubw (v4hi, v4hi)
20719 v2si __builtin_ia32_psubd (v2si, v2si)
20720 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
20721 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
20722 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
20723 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
20724 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
20725 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
20726 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
20727 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
20728 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
20729 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
20730 di __builtin_ia32_pand (di, di)
20731 di __builtin_ia32_pandn (di,di)
20732 di __builtin_ia32_por (di, di)
20733 di __builtin_ia32_pxor (di, di)
20734 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
20735 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
20736 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
20737 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
20738 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
20739 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
20740 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
20741 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
20742 v2si __builtin_ia32_punpckhdq (v2si, v2si)
20743 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
20744 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
20745 v2si __builtin_ia32_punpckldq (v2si, v2si)
20746 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
20747 v4hi __builtin_ia32_packssdw (v2si, v2si)
20748 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
20750 v4hi __builtin_ia32_psllw (v4hi, v4hi)
20751 v2si __builtin_ia32_pslld (v2si, v2si)
20752 v1di __builtin_ia32_psllq (v1di, v1di)
20753 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
20754 v2si __builtin_ia32_psrld (v2si, v2si)
20755 v1di __builtin_ia32_psrlq (v1di, v1di)
20756 v4hi __builtin_ia32_psraw (v4hi, v4hi)
20757 v2si __builtin_ia32_psrad (v2si, v2si)
20758 v4hi __builtin_ia32_psllwi (v4hi, int)
20759 v2si __builtin_ia32_pslldi (v2si, int)
20760 v1di __builtin_ia32_psllqi (v1di, int)
20761 v4hi __builtin_ia32_psrlwi (v4hi, int)
20762 v2si __builtin_ia32_psrldi (v2si, int)
20763 v1di __builtin_ia32_psrlqi (v1di, int)
20764 v4hi __builtin_ia32_psrawi (v4hi, int)
20765 v2si __builtin_ia32_psradi (v2si, int)
20769 The following built-in functions are made available either with
20770 @option{-msse}, or with @option{-m3dnowa}. All of them generate
20771 the machine instruction that is part of the name.
20774 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
20775 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
20776 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
20777 v1di __builtin_ia32_psadbw (v8qi, v8qi)
20778 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
20779 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
20780 v8qi __builtin_ia32_pminub (v8qi, v8qi)
20781 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
20782 int __builtin_ia32_pmovmskb (v8qi)
20783 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
20784 void __builtin_ia32_movntq (di *, di)
20785 void __builtin_ia32_sfence (void)
20788 The following built-in functions are available when @option{-msse} is used.
20789 All of them generate the machine instruction that is part of the name.
20792 int __builtin_ia32_comieq (v4sf, v4sf)
20793 int __builtin_ia32_comineq (v4sf, v4sf)
20794 int __builtin_ia32_comilt (v4sf, v4sf)
20795 int __builtin_ia32_comile (v4sf, v4sf)
20796 int __builtin_ia32_comigt (v4sf, v4sf)
20797 int __builtin_ia32_comige (v4sf, v4sf)
20798 int __builtin_ia32_ucomieq (v4sf, v4sf)
20799 int __builtin_ia32_ucomineq (v4sf, v4sf)
20800 int __builtin_ia32_ucomilt (v4sf, v4sf)
20801 int __builtin_ia32_ucomile (v4sf, v4sf)
20802 int __builtin_ia32_ucomigt (v4sf, v4sf)
20803 int __builtin_ia32_ucomige (v4sf, v4sf)
20804 v4sf __builtin_ia32_addps (v4sf, v4sf)
20805 v4sf __builtin_ia32_subps (v4sf, v4sf)
20806 v4sf __builtin_ia32_mulps (v4sf, v4sf)
20807 v4sf __builtin_ia32_divps (v4sf, v4sf)
20808 v4sf __builtin_ia32_addss (v4sf, v4sf)
20809 v4sf __builtin_ia32_subss (v4sf, v4sf)
20810 v4sf __builtin_ia32_mulss (v4sf, v4sf)
20811 v4sf __builtin_ia32_divss (v4sf, v4sf)
20812 v4sf __builtin_ia32_cmpeqps (v4sf, v4sf)
20813 v4sf __builtin_ia32_cmpltps (v4sf, v4sf)
20814 v4sf __builtin_ia32_cmpleps (v4sf, v4sf)
20815 v4sf __builtin_ia32_cmpgtps (v4sf, v4sf)
20816 v4sf __builtin_ia32_cmpgeps (v4sf, v4sf)
20817 v4sf __builtin_ia32_cmpunordps (v4sf, v4sf)
20818 v4sf __builtin_ia32_cmpneqps (v4sf, v4sf)
20819 v4sf __builtin_ia32_cmpnltps (v4sf, v4sf)
20820 v4sf __builtin_ia32_cmpnleps (v4sf, v4sf)
20821 v4sf __builtin_ia32_cmpngtps (v4sf, v4sf)
20822 v4sf __builtin_ia32_cmpngeps (v4sf, v4sf)
20823 v4sf __builtin_ia32_cmpordps (v4sf, v4sf)
20824 v4sf __builtin_ia32_cmpeqss (v4sf, v4sf)
20825 v4sf __builtin_ia32_cmpltss (v4sf, v4sf)
20826 v4sf __builtin_ia32_cmpless (v4sf, v4sf)
20827 v4sf __builtin_ia32_cmpunordss (v4sf, v4sf)
20828 v4sf __builtin_ia32_cmpneqss (v4sf, v4sf)
20829 v4sf __builtin_ia32_cmpnltss (v4sf, v4sf)
20830 v4sf __builtin_ia32_cmpnless (v4sf, v4sf)
20831 v4sf __builtin_ia32_cmpordss (v4sf, v4sf)
20832 v4sf __builtin_ia32_maxps (v4sf, v4sf)
20833 v4sf __builtin_ia32_maxss (v4sf, v4sf)
20834 v4sf __builtin_ia32_minps (v4sf, v4sf)
20835 v4sf __builtin_ia32_minss (v4sf, v4sf)
20836 v4sf __builtin_ia32_andps (v4sf, v4sf)
20837 v4sf __builtin_ia32_andnps (v4sf, v4sf)
20838 v4sf __builtin_ia32_orps (v4sf, v4sf)
20839 v4sf __builtin_ia32_xorps (v4sf, v4sf)
20840 v4sf __builtin_ia32_movss (v4sf, v4sf)
20841 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
20842 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
20843 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
20844 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
20845 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
20846 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
20847 v2si __builtin_ia32_cvtps2pi (v4sf)
20848 int __builtin_ia32_cvtss2si (v4sf)
20849 v2si __builtin_ia32_cvttps2pi (v4sf)
20850 int __builtin_ia32_cvttss2si (v4sf)
20851 v4sf __builtin_ia32_rcpps (v4sf)
20852 v4sf __builtin_ia32_rsqrtps (v4sf)
20853 v4sf __builtin_ia32_sqrtps (v4sf)
20854 v4sf __builtin_ia32_rcpss (v4sf)
20855 v4sf __builtin_ia32_rsqrtss (v4sf)
20856 v4sf __builtin_ia32_sqrtss (v4sf)
20857 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
20858 void __builtin_ia32_movntps (float *, v4sf)
20859 int __builtin_ia32_movmskps (v4sf)
20862 The following built-in functions are available when @option{-msse} is used.
20865 @item v4sf __builtin_ia32_loadups (float *)
20866 Generates the @code{movups} machine instruction as a load from memory.
20867 @item void __builtin_ia32_storeups (float *, v4sf)
20868 Generates the @code{movups} machine instruction as a store to memory.
20869 @item v4sf __builtin_ia32_loadss (float *)
20870 Generates the @code{movss} machine instruction as a load from memory.
20871 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
20872 Generates the @code{movhps} machine instruction as a load from memory.
20873 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
20874 Generates the @code{movlps} machine instruction as a load from memory
20875 @item void __builtin_ia32_storehps (v2sf *, v4sf)
20876 Generates the @code{movhps} machine instruction as a store to memory.
20877 @item void __builtin_ia32_storelps (v2sf *, v4sf)
20878 Generates the @code{movlps} machine instruction as a store to memory.
20881 The following built-in functions are available when @option{-msse2} is used.
20882 All of them generate the machine instruction that is part of the name.
20885 int __builtin_ia32_comisdeq (v2df, v2df)
20886 int __builtin_ia32_comisdlt (v2df, v2df)
20887 int __builtin_ia32_comisdle (v2df, v2df)
20888 int __builtin_ia32_comisdgt (v2df, v2df)
20889 int __builtin_ia32_comisdge (v2df, v2df)
20890 int __builtin_ia32_comisdneq (v2df, v2df)
20891 int __builtin_ia32_ucomisdeq (v2df, v2df)
20892 int __builtin_ia32_ucomisdlt (v2df, v2df)
20893 int __builtin_ia32_ucomisdle (v2df, v2df)
20894 int __builtin_ia32_ucomisdgt (v2df, v2df)
20895 int __builtin_ia32_ucomisdge (v2df, v2df)
20896 int __builtin_ia32_ucomisdneq (v2df, v2df)
20897 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
20898 v2df __builtin_ia32_cmpltpd (v2df, v2df)
20899 v2df __builtin_ia32_cmplepd (v2df, v2df)
20900 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
20901 v2df __builtin_ia32_cmpgepd (v2df, v2df)
20902 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
20903 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
20904 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
20905 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
20906 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
20907 v2df __builtin_ia32_cmpngepd (v2df, v2df)
20908 v2df __builtin_ia32_cmpordpd (v2df, v2df)
20909 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
20910 v2df __builtin_ia32_cmpltsd (v2df, v2df)
20911 v2df __builtin_ia32_cmplesd (v2df, v2df)
20912 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
20913 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
20914 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
20915 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
20916 v2df __builtin_ia32_cmpordsd (v2df, v2df)
20917 v2di __builtin_ia32_paddq (v2di, v2di)
20918 v2di __builtin_ia32_psubq (v2di, v2di)
20919 v2df __builtin_ia32_addpd (v2df, v2df)
20920 v2df __builtin_ia32_subpd (v2df, v2df)
20921 v2df __builtin_ia32_mulpd (v2df, v2df)
20922 v2df __builtin_ia32_divpd (v2df, v2df)
20923 v2df __builtin_ia32_addsd (v2df, v2df)
20924 v2df __builtin_ia32_subsd (v2df, v2df)
20925 v2df __builtin_ia32_mulsd (v2df, v2df)
20926 v2df __builtin_ia32_divsd (v2df, v2df)
20927 v2df __builtin_ia32_minpd (v2df, v2df)
20928 v2df __builtin_ia32_maxpd (v2df, v2df)
20929 v2df __builtin_ia32_minsd (v2df, v2df)
20930 v2df __builtin_ia32_maxsd (v2df, v2df)
20931 v2df __builtin_ia32_andpd (v2df, v2df)
20932 v2df __builtin_ia32_andnpd (v2df, v2df)
20933 v2df __builtin_ia32_orpd (v2df, v2df)
20934 v2df __builtin_ia32_xorpd (v2df, v2df)
20935 v2df __builtin_ia32_movsd (v2df, v2df)
20936 v2df __builtin_ia32_unpckhpd (v2df, v2df)
20937 v2df __builtin_ia32_unpcklpd (v2df, v2df)
20938 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
20939 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
20940 v4si __builtin_ia32_paddd128 (v4si, v4si)
20941 v2di __builtin_ia32_paddq128 (v2di, v2di)
20942 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
20943 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
20944 v4si __builtin_ia32_psubd128 (v4si, v4si)
20945 v2di __builtin_ia32_psubq128 (v2di, v2di)
20946 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
20947 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
20948 v2di __builtin_ia32_pand128 (v2di, v2di)
20949 v2di __builtin_ia32_pandn128 (v2di, v2di)
20950 v2di __builtin_ia32_por128 (v2di, v2di)
20951 v2di __builtin_ia32_pxor128 (v2di, v2di)
20952 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
20953 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
20954 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
20955 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
20956 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
20957 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
20958 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
20959 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
20960 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
20961 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
20962 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
20963 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
20964 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
20965 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
20966 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
20967 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
20968 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
20969 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
20970 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
20971 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
20972 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
20973 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
20974 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
20975 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
20976 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
20977 v2df __builtin_ia32_loadupd (double *)
20978 void __builtin_ia32_storeupd (double *, v2df)
20979 v2df __builtin_ia32_loadhpd (v2df, double const *)
20980 v2df __builtin_ia32_loadlpd (v2df, double const *)
20981 int __builtin_ia32_movmskpd (v2df)
20982 int __builtin_ia32_pmovmskb128 (v16qi)
20983 void __builtin_ia32_movnti (int *, int)
20984 void __builtin_ia32_movnti64 (long long int *, long long int)
20985 void __builtin_ia32_movntpd (double *, v2df)
20986 void __builtin_ia32_movntdq (v2df *, v2df)
20987 v4si __builtin_ia32_pshufd (v4si, int)
20988 v8hi __builtin_ia32_pshuflw (v8hi, int)
20989 v8hi __builtin_ia32_pshufhw (v8hi, int)
20990 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
20991 v2df __builtin_ia32_sqrtpd (v2df)
20992 v2df __builtin_ia32_sqrtsd (v2df)
20993 v2df __builtin_ia32_shufpd (v2df, v2df, int)
20994 v2df __builtin_ia32_cvtdq2pd (v4si)
20995 v4sf __builtin_ia32_cvtdq2ps (v4si)
20996 v4si __builtin_ia32_cvtpd2dq (v2df)
20997 v2si __builtin_ia32_cvtpd2pi (v2df)
20998 v4sf __builtin_ia32_cvtpd2ps (v2df)
20999 v4si __builtin_ia32_cvttpd2dq (v2df)
21000 v2si __builtin_ia32_cvttpd2pi (v2df)
21001 v2df __builtin_ia32_cvtpi2pd (v2si)
21002 int __builtin_ia32_cvtsd2si (v2df)
21003 int __builtin_ia32_cvttsd2si (v2df)
21004 long long __builtin_ia32_cvtsd2si64 (v2df)
21005 long long __builtin_ia32_cvttsd2si64 (v2df)
21006 v4si __builtin_ia32_cvtps2dq (v4sf)
21007 v2df __builtin_ia32_cvtps2pd (v4sf)
21008 v4si __builtin_ia32_cvttps2dq (v4sf)
21009 v2df __builtin_ia32_cvtsi2sd (v2df, int)
21010 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
21011 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
21012 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
21013 void __builtin_ia32_clflush (const void *)
21014 void __builtin_ia32_lfence (void)
21015 void __builtin_ia32_mfence (void)
21016 v16qi __builtin_ia32_loaddqu (const char *)
21017 void __builtin_ia32_storedqu (char *, v16qi)
21018 v1di __builtin_ia32_pmuludq (v2si, v2si)
21019 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
21020 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
21021 v4si __builtin_ia32_pslld128 (v4si, v4si)
21022 v2di __builtin_ia32_psllq128 (v2di, v2di)
21023 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
21024 v4si __builtin_ia32_psrld128 (v4si, v4si)
21025 v2di __builtin_ia32_psrlq128 (v2di, v2di)
21026 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
21027 v4si __builtin_ia32_psrad128 (v4si, v4si)
21028 v2di __builtin_ia32_pslldqi128 (v2di, int)
21029 v8hi __builtin_ia32_psllwi128 (v8hi, int)
21030 v4si __builtin_ia32_pslldi128 (v4si, int)
21031 v2di __builtin_ia32_psllqi128 (v2di, int)
21032 v2di __builtin_ia32_psrldqi128 (v2di, int)
21033 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
21034 v4si __builtin_ia32_psrldi128 (v4si, int)
21035 v2di __builtin_ia32_psrlqi128 (v2di, int)
21036 v8hi __builtin_ia32_psrawi128 (v8hi, int)
21037 v4si __builtin_ia32_psradi128 (v4si, int)
21038 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
21039 v2di __builtin_ia32_movq128 (v2di)
21042 The following built-in functions are available when @option{-msse3} is used.
21043 All of them generate the machine instruction that is part of the name.
21046 v2df __builtin_ia32_addsubpd (v2df, v2df)
21047 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
21048 v2df __builtin_ia32_haddpd (v2df, v2df)
21049 v4sf __builtin_ia32_haddps (v4sf, v4sf)
21050 v2df __builtin_ia32_hsubpd (v2df, v2df)
21051 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
21052 v16qi __builtin_ia32_lddqu (char const *)
21053 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
21054 v4sf __builtin_ia32_movshdup (v4sf)
21055 v4sf __builtin_ia32_movsldup (v4sf)
21056 void __builtin_ia32_mwait (unsigned int, unsigned int)
21059 The following built-in functions are available when @option{-mssse3} is used.
21060 All of them generate the machine instruction that is part of the name.
21063 v2si __builtin_ia32_phaddd (v2si, v2si)
21064 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
21065 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
21066 v2si __builtin_ia32_phsubd (v2si, v2si)
21067 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
21068 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
21069 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
21070 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
21071 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
21072 v8qi __builtin_ia32_psignb (v8qi, v8qi)
21073 v2si __builtin_ia32_psignd (v2si, v2si)
21074 v4hi __builtin_ia32_psignw (v4hi, v4hi)
21075 v1di __builtin_ia32_palignr (v1di, v1di, int)
21076 v8qi __builtin_ia32_pabsb (v8qi)
21077 v2si __builtin_ia32_pabsd (v2si)
21078 v4hi __builtin_ia32_pabsw (v4hi)
21081 The following built-in functions are available when @option{-mssse3} is used.
21082 All of them generate the machine instruction that is part of the name.
21085 v4si __builtin_ia32_phaddd128 (v4si, v4si)
21086 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
21087 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
21088 v4si __builtin_ia32_phsubd128 (v4si, v4si)
21089 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
21090 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
21091 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
21092 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
21093 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
21094 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
21095 v4si __builtin_ia32_psignd128 (v4si, v4si)
21096 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
21097 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
21098 v16qi __builtin_ia32_pabsb128 (v16qi)
21099 v4si __builtin_ia32_pabsd128 (v4si)
21100 v8hi __builtin_ia32_pabsw128 (v8hi)
21103 The following built-in functions are available when @option{-msse4.1} is
21104 used. All of them generate the machine instruction that is part of the
21108 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
21109 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
21110 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
21111 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
21112 v2df __builtin_ia32_dppd (v2df, v2df, const int)
21113 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
21114 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
21115 v2di __builtin_ia32_movntdqa (v2di *);
21116 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
21117 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
21118 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
21119 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
21120 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
21121 v8hi __builtin_ia32_phminposuw128 (v8hi)
21122 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
21123 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
21124 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
21125 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
21126 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
21127 v4si __builtin_ia32_pminsd128 (v4si, v4si)
21128 v4si __builtin_ia32_pminud128 (v4si, v4si)
21129 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
21130 v4si __builtin_ia32_pmovsxbd128 (v16qi)
21131 v2di __builtin_ia32_pmovsxbq128 (v16qi)
21132 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
21133 v2di __builtin_ia32_pmovsxdq128 (v4si)
21134 v4si __builtin_ia32_pmovsxwd128 (v8hi)
21135 v2di __builtin_ia32_pmovsxwq128 (v8hi)
21136 v4si __builtin_ia32_pmovzxbd128 (v16qi)
21137 v2di __builtin_ia32_pmovzxbq128 (v16qi)
21138 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
21139 v2di __builtin_ia32_pmovzxdq128 (v4si)
21140 v4si __builtin_ia32_pmovzxwd128 (v8hi)
21141 v2di __builtin_ia32_pmovzxwq128 (v8hi)
21142 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
21143 v4si __builtin_ia32_pmulld128 (v4si, v4si)
21144 int __builtin_ia32_ptestc128 (v2di, v2di)
21145 int __builtin_ia32_ptestnzc128 (v2di, v2di)
21146 int __builtin_ia32_ptestz128 (v2di, v2di)
21147 v2df __builtin_ia32_roundpd (v2df, const int)
21148 v4sf __builtin_ia32_roundps (v4sf, const int)
21149 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
21150 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
21153 The following built-in functions are available when @option{-msse4.1} is
21157 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
21158 Generates the @code{insertps} machine instruction.
21159 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
21160 Generates the @code{pextrb} machine instruction.
21161 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
21162 Generates the @code{pinsrb} machine instruction.
21163 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
21164 Generates the @code{pinsrd} machine instruction.
21165 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
21166 Generates the @code{pinsrq} machine instruction in 64bit mode.
21169 The following built-in functions are changed to generate new SSE4.1
21170 instructions when @option{-msse4.1} is used.
21173 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
21174 Generates the @code{extractps} machine instruction.
21175 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
21176 Generates the @code{pextrd} machine instruction.
21177 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
21178 Generates the @code{pextrq} machine instruction in 64bit mode.
21181 The following built-in functions are available when @option{-msse4.2} is
21182 used. All of them generate the machine instruction that is part of the
21186 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
21187 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
21188 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
21189 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
21190 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
21191 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
21192 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
21193 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
21194 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
21195 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
21196 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
21197 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
21198 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
21199 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
21200 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
21203 The following built-in functions are available when @option{-msse4.2} is
21207 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
21208 Generates the @code{crc32b} machine instruction.
21209 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
21210 Generates the @code{crc32w} machine instruction.
21211 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
21212 Generates the @code{crc32l} machine instruction.
21213 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
21214 Generates the @code{crc32q} machine instruction.
21217 The following built-in functions are changed to generate new SSE4.2
21218 instructions when @option{-msse4.2} is used.
21221 @item int __builtin_popcount (unsigned int)
21222 Generates the @code{popcntl} machine instruction.
21223 @item int __builtin_popcountl (unsigned long)
21224 Generates the @code{popcntl} or @code{popcntq} machine instruction,
21225 depending on the size of @code{unsigned long}.
21226 @item int __builtin_popcountll (unsigned long long)
21227 Generates the @code{popcntq} machine instruction.
21230 The following built-in functions are available when @option{-mavx} is
21231 used. All of them generate the machine instruction that is part of the
21235 v4df __builtin_ia32_addpd256 (v4df,v4df)
21236 v8sf __builtin_ia32_addps256 (v8sf,v8sf)
21237 v4df __builtin_ia32_addsubpd256 (v4df,v4df)
21238 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf)
21239 v4df __builtin_ia32_andnpd256 (v4df,v4df)
21240 v8sf __builtin_ia32_andnps256 (v8sf,v8sf)
21241 v4df __builtin_ia32_andpd256 (v4df,v4df)
21242 v8sf __builtin_ia32_andps256 (v8sf,v8sf)
21243 v4df __builtin_ia32_blendpd256 (v4df,v4df,int)
21244 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int)
21245 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df)
21246 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf)
21247 v2df __builtin_ia32_cmppd (v2df,v2df,int)
21248 v4df __builtin_ia32_cmppd256 (v4df,v4df,int)
21249 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int)
21250 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int)
21251 v2df __builtin_ia32_cmpsd (v2df,v2df,int)
21252 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int)
21253 v4df __builtin_ia32_cvtdq2pd256 (v4si)
21254 v8sf __builtin_ia32_cvtdq2ps256 (v8si)
21255 v4si __builtin_ia32_cvtpd2dq256 (v4df)
21256 v4sf __builtin_ia32_cvtpd2ps256 (v4df)
21257 v8si __builtin_ia32_cvtps2dq256 (v8sf)
21258 v4df __builtin_ia32_cvtps2pd256 (v4sf)
21259 v4si __builtin_ia32_cvttpd2dq256 (v4df)
21260 v8si __builtin_ia32_cvttps2dq256 (v8sf)
21261 v4df __builtin_ia32_divpd256 (v4df,v4df)
21262 v8sf __builtin_ia32_divps256 (v8sf,v8sf)
21263 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int)
21264 v4df __builtin_ia32_haddpd256 (v4df,v4df)
21265 v8sf __builtin_ia32_haddps256 (v8sf,v8sf)
21266 v4df __builtin_ia32_hsubpd256 (v4df,v4df)
21267 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf)
21268 v32qi __builtin_ia32_lddqu256 (pcchar)
21269 v32qi __builtin_ia32_loaddqu256 (pcchar)
21270 v4df __builtin_ia32_loadupd256 (pcdouble)
21271 v8sf __builtin_ia32_loadups256 (pcfloat)
21272 v2df __builtin_ia32_maskloadpd (pcv2df,v2df)
21273 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df)
21274 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf)
21275 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf)
21276 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df)
21277 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df)
21278 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf)
21279 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf)
21280 v4df __builtin_ia32_maxpd256 (v4df,v4df)
21281 v8sf __builtin_ia32_maxps256 (v8sf,v8sf)
21282 v4df __builtin_ia32_minpd256 (v4df,v4df)
21283 v8sf __builtin_ia32_minps256 (v8sf,v8sf)
21284 v4df __builtin_ia32_movddup256 (v4df)
21285 int __builtin_ia32_movmskpd256 (v4df)
21286 int __builtin_ia32_movmskps256 (v8sf)
21287 v8sf __builtin_ia32_movshdup256 (v8sf)
21288 v8sf __builtin_ia32_movsldup256 (v8sf)
21289 v4df __builtin_ia32_mulpd256 (v4df,v4df)
21290 v8sf __builtin_ia32_mulps256 (v8sf,v8sf)
21291 v4df __builtin_ia32_orpd256 (v4df,v4df)
21292 v8sf __builtin_ia32_orps256 (v8sf,v8sf)
21293 v2df __builtin_ia32_pd_pd256 (v4df)
21294 v4df __builtin_ia32_pd256_pd (v2df)
21295 v4sf __builtin_ia32_ps_ps256 (v8sf)
21296 v8sf __builtin_ia32_ps256_ps (v4sf)
21297 int __builtin_ia32_ptestc256 (v4di,v4di,ptest)
21298 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest)
21299 int __builtin_ia32_ptestz256 (v4di,v4di,ptest)
21300 v8sf __builtin_ia32_rcpps256 (v8sf)
21301 v4df __builtin_ia32_roundpd256 (v4df,int)
21302 v8sf __builtin_ia32_roundps256 (v8sf,int)
21303 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf)
21304 v8sf __builtin_ia32_rsqrtps256 (v8sf)
21305 v4df __builtin_ia32_shufpd256 (v4df,v4df,int)
21306 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int)
21307 v4si __builtin_ia32_si_si256 (v8si)
21308 v8si __builtin_ia32_si256_si (v4si)
21309 v4df __builtin_ia32_sqrtpd256 (v4df)
21310 v8sf __builtin_ia32_sqrtps_nr256 (v8sf)
21311 v8sf __builtin_ia32_sqrtps256 (v8sf)
21312 void __builtin_ia32_storedqu256 (pchar,v32qi)
21313 void __builtin_ia32_storeupd256 (pdouble,v4df)
21314 void __builtin_ia32_storeups256 (pfloat,v8sf)
21315 v4df __builtin_ia32_subpd256 (v4df,v4df)
21316 v8sf __builtin_ia32_subps256 (v8sf,v8sf)
21317 v4df __builtin_ia32_unpckhpd256 (v4df,v4df)
21318 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf)
21319 v4df __builtin_ia32_unpcklpd256 (v4df,v4df)
21320 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf)
21321 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df)
21322 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf)
21323 v4df __builtin_ia32_vbroadcastsd256 (pcdouble)
21324 v4sf __builtin_ia32_vbroadcastss (pcfloat)
21325 v8sf __builtin_ia32_vbroadcastss256 (pcfloat)
21326 v2df __builtin_ia32_vextractf128_pd256 (v4df,int)
21327 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int)
21328 v4si __builtin_ia32_vextractf128_si256 (v8si,int)
21329 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int)
21330 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int)
21331 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int)
21332 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int)
21333 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int)
21334 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int)
21335 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int)
21336 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int)
21337 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int)
21338 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int)
21339 v2df __builtin_ia32_vpermilpd (v2df,int)
21340 v4df __builtin_ia32_vpermilpd256 (v4df,int)
21341 v4sf __builtin_ia32_vpermilps (v4sf,int)
21342 v8sf __builtin_ia32_vpermilps256 (v8sf,int)
21343 v2df __builtin_ia32_vpermilvarpd (v2df,v2di)
21344 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di)
21345 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si)
21346 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si)
21347 int __builtin_ia32_vtestcpd (v2df,v2df,ptest)
21348 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest)
21349 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest)
21350 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest)
21351 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest)
21352 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest)
21353 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest)
21354 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest)
21355 int __builtin_ia32_vtestzpd (v2df,v2df,ptest)
21356 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest)
21357 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest)
21358 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest)
21359 void __builtin_ia32_vzeroall (void)
21360 void __builtin_ia32_vzeroupper (void)
21361 v4df __builtin_ia32_xorpd256 (v4df,v4df)
21362 v8sf __builtin_ia32_xorps256 (v8sf,v8sf)
21365 The following built-in functions are available when @option{-mavx2} is
21366 used. All of them generate the machine instruction that is part of the
21370 v32qi __builtin_ia32_mpsadbw256 (v32qi,v32qi,int)
21371 v32qi __builtin_ia32_pabsb256 (v32qi)
21372 v16hi __builtin_ia32_pabsw256 (v16hi)
21373 v8si __builtin_ia32_pabsd256 (v8si)
21374 v16hi __builtin_ia32_packssdw256 (v8si,v8si)
21375 v32qi __builtin_ia32_packsswb256 (v16hi,v16hi)
21376 v16hi __builtin_ia32_packusdw256 (v8si,v8si)
21377 v32qi __builtin_ia32_packuswb256 (v16hi,v16hi)
21378 v32qi __builtin_ia32_paddb256 (v32qi,v32qi)
21379 v16hi __builtin_ia32_paddw256 (v16hi,v16hi)
21380 v8si __builtin_ia32_paddd256 (v8si,v8si)
21381 v4di __builtin_ia32_paddq256 (v4di,v4di)
21382 v32qi __builtin_ia32_paddsb256 (v32qi,v32qi)
21383 v16hi __builtin_ia32_paddsw256 (v16hi,v16hi)
21384 v32qi __builtin_ia32_paddusb256 (v32qi,v32qi)
21385 v16hi __builtin_ia32_paddusw256 (v16hi,v16hi)
21386 v4di __builtin_ia32_palignr256 (v4di,v4di,int)
21387 v4di __builtin_ia32_andsi256 (v4di,v4di)
21388 v4di __builtin_ia32_andnotsi256 (v4di,v4di)
21389 v32qi __builtin_ia32_pavgb256 (v32qi,v32qi)
21390 v16hi __builtin_ia32_pavgw256 (v16hi,v16hi)
21391 v32qi __builtin_ia32_pblendvb256 (v32qi,v32qi,v32qi)
21392 v16hi __builtin_ia32_pblendw256 (v16hi,v16hi,int)
21393 v32qi __builtin_ia32_pcmpeqb256 (v32qi,v32qi)
21394 v16hi __builtin_ia32_pcmpeqw256 (v16hi,v16hi)
21395 v8si __builtin_ia32_pcmpeqd256 (c8si,v8si)
21396 v4di __builtin_ia32_pcmpeqq256 (v4di,v4di)
21397 v32qi __builtin_ia32_pcmpgtb256 (v32qi,v32qi)
21398 v16hi __builtin_ia32_pcmpgtw256 (16hi,v16hi)
21399 v8si __builtin_ia32_pcmpgtd256 (v8si,v8si)
21400 v4di __builtin_ia32_pcmpgtq256 (v4di,v4di)
21401 v16hi __builtin_ia32_phaddw256 (v16hi,v16hi)
21402 v8si __builtin_ia32_phaddd256 (v8si,v8si)
21403 v16hi __builtin_ia32_phaddsw256 (v16hi,v16hi)
21404 v16hi __builtin_ia32_phsubw256 (v16hi,v16hi)
21405 v8si __builtin_ia32_phsubd256 (v8si,v8si)
21406 v16hi __builtin_ia32_phsubsw256 (v16hi,v16hi)
21407 v32qi __builtin_ia32_pmaddubsw256 (v32qi,v32qi)
21408 v16hi __builtin_ia32_pmaddwd256 (v16hi,v16hi)
21409 v32qi __builtin_ia32_pmaxsb256 (v32qi,v32qi)
21410 v16hi __builtin_ia32_pmaxsw256 (v16hi,v16hi)
21411 v8si __builtin_ia32_pmaxsd256 (v8si,v8si)
21412 v32qi __builtin_ia32_pmaxub256 (v32qi,v32qi)
21413 v16hi __builtin_ia32_pmaxuw256 (v16hi,v16hi)
21414 v8si __builtin_ia32_pmaxud256 (v8si,v8si)
21415 v32qi __builtin_ia32_pminsb256 (v32qi,v32qi)
21416 v16hi __builtin_ia32_pminsw256 (v16hi,v16hi)
21417 v8si __builtin_ia32_pminsd256 (v8si,v8si)
21418 v32qi __builtin_ia32_pminub256 (v32qi,v32qi)
21419 v16hi __builtin_ia32_pminuw256 (v16hi,v16hi)
21420 v8si __builtin_ia32_pminud256 (v8si,v8si)
21421 int __builtin_ia32_pmovmskb256 (v32qi)
21422 v16hi __builtin_ia32_pmovsxbw256 (v16qi)
21423 v8si __builtin_ia32_pmovsxbd256 (v16qi)
21424 v4di __builtin_ia32_pmovsxbq256 (v16qi)
21425 v8si __builtin_ia32_pmovsxwd256 (v8hi)
21426 v4di __builtin_ia32_pmovsxwq256 (v8hi)
21427 v4di __builtin_ia32_pmovsxdq256 (v4si)
21428 v16hi __builtin_ia32_pmovzxbw256 (v16qi)
21429 v8si __builtin_ia32_pmovzxbd256 (v16qi)
21430 v4di __builtin_ia32_pmovzxbq256 (v16qi)
21431 v8si __builtin_ia32_pmovzxwd256 (v8hi)
21432 v4di __builtin_ia32_pmovzxwq256 (v8hi)
21433 v4di __builtin_ia32_pmovzxdq256 (v4si)
21434 v4di __builtin_ia32_pmuldq256 (v8si,v8si)
21435 v16hi __builtin_ia32_pmulhrsw256 (v16hi, v16hi)
21436 v16hi __builtin_ia32_pmulhuw256 (v16hi,v16hi)
21437 v16hi __builtin_ia32_pmulhw256 (v16hi,v16hi)
21438 v16hi __builtin_ia32_pmullw256 (v16hi,v16hi)
21439 v8si __builtin_ia32_pmulld256 (v8si,v8si)
21440 v4di __builtin_ia32_pmuludq256 (v8si,v8si)
21441 v4di __builtin_ia32_por256 (v4di,v4di)
21442 v16hi __builtin_ia32_psadbw256 (v32qi,v32qi)
21443 v32qi __builtin_ia32_pshufb256 (v32qi,v32qi)
21444 v8si __builtin_ia32_pshufd256 (v8si,int)
21445 v16hi __builtin_ia32_pshufhw256 (v16hi,int)
21446 v16hi __builtin_ia32_pshuflw256 (v16hi,int)
21447 v32qi __builtin_ia32_psignb256 (v32qi,v32qi)
21448 v16hi __builtin_ia32_psignw256 (v16hi,v16hi)
21449 v8si __builtin_ia32_psignd256 (v8si,v8si)
21450 v4di __builtin_ia32_pslldqi256 (v4di,int)
21451 v16hi __builtin_ia32_psllwi256 (16hi,int)
21452 v16hi __builtin_ia32_psllw256(v16hi,v8hi)
21453 v8si __builtin_ia32_pslldi256 (v8si,int)
21454 v8si __builtin_ia32_pslld256(v8si,v4si)
21455 v4di __builtin_ia32_psllqi256 (v4di,int)
21456 v4di __builtin_ia32_psllq256(v4di,v2di)
21457 v16hi __builtin_ia32_psrawi256 (v16hi,int)
21458 v16hi __builtin_ia32_psraw256 (v16hi,v8hi)
21459 v8si __builtin_ia32_psradi256 (v8si,int)
21460 v8si __builtin_ia32_psrad256 (v8si,v4si)
21461 v4di __builtin_ia32_psrldqi256 (v4di, int)
21462 v16hi __builtin_ia32_psrlwi256 (v16hi,int)
21463 v16hi __builtin_ia32_psrlw256 (v16hi,v8hi)
21464 v8si __builtin_ia32_psrldi256 (v8si,int)
21465 v8si __builtin_ia32_psrld256 (v8si,v4si)
21466 v4di __builtin_ia32_psrlqi256 (v4di,int)
21467 v4di __builtin_ia32_psrlq256(v4di,v2di)
21468 v32qi __builtin_ia32_psubb256 (v32qi,v32qi)
21469 v32hi __builtin_ia32_psubw256 (v16hi,v16hi)
21470 v8si __builtin_ia32_psubd256 (v8si,v8si)
21471 v4di __builtin_ia32_psubq256 (v4di,v4di)
21472 v32qi __builtin_ia32_psubsb256 (v32qi,v32qi)
21473 v16hi __builtin_ia32_psubsw256 (v16hi,v16hi)
21474 v32qi __builtin_ia32_psubusb256 (v32qi,v32qi)
21475 v16hi __builtin_ia32_psubusw256 (v16hi,v16hi)
21476 v32qi __builtin_ia32_punpckhbw256 (v32qi,v32qi)
21477 v16hi __builtin_ia32_punpckhwd256 (v16hi,v16hi)
21478 v8si __builtin_ia32_punpckhdq256 (v8si,v8si)
21479 v4di __builtin_ia32_punpckhqdq256 (v4di,v4di)
21480 v32qi __builtin_ia32_punpcklbw256 (v32qi,v32qi)
21481 v16hi __builtin_ia32_punpcklwd256 (v16hi,v16hi)
21482 v8si __builtin_ia32_punpckldq256 (v8si,v8si)
21483 v4di __builtin_ia32_punpcklqdq256 (v4di,v4di)
21484 v4di __builtin_ia32_pxor256 (v4di,v4di)
21485 v4di __builtin_ia32_movntdqa256 (pv4di)
21486 v4sf __builtin_ia32_vbroadcastss_ps (v4sf)
21487 v8sf __builtin_ia32_vbroadcastss_ps256 (v4sf)
21488 v4df __builtin_ia32_vbroadcastsd_pd256 (v2df)
21489 v4di __builtin_ia32_vbroadcastsi256 (v2di)
21490 v4si __builtin_ia32_pblendd128 (v4si,v4si)
21491 v8si __builtin_ia32_pblendd256 (v8si,v8si)
21492 v32qi __builtin_ia32_pbroadcastb256 (v16qi)
21493 v16hi __builtin_ia32_pbroadcastw256 (v8hi)
21494 v8si __builtin_ia32_pbroadcastd256 (v4si)
21495 v4di __builtin_ia32_pbroadcastq256 (v2di)
21496 v16qi __builtin_ia32_pbroadcastb128 (v16qi)
21497 v8hi __builtin_ia32_pbroadcastw128 (v8hi)
21498 v4si __builtin_ia32_pbroadcastd128 (v4si)
21499 v2di __builtin_ia32_pbroadcastq128 (v2di)
21500 v8si __builtin_ia32_permvarsi256 (v8si,v8si)
21501 v4df __builtin_ia32_permdf256 (v4df,int)
21502 v8sf __builtin_ia32_permvarsf256 (v8sf,v8sf)
21503 v4di __builtin_ia32_permdi256 (v4di,int)
21504 v4di __builtin_ia32_permti256 (v4di,v4di,int)
21505 v4di __builtin_ia32_extract128i256 (v4di,int)
21506 v4di __builtin_ia32_insert128i256 (v4di,v2di,int)
21507 v8si __builtin_ia32_maskloadd256 (pcv8si,v8si)
21508 v4di __builtin_ia32_maskloadq256 (pcv4di,v4di)
21509 v4si __builtin_ia32_maskloadd (pcv4si,v4si)
21510 v2di __builtin_ia32_maskloadq (pcv2di,v2di)
21511 void __builtin_ia32_maskstored256 (pv8si,v8si,v8si)
21512 void __builtin_ia32_maskstoreq256 (pv4di,v4di,v4di)
21513 void __builtin_ia32_maskstored (pv4si,v4si,v4si)
21514 void __builtin_ia32_maskstoreq (pv2di,v2di,v2di)
21515 v8si __builtin_ia32_psllv8si (v8si,v8si)
21516 v4si __builtin_ia32_psllv4si (v4si,v4si)
21517 v4di __builtin_ia32_psllv4di (v4di,v4di)
21518 v2di __builtin_ia32_psllv2di (v2di,v2di)
21519 v8si __builtin_ia32_psrav8si (v8si,v8si)
21520 v4si __builtin_ia32_psrav4si (v4si,v4si)
21521 v8si __builtin_ia32_psrlv8si (v8si,v8si)
21522 v4si __builtin_ia32_psrlv4si (v4si,v4si)
21523 v4di __builtin_ia32_psrlv4di (v4di,v4di)
21524 v2di __builtin_ia32_psrlv2di (v2di,v2di)
21525 v2df __builtin_ia32_gathersiv2df (v2df, pcdouble,v4si,v2df,int)
21526 v4df __builtin_ia32_gathersiv4df (v4df, pcdouble,v4si,v4df,int)
21527 v2df __builtin_ia32_gatherdiv2df (v2df, pcdouble,v2di,v2df,int)
21528 v4df __builtin_ia32_gatherdiv4df (v4df, pcdouble,v4di,v4df,int)
21529 v4sf __builtin_ia32_gathersiv4sf (v4sf, pcfloat,v4si,v4sf,int)
21530 v8sf __builtin_ia32_gathersiv8sf (v8sf, pcfloat,v8si,v8sf,int)
21531 v4sf __builtin_ia32_gatherdiv4sf (v4sf, pcfloat,v2di,v4sf,int)
21532 v4sf __builtin_ia32_gatherdiv4sf256 (v4sf, pcfloat,v4di,v4sf,int)
21533 v2di __builtin_ia32_gathersiv2di (v2di, pcint64,v4si,v2di,int)
21534 v4di __builtin_ia32_gathersiv4di (v4di, pcint64,v4si,v4di,int)
21535 v2di __builtin_ia32_gatherdiv2di (v2di, pcint64,v2di,v2di,int)
21536 v4di __builtin_ia32_gatherdiv4di (v4di, pcint64,v4di,v4di,int)
21537 v4si __builtin_ia32_gathersiv4si (v4si, pcint,v4si,v4si,int)
21538 v8si __builtin_ia32_gathersiv8si (v8si, pcint,v8si,v8si,int)
21539 v4si __builtin_ia32_gatherdiv4si (v4si, pcint,v2di,v4si,int)
21540 v4si __builtin_ia32_gatherdiv4si256 (v4si, pcint,v4di,v4si,int)
21543 The following built-in functions are available when @option{-maes} is
21544 used. All of them generate the machine instruction that is part of the
21548 v2di __builtin_ia32_aesenc128 (v2di, v2di)
21549 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
21550 v2di __builtin_ia32_aesdec128 (v2di, v2di)
21551 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
21552 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
21553 v2di __builtin_ia32_aesimc128 (v2di)
21556 The following built-in function is available when @option{-mpclmul} is
21560 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
21561 Generates the @code{pclmulqdq} machine instruction.
21564 The following built-in function is available when @option{-mfsgsbase} is
21565 used. All of them generate the machine instruction that is part of the
21569 unsigned int __builtin_ia32_rdfsbase32 (void)
21570 unsigned long long __builtin_ia32_rdfsbase64 (void)
21571 unsigned int __builtin_ia32_rdgsbase32 (void)
21572 unsigned long long __builtin_ia32_rdgsbase64 (void)
21573 void _writefsbase_u32 (unsigned int)
21574 void _writefsbase_u64 (unsigned long long)
21575 void _writegsbase_u32 (unsigned int)
21576 void _writegsbase_u64 (unsigned long long)
21579 The following built-in function is available when @option{-mrdrnd} is
21580 used. All of them generate the machine instruction that is part of the
21584 unsigned int __builtin_ia32_rdrand16_step (unsigned short *)
21585 unsigned int __builtin_ia32_rdrand32_step (unsigned int *)
21586 unsigned int __builtin_ia32_rdrand64_step (unsigned long long *)
21589 The following built-in functions are available when @option{-msse4a} is used.
21590 All of them generate the machine instruction that is part of the name.
21593 void __builtin_ia32_movntsd (double *, v2df)
21594 void __builtin_ia32_movntss (float *, v4sf)
21595 v2di __builtin_ia32_extrq (v2di, v16qi)
21596 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
21597 v2di __builtin_ia32_insertq (v2di, v2di)
21598 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
21601 The following built-in functions are available when @option{-mxop} is used.
21603 v2df __builtin_ia32_vfrczpd (v2df)
21604 v4sf __builtin_ia32_vfrczps (v4sf)
21605 v2df __builtin_ia32_vfrczsd (v2df)
21606 v4sf __builtin_ia32_vfrczss (v4sf)
21607 v4df __builtin_ia32_vfrczpd256 (v4df)
21608 v8sf __builtin_ia32_vfrczps256 (v8sf)
21609 v2di __builtin_ia32_vpcmov (v2di, v2di, v2di)
21610 v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di)
21611 v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si)
21612 v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi)
21613 v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi)
21614 v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df)
21615 v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf)
21616 v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di)
21617 v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si)
21618 v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi)
21619 v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi)
21620 v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df)
21621 v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf)
21622 v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi)
21623 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
21624 v4si __builtin_ia32_vpcomeqd (v4si, v4si)
21625 v2di __builtin_ia32_vpcomeqq (v2di, v2di)
21626 v16qi __builtin_ia32_vpcomequb (v16qi, v16qi)
21627 v4si __builtin_ia32_vpcomequd (v4si, v4si)
21628 v2di __builtin_ia32_vpcomequq (v2di, v2di)
21629 v8hi __builtin_ia32_vpcomequw (v8hi, v8hi)
21630 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
21631 v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi)
21632 v4si __builtin_ia32_vpcomfalsed (v4si, v4si)
21633 v2di __builtin_ia32_vpcomfalseq (v2di, v2di)
21634 v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi)
21635 v4si __builtin_ia32_vpcomfalseud (v4si, v4si)
21636 v2di __builtin_ia32_vpcomfalseuq (v2di, v2di)
21637 v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi)
21638 v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi)
21639 v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi)
21640 v4si __builtin_ia32_vpcomged (v4si, v4si)
21641 v2di __builtin_ia32_vpcomgeq (v2di, v2di)
21642 v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi)
21643 v4si __builtin_ia32_vpcomgeud (v4si, v4si)
21644 v2di __builtin_ia32_vpcomgeuq (v2di, v2di)
21645 v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi)
21646 v8hi __builtin_ia32_vpcomgew (v8hi, v8hi)
21647 v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi)
21648 v4si __builtin_ia32_vpcomgtd (v4si, v4si)
21649 v2di __builtin_ia32_vpcomgtq (v2di, v2di)
21650 v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi)
21651 v4si __builtin_ia32_vpcomgtud (v4si, v4si)
21652 v2di __builtin_ia32_vpcomgtuq (v2di, v2di)
21653 v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi)
21654 v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi)
21655 v16qi __builtin_ia32_vpcomleb (v16qi, v16qi)
21656 v4si __builtin_ia32_vpcomled (v4si, v4si)
21657 v2di __builtin_ia32_vpcomleq (v2di, v2di)
21658 v16qi __builtin_ia32_vpcomleub (v16qi, v16qi)
21659 v4si __builtin_ia32_vpcomleud (v4si, v4si)
21660 v2di __builtin_ia32_vpcomleuq (v2di, v2di)
21661 v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi)
21662 v8hi __builtin_ia32_vpcomlew (v8hi, v8hi)
21663 v16qi __builtin_ia32_vpcomltb (v16qi, v16qi)
21664 v4si __builtin_ia32_vpcomltd (v4si, v4si)
21665 v2di __builtin_ia32_vpcomltq (v2di, v2di)
21666 v16qi __builtin_ia32_vpcomltub (v16qi, v16qi)
21667 v4si __builtin_ia32_vpcomltud (v4si, v4si)
21668 v2di __builtin_ia32_vpcomltuq (v2di, v2di)
21669 v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi)
21670 v8hi __builtin_ia32_vpcomltw (v8hi, v8hi)
21671 v16qi __builtin_ia32_vpcomneb (v16qi, v16qi)
21672 v4si __builtin_ia32_vpcomned (v4si, v4si)
21673 v2di __builtin_ia32_vpcomneq (v2di, v2di)
21674 v16qi __builtin_ia32_vpcomneub (v16qi, v16qi)
21675 v4si __builtin_ia32_vpcomneud (v4si, v4si)
21676 v2di __builtin_ia32_vpcomneuq (v2di, v2di)
21677 v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi)
21678 v8hi __builtin_ia32_vpcomnew (v8hi, v8hi)
21679 v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi)
21680 v4si __builtin_ia32_vpcomtrued (v4si, v4si)
21681 v2di __builtin_ia32_vpcomtrueq (v2di, v2di)
21682 v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi)
21683 v4si __builtin_ia32_vpcomtrueud (v4si, v4si)
21684 v2di __builtin_ia32_vpcomtrueuq (v2di, v2di)
21685 v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi)
21686 v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi)
21687 v4si __builtin_ia32_vphaddbd (v16qi)
21688 v2di __builtin_ia32_vphaddbq (v16qi)
21689 v8hi __builtin_ia32_vphaddbw (v16qi)
21690 v2di __builtin_ia32_vphadddq (v4si)
21691 v4si __builtin_ia32_vphaddubd (v16qi)
21692 v2di __builtin_ia32_vphaddubq (v16qi)
21693 v8hi __builtin_ia32_vphaddubw (v16qi)
21694 v2di __builtin_ia32_vphaddudq (v4si)
21695 v4si __builtin_ia32_vphadduwd (v8hi)
21696 v2di __builtin_ia32_vphadduwq (v8hi)
21697 v4si __builtin_ia32_vphaddwd (v8hi)
21698 v2di __builtin_ia32_vphaddwq (v8hi)
21699 v8hi __builtin_ia32_vphsubbw (v16qi)
21700 v2di __builtin_ia32_vphsubdq (v4si)
21701 v4si __builtin_ia32_vphsubwd (v8hi)
21702 v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si)
21703 v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di)
21704 v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di)
21705 v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si)
21706 v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di)
21707 v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di)
21708 v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si)
21709 v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi)
21710 v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si)
21711 v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi)
21712 v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si)
21713 v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si)
21714 v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi)
21715 v16qi __builtin_ia32_vprotb (v16qi, v16qi)
21716 v4si __builtin_ia32_vprotd (v4si, v4si)
21717 v2di __builtin_ia32_vprotq (v2di, v2di)
21718 v8hi __builtin_ia32_vprotw (v8hi, v8hi)
21719 v16qi __builtin_ia32_vpshab (v16qi, v16qi)
21720 v4si __builtin_ia32_vpshad (v4si, v4si)
21721 v2di __builtin_ia32_vpshaq (v2di, v2di)
21722 v8hi __builtin_ia32_vpshaw (v8hi, v8hi)
21723 v16qi __builtin_ia32_vpshlb (v16qi, v16qi)
21724 v4si __builtin_ia32_vpshld (v4si, v4si)
21725 v2di __builtin_ia32_vpshlq (v2di, v2di)
21726 v8hi __builtin_ia32_vpshlw (v8hi, v8hi)
21729 The following built-in functions are available when @option{-mfma4} is used.
21730 All of them generate the machine instruction that is part of the name.
21733 v2df __builtin_ia32_vfmaddpd (v2df, v2df, v2df)
21734 v4sf __builtin_ia32_vfmaddps (v4sf, v4sf, v4sf)
21735 v2df __builtin_ia32_vfmaddsd (v2df, v2df, v2df)
21736 v4sf __builtin_ia32_vfmaddss (v4sf, v4sf, v4sf)
21737 v2df __builtin_ia32_vfmsubpd (v2df, v2df, v2df)
21738 v4sf __builtin_ia32_vfmsubps (v4sf, v4sf, v4sf)
21739 v2df __builtin_ia32_vfmsubsd (v2df, v2df, v2df)
21740 v4sf __builtin_ia32_vfmsubss (v4sf, v4sf, v4sf)
21741 v2df __builtin_ia32_vfnmaddpd (v2df, v2df, v2df)
21742 v4sf __builtin_ia32_vfnmaddps (v4sf, v4sf, v4sf)
21743 v2df __builtin_ia32_vfnmaddsd (v2df, v2df, v2df)
21744 v4sf __builtin_ia32_vfnmaddss (v4sf, v4sf, v4sf)
21745 v2df __builtin_ia32_vfnmsubpd (v2df, v2df, v2df)
21746 v4sf __builtin_ia32_vfnmsubps (v4sf, v4sf, v4sf)
21747 v2df __builtin_ia32_vfnmsubsd (v2df, v2df, v2df)
21748 v4sf __builtin_ia32_vfnmsubss (v4sf, v4sf, v4sf)
21749 v2df __builtin_ia32_vfmaddsubpd (v2df, v2df, v2df)
21750 v4sf __builtin_ia32_vfmaddsubps (v4sf, v4sf, v4sf)
21751 v2df __builtin_ia32_vfmsubaddpd (v2df, v2df, v2df)
21752 v4sf __builtin_ia32_vfmsubaddps (v4sf, v4sf, v4sf)
21753 v4df __builtin_ia32_vfmaddpd256 (v4df, v4df, v4df)
21754 v8sf __builtin_ia32_vfmaddps256 (v8sf, v8sf, v8sf)
21755 v4df __builtin_ia32_vfmsubpd256 (v4df, v4df, v4df)
21756 v8sf __builtin_ia32_vfmsubps256 (v8sf, v8sf, v8sf)
21757 v4df __builtin_ia32_vfnmaddpd256 (v4df, v4df, v4df)
21758 v8sf __builtin_ia32_vfnmaddps256 (v8sf, v8sf, v8sf)
21759 v4df __builtin_ia32_vfnmsubpd256 (v4df, v4df, v4df)
21760 v8sf __builtin_ia32_vfnmsubps256 (v8sf, v8sf, v8sf)
21761 v4df __builtin_ia32_vfmaddsubpd256 (v4df, v4df, v4df)
21762 v8sf __builtin_ia32_vfmaddsubps256 (v8sf, v8sf, v8sf)
21763 v4df __builtin_ia32_vfmsubaddpd256 (v4df, v4df, v4df)
21764 v8sf __builtin_ia32_vfmsubaddps256 (v8sf, v8sf, v8sf)
21768 The following built-in functions are available when @option{-mlwp} is used.
21771 void __builtin_ia32_llwpcb16 (void *);
21772 void __builtin_ia32_llwpcb32 (void *);
21773 void __builtin_ia32_llwpcb64 (void *);
21774 void * __builtin_ia32_llwpcb16 (void);
21775 void * __builtin_ia32_llwpcb32 (void);
21776 void * __builtin_ia32_llwpcb64 (void);
21777 void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short)
21778 void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int)
21779 void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int)
21780 unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short)
21781 unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int)
21782 unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int)
21785 The following built-in functions are available when @option{-mbmi} is used.
21786 All of them generate the machine instruction that is part of the name.
21788 unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int);
21789 unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long);
21792 The following built-in functions are available when @option{-mbmi2} is used.
21793 All of them generate the machine instruction that is part of the name.
21795 unsigned int _bzhi_u32 (unsigned int, unsigned int)
21796 unsigned int _pdep_u32 (unsigned int, unsigned int)
21797 unsigned int _pext_u32 (unsigned int, unsigned int)
21798 unsigned long long _bzhi_u64 (unsigned long long, unsigned long long)
21799 unsigned long long _pdep_u64 (unsigned long long, unsigned long long)
21800 unsigned long long _pext_u64 (unsigned long long, unsigned long long)
21803 The following built-in functions are available when @option{-mlzcnt} is used.
21804 All of them generate the machine instruction that is part of the name.
21806 unsigned short __builtin_ia32_lzcnt_u16(unsigned short);
21807 unsigned int __builtin_ia32_lzcnt_u32(unsigned int);
21808 unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long);
21811 The following built-in functions are available when @option{-mfxsr} is used.
21812 All of them generate the machine instruction that is part of the name.
21814 void __builtin_ia32_fxsave (void *)
21815 void __builtin_ia32_fxrstor (void *)
21816 void __builtin_ia32_fxsave64 (void *)
21817 void __builtin_ia32_fxrstor64 (void *)
21820 The following built-in functions are available when @option{-mxsave} is used.
21821 All of them generate the machine instruction that is part of the name.
21823 void __builtin_ia32_xsave (void *, long long)
21824 void __builtin_ia32_xrstor (void *, long long)
21825 void __builtin_ia32_xsave64 (void *, long long)
21826 void __builtin_ia32_xrstor64 (void *, long long)
21829 The following built-in functions are available when @option{-mxsaveopt} is used.
21830 All of them generate the machine instruction that is part of the name.
21832 void __builtin_ia32_xsaveopt (void *, long long)
21833 void __builtin_ia32_xsaveopt64 (void *, long long)
21836 The following built-in functions are available when @option{-mtbm} is used.
21837 Both of them generate the immediate form of the bextr machine instruction.
21839 unsigned int __builtin_ia32_bextri_u32 (unsigned int,
21840 const unsigned int);
21841 unsigned long long __builtin_ia32_bextri_u64 (unsigned long long,
21842 const unsigned long long);
21846 The following built-in functions are available when @option{-m3dnow} is used.
21847 All of them generate the machine instruction that is part of the name.
21850 void __builtin_ia32_femms (void)
21851 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
21852 v2si __builtin_ia32_pf2id (v2sf)
21853 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
21854 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
21855 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
21856 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
21857 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
21858 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
21859 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
21860 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
21861 v2sf __builtin_ia32_pfrcp (v2sf)
21862 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
21863 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
21864 v2sf __builtin_ia32_pfrsqrt (v2sf)
21865 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
21866 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
21867 v2sf __builtin_ia32_pi2fd (v2si)
21868 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
21871 The following built-in functions are available when @option{-m3dnowa} is used.
21872 All of them generate the machine instruction that is part of the name.
21875 v2si __builtin_ia32_pf2iw (v2sf)
21876 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
21877 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
21878 v2sf __builtin_ia32_pi2fw (v2si)
21879 v2sf __builtin_ia32_pswapdsf (v2sf)
21880 v2si __builtin_ia32_pswapdsi (v2si)
21883 The following built-in functions are available when @option{-mrtm} is used
21884 They are used for restricted transactional memory. These are the internal
21885 low level functions. Normally the functions in
21886 @ref{x86 transactional memory intrinsics} should be used instead.
21889 int __builtin_ia32_xbegin ()
21890 void __builtin_ia32_xend ()
21891 void __builtin_ia32_xabort (status)
21892 int __builtin_ia32_xtest ()
21895 The following built-in functions are available when @option{-mmwaitx} is used.
21896 All of them generate the machine instruction that is part of the name.
21898 void __builtin_ia32_monitorx (void *, unsigned int, unsigned int)
21899 void __builtin_ia32_mwaitx (unsigned int, unsigned int, unsigned int)
21902 The following built-in functions are available when @option{-mclzero} is used.
21903 All of them generate the machine instruction that is part of the name.
21905 void __builtin_i32_clzero (void *)
21908 The following built-in functions are available when @option{-mpku} is used.
21909 They generate reads and writes to PKRU.
21911 void __builtin_ia32_wrpkru (unsigned int)
21912 unsigned int __builtin_ia32_rdpkru ()
21915 The following built-in functions are available when @option{-mcet} or
21916 @option{-mshstk} option is used. They support shadow stack
21917 machine instructions from Intel Control-flow Enforcement Technology (CET).
21918 Each built-in function generates the machine instruction that is part
21919 of the function's name. These are the internal low-level functions.
21920 Normally the functions in @ref{x86 control-flow protection intrinsics}
21921 should be used instead.
21924 unsigned int __builtin_ia32_rdsspd (void)
21925 unsigned long long __builtin_ia32_rdsspq (void)
21926 void __builtin_ia32_incsspd (unsigned int)
21927 void __builtin_ia32_incsspq (unsigned long long)
21928 void __builtin_ia32_saveprevssp(void);
21929 void __builtin_ia32_rstorssp(void *);
21930 void __builtin_ia32_wrssd(unsigned int, void *);
21931 void __builtin_ia32_wrssq(unsigned long long, void *);
21932 void __builtin_ia32_wrussd(unsigned int, void *);
21933 void __builtin_ia32_wrussq(unsigned long long, void *);
21934 void __builtin_ia32_setssbsy(void);
21935 void __builtin_ia32_clrssbsy(void *);
21938 @node x86 transactional memory intrinsics
21939 @subsection x86 Transactional Memory Intrinsics
21941 These hardware transactional memory intrinsics for x86 allow you to use
21942 memory transactions with RTM (Restricted Transactional Memory).
21943 This support is enabled with the @option{-mrtm} option.
21944 For using HLE (Hardware Lock Elision) see
21945 @ref{x86 specific memory model extensions for transactional memory} instead.
21947 A memory transaction commits all changes to memory in an atomic way,
21948 as visible to other threads. If the transaction fails it is rolled back
21949 and all side effects discarded.
21951 Generally there is no guarantee that a memory transaction ever succeeds
21952 and suitable fallback code always needs to be supplied.
21954 @deftypefn {RTM Function} {unsigned} _xbegin ()
21955 Start a RTM (Restricted Transactional Memory) transaction.
21956 Returns @code{_XBEGIN_STARTED} when the transaction
21957 started successfully (note this is not 0, so the constant has to be
21958 explicitly tested).
21960 If the transaction aborts, all side effects
21961 are undone and an abort code encoded as a bit mask is returned.
21962 The following macros are defined:
21965 @item _XABORT_EXPLICIT
21966 Transaction was explicitly aborted with @code{_xabort}. The parameter passed
21967 to @code{_xabort} is available with @code{_XABORT_CODE(status)}.
21968 @item _XABORT_RETRY
21969 Transaction retry is possible.
21970 @item _XABORT_CONFLICT
21971 Transaction abort due to a memory conflict with another thread.
21972 @item _XABORT_CAPACITY
21973 Transaction abort due to the transaction using too much memory.
21974 @item _XABORT_DEBUG
21975 Transaction abort due to a debug trap.
21976 @item _XABORT_NESTED
21977 Transaction abort in an inner nested transaction.
21980 There is no guarantee
21981 any transaction ever succeeds, so there always needs to be a valid
21985 @deftypefn {RTM Function} {void} _xend ()
21986 Commit the current transaction. When no transaction is active this faults.
21987 All memory side effects of the transaction become visible
21988 to other threads in an atomic manner.
21991 @deftypefn {RTM Function} {int} _xtest ()
21992 Return a nonzero value if a transaction is currently active, otherwise 0.
21995 @deftypefn {RTM Function} {void} _xabort (status)
21996 Abort the current transaction. When no transaction is active this is a no-op.
21997 The @var{status} is an 8-bit constant; its value is encoded in the return
21998 value from @code{_xbegin}.
22001 Here is an example showing handling for @code{_XABORT_RETRY}
22002 and a fallback path for other failures:
22005 #include <immintrin.h>
22007 int n_tries, max_tries;
22008 unsigned status = _XABORT_EXPLICIT;
22011 for (n_tries = 0; n_tries < max_tries; n_tries++)
22013 status = _xbegin ();
22014 if (status == _XBEGIN_STARTED || !(status & _XABORT_RETRY))
22017 if (status == _XBEGIN_STARTED)
22019 ... transaction code...
22024 ... non-transactional fallback path...
22029 Note that, in most cases, the transactional and non-transactional code
22030 must synchronize together to ensure consistency.
22032 @node x86 control-flow protection intrinsics
22033 @subsection x86 Control-Flow Protection Intrinsics
22035 @deftypefn {CET Function} {ret_type} _get_ssp (void)
22036 Get the current value of shadow stack pointer if shadow stack support
22037 from Intel CET is enabled in the hardware or @code{0} otherwise.
22038 The @code{ret_type} is @code{unsigned long long} for 64-bit targets
22039 and @code{unsigned int} for 32-bit targets.
22042 @deftypefn {CET Function} void _inc_ssp (unsigned int)
22043 Increment the current shadow stack pointer by the size specified by the
22044 function argument. The argument is masked to a byte value for security
22045 reasons, so to increment by more than 255 bytes you must call the function
22049 The shadow stack unwind code looks like:
22052 #include <immintrin.h>
22054 /* Unwind the shadow stack for EH. */
22055 #define _Unwind_Frames_Extra(x) \
22058 _Unwind_Word ssp = _get_ssp (); \
22061 _Unwind_Word tmp = (x); \
22062 while (tmp > 255) \
22074 This code runs unconditionally on all 64-bit processors. For 32-bit
22075 processors the code runs on those that support multi-byte NOP instructions.
22077 @node Target Format Checks
22078 @section Format Checks Specific to Particular Target Machines
22080 For some target machines, GCC supports additional options to the
22082 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
22085 * Solaris Format Checks::
22086 * Darwin Format Checks::
22089 @node Solaris Format Checks
22090 @subsection Solaris Format Checks
22092 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
22093 check. @code{cmn_err} accepts a subset of the standard @code{printf}
22094 conversions, and the two-argument @code{%b} conversion for displaying
22095 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
22097 @node Darwin Format Checks
22098 @subsection Darwin Format Checks
22100 Darwin targets support the @code{CFString} (or @code{__CFString__}) in the format
22101 attribute context. Declarations made with such attribution are parsed for correct syntax
22102 and format argument types. However, parsing of the format string itself is currently undefined
22103 and is not carried out by this version of the compiler.
22105 Additionally, @code{CFStringRefs} (defined by the @code{CoreFoundation} headers) may
22106 also be used as format arguments. Note that the relevant headers are only likely to be
22107 available on Darwin (OSX) installations. On such installations, the XCode and system
22108 documentation provide descriptions of @code{CFString}, @code{CFStringRefs} and
22109 associated functions.
22112 @section Pragmas Accepted by GCC
22114 @cindex @code{#pragma}
22116 GCC supports several types of pragmas, primarily in order to compile
22117 code originally written for other compilers. Note that in general
22118 we do not recommend the use of pragmas; @xref{Function Attributes},
22119 for further explanation.
22122 * AArch64 Pragmas::
22126 * RS/6000 and PowerPC Pragmas::
22129 * Solaris Pragmas::
22130 * Symbol-Renaming Pragmas::
22131 * Structure-Layout Pragmas::
22133 * Diagnostic Pragmas::
22134 * Visibility Pragmas::
22135 * Push/Pop Macro Pragmas::
22136 * Function Specific Option Pragmas::
22137 * Loop-Specific Pragmas::
22140 @node AArch64 Pragmas
22141 @subsection AArch64 Pragmas
22143 The pragmas defined by the AArch64 target correspond to the AArch64
22144 target function attributes. They can be specified as below:
22146 #pragma GCC target("string")
22149 where @code{@var{string}} can be any string accepted as an AArch64 target
22150 attribute. @xref{AArch64 Function Attributes}, for more details
22151 on the permissible values of @code{string}.
22154 @subsection ARM Pragmas
22156 The ARM target defines pragmas for controlling the default addition of
22157 @code{long_call} and @code{short_call} attributes to functions.
22158 @xref{Function Attributes}, for information about the effects of these
22163 @cindex pragma, long_calls
22164 Set all subsequent functions to have the @code{long_call} attribute.
22166 @item no_long_calls
22167 @cindex pragma, no_long_calls
22168 Set all subsequent functions to have the @code{short_call} attribute.
22170 @item long_calls_off
22171 @cindex pragma, long_calls_off
22172 Do not affect the @code{long_call} or @code{short_call} attributes of
22173 subsequent functions.
22177 @subsection M32C Pragmas
22180 @item GCC memregs @var{number}
22181 @cindex pragma, memregs
22182 Overrides the command-line option @code{-memregs=} for the current
22183 file. Use with care! This pragma must be before any function in the
22184 file, and mixing different memregs values in different objects may
22185 make them incompatible. This pragma is useful when a
22186 performance-critical function uses a memreg for temporary values,
22187 as it may allow you to reduce the number of memregs used.
22189 @item ADDRESS @var{name} @var{address}
22190 @cindex pragma, address
22191 For any declared symbols matching @var{name}, this does three things
22192 to that symbol: it forces the symbol to be located at the given
22193 address (a number), it forces the symbol to be volatile, and it
22194 changes the symbol's scope to be static. This pragma exists for
22195 compatibility with other compilers, but note that the common
22196 @code{1234H} numeric syntax is not supported (use @code{0x1234}
22200 #pragma ADDRESS port3 0x103
22207 @subsection MeP Pragmas
22211 @item custom io_volatile (on|off)
22212 @cindex pragma, custom io_volatile
22213 Overrides the command-line option @code{-mio-volatile} for the current
22214 file. Note that for compatibility with future GCC releases, this
22215 option should only be used once before any @code{io} variables in each
22218 @item GCC coprocessor available @var{registers}
22219 @cindex pragma, coprocessor available
22220 Specifies which coprocessor registers are available to the register
22221 allocator. @var{registers} may be a single register, register range
22222 separated by ellipses, or comma-separated list of those. Example:
22225 #pragma GCC coprocessor available $c0...$c10, $c28
22228 @item GCC coprocessor call_saved @var{registers}
22229 @cindex pragma, coprocessor call_saved
22230 Specifies which coprocessor registers are to be saved and restored by
22231 any function using them. @var{registers} may be a single register,
22232 register range separated by ellipses, or comma-separated list of
22236 #pragma GCC coprocessor call_saved $c4...$c6, $c31
22239 @item GCC coprocessor subclass '(A|B|C|D)' = @var{registers}
22240 @cindex pragma, coprocessor subclass
22241 Creates and defines a register class. These register classes can be
22242 used by inline @code{asm} constructs. @var{registers} may be a single
22243 register, register range separated by ellipses, or comma-separated
22244 list of those. Example:
22247 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6
22249 asm ("cpfoo %0" : "=B" (x));
22252 @item GCC disinterrupt @var{name} , @var{name} @dots{}
22253 @cindex pragma, disinterrupt
22254 For the named functions, the compiler adds code to disable interrupts
22255 for the duration of those functions. If any functions so named
22256 are not encountered in the source, a warning is emitted that the pragma is
22257 not used. Examples:
22260 #pragma disinterrupt foo
22261 #pragma disinterrupt bar, grill
22262 int foo () @{ @dots{} @}
22265 @item GCC call @var{name} , @var{name} @dots{}
22266 @cindex pragma, call
22267 For the named functions, the compiler always uses a register-indirect
22268 call model when calling the named functions. Examples:
22277 @node RS/6000 and PowerPC Pragmas
22278 @subsection RS/6000 and PowerPC Pragmas
22280 The RS/6000 and PowerPC targets define one pragma for controlling
22281 whether or not the @code{longcall} attribute is added to function
22282 declarations by default. This pragma overrides the @option{-mlongcall}
22283 option, but not the @code{longcall} and @code{shortcall} attributes.
22284 @xref{RS/6000 and PowerPC Options}, for more information about when long
22285 calls are and are not necessary.
22289 @cindex pragma, longcall
22290 Apply the @code{longcall} attribute to all subsequent function
22294 Do not apply the @code{longcall} attribute to subsequent function
22298 @c Describe h8300 pragmas here.
22299 @c Describe sh pragmas here.
22300 @c Describe v850 pragmas here.
22302 @node S/390 Pragmas
22303 @subsection S/390 Pragmas
22305 The pragmas defined by the S/390 target correspond to the S/390
22306 target function attributes and some the additional options:
22313 Note that options of the pragma, unlike options of the target
22314 attribute, do change the value of preprocessor macros like
22315 @code{__VEC__}. They can be specified as below:
22318 #pragma GCC target("string[,string]...")
22319 #pragma GCC target("string"[,"string"]...)
22322 @node Darwin Pragmas
22323 @subsection Darwin Pragmas
22325 The following pragmas are available for all architectures running the
22326 Darwin operating system. These are useful for compatibility with other
22330 @item mark @var{tokens}@dots{}
22331 @cindex pragma, mark
22332 This pragma is accepted, but has no effect.
22334 @item options align=@var{alignment}
22335 @cindex pragma, options align
22336 This pragma sets the alignment of fields in structures. The values of
22337 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
22338 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
22339 properly; to restore the previous setting, use @code{reset} for the
22342 @item segment @var{tokens}@dots{}
22343 @cindex pragma, segment
22344 This pragma is accepted, but has no effect.
22346 @item unused (@var{var} [, @var{var}]@dots{})
22347 @cindex pragma, unused
22348 This pragma declares variables to be possibly unused. GCC does not
22349 produce warnings for the listed variables. The effect is similar to
22350 that of the @code{unused} attribute, except that this pragma may appear
22351 anywhere within the variables' scopes.
22354 @node Solaris Pragmas
22355 @subsection Solaris Pragmas
22357 The Solaris target supports @code{#pragma redefine_extname}
22358 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
22359 @code{#pragma} directives for compatibility with the system compiler.
22362 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
22363 @cindex pragma, align
22365 Increase the minimum alignment of each @var{variable} to @var{alignment}.
22366 This is the same as GCC's @code{aligned} attribute @pxref{Variable
22367 Attributes}). Macro expansion occurs on the arguments to this pragma
22368 when compiling C and Objective-C@. It does not currently occur when
22369 compiling C++, but this is a bug which may be fixed in a future
22372 @item fini (@var{function} [, @var{function}]...)
22373 @cindex pragma, fini
22375 This pragma causes each listed @var{function} to be called after
22376 main, or during shared module unloading, by adding a call to the
22377 @code{.fini} section.
22379 @item init (@var{function} [, @var{function}]...)
22380 @cindex pragma, init
22382 This pragma causes each listed @var{function} to be called during
22383 initialization (before @code{main}) or during shared module loading, by
22384 adding a call to the @code{.init} section.
22388 @node Symbol-Renaming Pragmas
22389 @subsection Symbol-Renaming Pragmas
22391 GCC supports a @code{#pragma} directive that changes the name used in
22392 assembly for a given declaration. While this pragma is supported on all
22393 platforms, it is intended primarily to provide compatibility with the
22394 Solaris system headers. This effect can also be achieved using the asm
22395 labels extension (@pxref{Asm Labels}).
22398 @item redefine_extname @var{oldname} @var{newname}
22399 @cindex pragma, redefine_extname
22401 This pragma gives the C function @var{oldname} the assembly symbol
22402 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
22403 is defined if this pragma is available (currently on all platforms).
22406 This pragma and the asm labels extension interact in a complicated
22407 manner. Here are some corner cases you may want to be aware of:
22410 @item This pragma silently applies only to declarations with external
22411 linkage. Asm labels do not have this restriction.
22413 @item In C++, this pragma silently applies only to declarations with
22414 ``C'' linkage. Again, asm labels do not have this restriction.
22416 @item If either of the ways of changing the assembly name of a
22417 declaration are applied to a declaration whose assembly name has
22418 already been determined (either by a previous use of one of these
22419 features, or because the compiler needed the assembly name in order to
22420 generate code), and the new name is different, a warning issues and
22421 the name does not change.
22423 @item The @var{oldname} used by @code{#pragma redefine_extname} is
22424 always the C-language name.
22427 @node Structure-Layout Pragmas
22428 @subsection Structure-Layout Pragmas
22430 For compatibility with Microsoft Windows compilers, GCC supports a
22431 set of @code{#pragma} directives that change the maximum alignment of
22432 members of structures (other than zero-width bit-fields), unions, and
22433 classes subsequently defined. The @var{n} value below always is required
22434 to be a small power of two and specifies the new alignment in bytes.
22437 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
22438 @item @code{#pragma pack()} sets the alignment to the one that was in
22439 effect when compilation started (see also command-line option
22440 @option{-fpack-struct[=@var{n}]} @pxref{Code Gen Options}).
22441 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
22442 setting on an internal stack and then optionally sets the new alignment.
22443 @item @code{#pragma pack(pop)} restores the alignment setting to the one
22444 saved at the top of the internal stack (and removes that stack entry).
22445 Note that @code{#pragma pack([@var{n}])} does not influence this internal
22446 stack; thus it is possible to have @code{#pragma pack(push)} followed by
22447 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
22448 @code{#pragma pack(pop)}.
22451 Some targets, e.g.@: x86 and PowerPC, support the @code{#pragma ms_struct}
22452 directive which lays out structures and unions subsequently defined as the
22453 documented @code{__attribute__ ((ms_struct))}.
22456 @item @code{#pragma ms_struct on} turns on the Microsoft layout.
22457 @item @code{#pragma ms_struct off} turns off the Microsoft layout.
22458 @item @code{#pragma ms_struct reset} goes back to the default layout.
22461 Most targets also support the @code{#pragma scalar_storage_order} directive
22462 which lays out structures and unions subsequently defined as the documented
22463 @code{__attribute__ ((scalar_storage_order))}.
22466 @item @code{#pragma scalar_storage_order big-endian} sets the storage order
22467 of the scalar fields to big-endian.
22468 @item @code{#pragma scalar_storage_order little-endian} sets the storage order
22469 of the scalar fields to little-endian.
22470 @item @code{#pragma scalar_storage_order default} goes back to the endianness
22471 that was in effect when compilation started (see also command-line option
22472 @option{-fsso-struct=@var{endianness}} @pxref{C Dialect Options}).
22476 @subsection Weak Pragmas
22478 For compatibility with SVR4, GCC supports a set of @code{#pragma}
22479 directives for declaring symbols to be weak, and defining weak
22483 @item #pragma weak @var{symbol}
22484 @cindex pragma, weak
22485 This pragma declares @var{symbol} to be weak, as if the declaration
22486 had the attribute of the same name. The pragma may appear before
22487 or after the declaration of @var{symbol}. It is not an error for
22488 @var{symbol} to never be defined at all.
22490 @item #pragma weak @var{symbol1} = @var{symbol2}
22491 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
22492 It is an error if @var{symbol2} is not defined in the current
22496 @node Diagnostic Pragmas
22497 @subsection Diagnostic Pragmas
22499 GCC allows the user to selectively enable or disable certain types of
22500 diagnostics, and change the kind of the diagnostic. For example, a
22501 project's policy might require that all sources compile with
22502 @option{-Werror} but certain files might have exceptions allowing
22503 specific types of warnings. Or, a project might selectively enable
22504 diagnostics and treat them as errors depending on which preprocessor
22505 macros are defined.
22508 @item #pragma GCC diagnostic @var{kind} @var{option}
22509 @cindex pragma, diagnostic
22511 Modifies the disposition of a diagnostic. Note that not all
22512 diagnostics are modifiable; at the moment only warnings (normally
22513 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
22514 Use @option{-fdiagnostics-show-option} to determine which diagnostics
22515 are controllable and which option controls them.
22517 @var{kind} is @samp{error} to treat this diagnostic as an error,
22518 @samp{warning} to treat it like a warning (even if @option{-Werror} is
22519 in effect), or @samp{ignored} if the diagnostic is to be ignored.
22520 @var{option} is a double quoted string that matches the command-line
22524 #pragma GCC diagnostic warning "-Wformat"
22525 #pragma GCC diagnostic error "-Wformat"
22526 #pragma GCC diagnostic ignored "-Wformat"
22529 Note that these pragmas override any command-line options. GCC keeps
22530 track of the location of each pragma, and issues diagnostics according
22531 to the state as of that point in the source file. Thus, pragmas occurring
22532 after a line do not affect diagnostics caused by that line.
22534 @item #pragma GCC diagnostic push
22535 @itemx #pragma GCC diagnostic pop
22537 Causes GCC to remember the state of the diagnostics as of each
22538 @code{push}, and restore to that point at each @code{pop}. If a
22539 @code{pop} has no matching @code{push}, the command-line options are
22543 #pragma GCC diagnostic error "-Wuninitialized"
22544 foo(a); /* error is given for this one */
22545 #pragma GCC diagnostic push
22546 #pragma GCC diagnostic ignored "-Wuninitialized"
22547 foo(b); /* no diagnostic for this one */
22548 #pragma GCC diagnostic pop
22549 foo(c); /* error is given for this one */
22550 #pragma GCC diagnostic pop
22551 foo(d); /* depends on command-line options */
22556 GCC also offers a simple mechanism for printing messages during
22560 @item #pragma message @var{string}
22561 @cindex pragma, diagnostic
22563 Prints @var{string} as a compiler message on compilation. The message
22564 is informational only, and is neither a compilation warning nor an error.
22567 #pragma message "Compiling " __FILE__ "..."
22570 @var{string} may be parenthesized, and is printed with location
22571 information. For example,
22574 #define DO_PRAGMA(x) _Pragma (#x)
22575 #define TODO(x) DO_PRAGMA(message ("TODO - " #x))
22577 TODO(Remember to fix this)
22581 prints @samp{/tmp/file.c:4: note: #pragma message:
22582 TODO - Remember to fix this}.
22586 @node Visibility Pragmas
22587 @subsection Visibility Pragmas
22590 @item #pragma GCC visibility push(@var{visibility})
22591 @itemx #pragma GCC visibility pop
22592 @cindex pragma, visibility
22594 This pragma allows the user to set the visibility for multiple
22595 declarations without having to give each a visibility attribute
22596 (@pxref{Function Attributes}).
22598 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
22599 declarations. Class members and template specializations are not
22600 affected; if you want to override the visibility for a particular
22601 member or instantiation, you must use an attribute.
22606 @node Push/Pop Macro Pragmas
22607 @subsection Push/Pop Macro Pragmas
22609 For compatibility with Microsoft Windows compilers, GCC supports
22610 @samp{#pragma push_macro(@var{"macro_name"})}
22611 and @samp{#pragma pop_macro(@var{"macro_name"})}.
22614 @item #pragma push_macro(@var{"macro_name"})
22615 @cindex pragma, push_macro
22616 This pragma saves the value of the macro named as @var{macro_name} to
22617 the top of the stack for this macro.
22619 @item #pragma pop_macro(@var{"macro_name"})
22620 @cindex pragma, pop_macro
22621 This pragma sets the value of the macro named as @var{macro_name} to
22622 the value on top of the stack for this macro. If the stack for
22623 @var{macro_name} is empty, the value of the macro remains unchanged.
22630 #pragma push_macro("X")
22633 #pragma pop_macro("X")
22638 In this example, the definition of X as 1 is saved by @code{#pragma
22639 push_macro} and restored by @code{#pragma pop_macro}.
22641 @node Function Specific Option Pragmas
22642 @subsection Function Specific Option Pragmas
22645 @item #pragma GCC target (@var{"string"}...)
22646 @cindex pragma GCC target
22648 This pragma allows you to set target specific options for functions
22649 defined later in the source file. One or more strings can be
22650 specified. Each function that is defined after this point is as
22651 if @code{attribute((target("STRING")))} was specified for that
22652 function. The parenthesis around the options is optional.
22653 @xref{Function Attributes}, for more information about the
22654 @code{target} attribute and the attribute syntax.
22656 The @code{#pragma GCC target} pragma is presently implemented for
22657 x86, ARM, AArch64, PowerPC, S/390, and Nios II targets only.
22659 @item #pragma GCC optimize (@var{"string"}...)
22660 @cindex pragma GCC optimize
22662 This pragma allows you to set global optimization options for functions
22663 defined later in the source file. One or more strings can be
22664 specified. Each function that is defined after this point is as
22665 if @code{attribute((optimize("STRING")))} was specified for that
22666 function. The parenthesis around the options is optional.
22667 @xref{Function Attributes}, for more information about the
22668 @code{optimize} attribute and the attribute syntax.
22670 @item #pragma GCC push_options
22671 @itemx #pragma GCC pop_options
22672 @cindex pragma GCC push_options
22673 @cindex pragma GCC pop_options
22675 These pragmas maintain a stack of the current target and optimization
22676 options. It is intended for include files where you temporarily want
22677 to switch to using a different @samp{#pragma GCC target} or
22678 @samp{#pragma GCC optimize} and then to pop back to the previous
22681 @item #pragma GCC reset_options
22682 @cindex pragma GCC reset_options
22684 This pragma clears the current @code{#pragma GCC target} and
22685 @code{#pragma GCC optimize} to use the default switches as specified
22686 on the command line.
22690 @node Loop-Specific Pragmas
22691 @subsection Loop-Specific Pragmas
22694 @item #pragma GCC ivdep
22695 @cindex pragma GCC ivdep
22697 With this pragma, the programmer asserts that there are no loop-carried
22698 dependencies which would prevent consecutive iterations of
22699 the following loop from executing concurrently with SIMD
22700 (single instruction multiple data) instructions.
22702 For example, the compiler can only unconditionally vectorize the following
22703 loop with the pragma:
22706 void foo (int n, int *a, int *b, int *c)
22710 for (i = 0; i < n; ++i)
22711 a[i] = b[i] + c[i];
22716 In this example, using the @code{restrict} qualifier had the same
22717 effect. In the following example, that would not be possible. Assume
22718 @math{k < -m} or @math{k >= m}. Only with the pragma, the compiler knows
22719 that it can unconditionally vectorize the following loop:
22722 void ignore_vec_dep (int *a, int k, int c, int m)
22725 for (int i = 0; i < m; i++)
22726 a[i] = a[i + k] * c;
22730 @item #pragma GCC unroll @var{n}
22731 @cindex pragma GCC unroll @var{n}
22733 You can use this pragma to control how many times a loop should be unrolled.
22734 It must be placed immediately before a @code{for}, @code{while} or @code{do}
22735 loop or a @code{#pragma GCC ivdep}, and applies only to the loop that follows.
22736 @var{n} is an integer constant expression specifying the unrolling factor.
22737 The values of @math{0} and @math{1} block any unrolling of the loop.
22741 @node Unnamed Fields
22742 @section Unnamed Structure and Union Fields
22743 @cindex @code{struct}
22744 @cindex @code{union}
22746 As permitted by ISO C11 and for compatibility with other compilers,
22747 GCC allows you to define
22748 a structure or union that contains, as fields, structures and unions
22749 without names. For example:
22763 In this example, you are able to access members of the unnamed
22764 union with code like @samp{foo.b}. Note that only unnamed structs and
22765 unions are allowed, you may not have, for example, an unnamed
22768 You must never create such structures that cause ambiguous field definitions.
22769 For example, in this structure:
22781 it is ambiguous which @code{a} is being referred to with @samp{foo.a}.
22782 The compiler gives errors for such constructs.
22784 @opindex fms-extensions
22785 Unless @option{-fms-extensions} is used, the unnamed field must be a
22786 structure or union definition without a tag (for example, @samp{struct
22787 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
22788 also be a definition with a tag such as @samp{struct foo @{ int a;
22789 @};}, a reference to a previously defined structure or union such as
22790 @samp{struct foo;}, or a reference to a @code{typedef} name for a
22791 previously defined structure or union type.
22793 @opindex fplan9-extensions
22794 The option @option{-fplan9-extensions} enables
22795 @option{-fms-extensions} as well as two other extensions. First, a
22796 pointer to a structure is automatically converted to a pointer to an
22797 anonymous field for assignments and function calls. For example:
22800 struct s1 @{ int a; @};
22801 struct s2 @{ struct s1; @};
22802 extern void f1 (struct s1 *);
22803 void f2 (struct s2 *p) @{ f1 (p); @}
22807 In the call to @code{f1} inside @code{f2}, the pointer @code{p} is
22808 converted into a pointer to the anonymous field.
22810 Second, when the type of an anonymous field is a @code{typedef} for a
22811 @code{struct} or @code{union}, code may refer to the field using the
22812 name of the @code{typedef}.
22815 typedef struct @{ int a; @} s1;
22816 struct s2 @{ s1; @};
22817 s1 f1 (struct s2 *p) @{ return p->s1; @}
22820 These usages are only permitted when they are not ambiguous.
22823 @section Thread-Local Storage
22824 @cindex Thread-Local Storage
22825 @cindex @acronym{TLS}
22826 @cindex @code{__thread}
22828 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
22829 are allocated such that there is one instance of the variable per extant
22830 thread. The runtime model GCC uses to implement this originates
22831 in the IA-64 processor-specific ABI, but has since been migrated
22832 to other processors as well. It requires significant support from
22833 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
22834 system libraries (@file{libc.so} and @file{libpthread.so}), so it
22835 is not available everywhere.
22837 At the user level, the extension is visible with a new storage
22838 class keyword: @code{__thread}. For example:
22842 extern __thread struct state s;
22843 static __thread char *p;
22846 The @code{__thread} specifier may be used alone, with the @code{extern}
22847 or @code{static} specifiers, but with no other storage class specifier.
22848 When used with @code{extern} or @code{static}, @code{__thread} must appear
22849 immediately after the other storage class specifier.
22851 The @code{__thread} specifier may be applied to any global, file-scoped
22852 static, function-scoped static, or static data member of a class. It may
22853 not be applied to block-scoped automatic or non-static data member.
22855 When the address-of operator is applied to a thread-local variable, it is
22856 evaluated at run time and returns the address of the current thread's
22857 instance of that variable. An address so obtained may be used by any
22858 thread. When a thread terminates, any pointers to thread-local variables
22859 in that thread become invalid.
22861 No static initialization may refer to the address of a thread-local variable.
22863 In C++, if an initializer is present for a thread-local variable, it must
22864 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
22867 See @uref{https://www.akkadia.org/drepper/tls.pdf,
22868 ELF Handling For Thread-Local Storage} for a detailed explanation of
22869 the four thread-local storage addressing models, and how the runtime
22870 is expected to function.
22873 * C99 Thread-Local Edits::
22874 * C++98 Thread-Local Edits::
22877 @node C99 Thread-Local Edits
22878 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
22880 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
22881 that document the exact semantics of the language extension.
22885 @cite{5.1.2 Execution environments}
22887 Add new text after paragraph 1
22890 Within either execution environment, a @dfn{thread} is a flow of
22891 control within a program. It is implementation defined whether
22892 or not there may be more than one thread associated with a program.
22893 It is implementation defined how threads beyond the first are
22894 created, the name and type of the function called at thread
22895 startup, and how threads may be terminated. However, objects
22896 with thread storage duration shall be initialized before thread
22901 @cite{6.2.4 Storage durations of objects}
22903 Add new text before paragraph 3
22906 An object whose identifier is declared with the storage-class
22907 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
22908 Its lifetime is the entire execution of the thread, and its
22909 stored value is initialized only once, prior to thread startup.
22913 @cite{6.4.1 Keywords}
22915 Add @code{__thread}.
22918 @cite{6.7.1 Storage-class specifiers}
22920 Add @code{__thread} to the list of storage class specifiers in
22923 Change paragraph 2 to
22926 With the exception of @code{__thread}, at most one storage-class
22927 specifier may be given [@dots{}]. The @code{__thread} specifier may
22928 be used alone, or immediately following @code{extern} or
22932 Add new text after paragraph 6
22935 The declaration of an identifier for a variable that has
22936 block scope that specifies @code{__thread} shall also
22937 specify either @code{extern} or @code{static}.
22939 The @code{__thread} specifier shall be used only with
22944 @node C++98 Thread-Local Edits
22945 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
22947 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
22948 that document the exact semantics of the language extension.
22952 @b{[intro.execution]}
22954 New text after paragraph 4
22957 A @dfn{thread} is a flow of control within the abstract machine.
22958 It is implementation defined whether or not there may be more than
22962 New text after paragraph 7
22965 It is unspecified whether additional action must be taken to
22966 ensure when and whether side effects are visible to other threads.
22972 Add @code{__thread}.
22975 @b{[basic.start.main]}
22977 Add after paragraph 5
22980 The thread that begins execution at the @code{main} function is called
22981 the @dfn{main thread}. It is implementation defined how functions
22982 beginning threads other than the main thread are designated or typed.
22983 A function so designated, as well as the @code{main} function, is called
22984 a @dfn{thread startup function}. It is implementation defined what
22985 happens if a thread startup function returns. It is implementation
22986 defined what happens to other threads when any thread calls @code{exit}.
22990 @b{[basic.start.init]}
22992 Add after paragraph 4
22995 The storage for an object of thread storage duration shall be
22996 statically initialized before the first statement of the thread startup
22997 function. An object of thread storage duration shall not require
22998 dynamic initialization.
23002 @b{[basic.start.term]}
23004 Add after paragraph 3
23007 The type of an object with thread storage duration shall not have a
23008 non-trivial destructor, nor shall it be an array type whose elements
23009 (directly or indirectly) have non-trivial destructors.
23015 Add ``thread storage duration'' to the list in paragraph 1.
23020 Thread, static, and automatic storage durations are associated with
23021 objects introduced by declarations [@dots{}].
23024 Add @code{__thread} to the list of specifiers in paragraph 3.
23027 @b{[basic.stc.thread]}
23029 New section before @b{[basic.stc.static]}
23032 The keyword @code{__thread} applied to a non-local object gives the
23033 object thread storage duration.
23035 A local variable or class data member declared both @code{static}
23036 and @code{__thread} gives the variable or member thread storage
23041 @b{[basic.stc.static]}
23046 All objects that have neither thread storage duration, dynamic
23047 storage duration nor are local [@dots{}].
23053 Add @code{__thread} to the list in paragraph 1.
23058 With the exception of @code{__thread}, at most one
23059 @var{storage-class-specifier} shall appear in a given
23060 @var{decl-specifier-seq}. The @code{__thread} specifier may
23061 be used alone, or immediately following the @code{extern} or
23062 @code{static} specifiers. [@dots{}]
23065 Add after paragraph 5
23068 The @code{__thread} specifier can be applied only to the names of objects
23069 and to anonymous unions.
23075 Add after paragraph 6
23078 Non-@code{static} members shall not be @code{__thread}.
23082 @node Binary constants
23083 @section Binary Constants using the @samp{0b} Prefix
23084 @cindex Binary constants using the @samp{0b} prefix
23086 Integer constants can be written as binary constants, consisting of a
23087 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
23088 @samp{0B}. This is particularly useful in environments that operate a
23089 lot on the bit level (like microcontrollers).
23091 The following statements are identical:
23100 The type of these constants follows the same rules as for octal or
23101 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
23104 @node C++ Extensions
23105 @chapter Extensions to the C++ Language
23106 @cindex extensions, C++ language
23107 @cindex C++ language extensions
23109 The GNU compiler provides these extensions to the C++ language (and you
23110 can also use most of the C language extensions in your C++ programs). If you
23111 want to write code that checks whether these features are available, you can
23112 test for the GNU compiler the same way as for C programs: check for a
23113 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
23114 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
23115 Predefined Macros,cpp,The GNU C Preprocessor}).
23118 * C++ Volatiles:: What constitutes an access to a volatile object.
23119 * Restricted Pointers:: C99 restricted pointers and references.
23120 * Vague Linkage:: Where G++ puts inlines, vtables and such.
23121 * C++ Interface:: You can use a single C++ header file for both
23122 declarations and definitions.
23123 * Template Instantiation:: Methods for ensuring that exactly one copy of
23124 each needed template instantiation is emitted.
23125 * Bound member functions:: You can extract a function pointer to the
23126 method denoted by a @samp{->*} or @samp{.*} expression.
23127 * C++ Attributes:: Variable, function, and type attributes for C++ only.
23128 * Function Multiversioning:: Declaring multiple function versions.
23129 * Type Traits:: Compiler support for type traits.
23130 * C++ Concepts:: Improved support for generic programming.
23131 * Deprecated Features:: Things will disappear from G++.
23132 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
23135 @node C++ Volatiles
23136 @section When is a Volatile C++ Object Accessed?
23137 @cindex accessing volatiles
23138 @cindex volatile read
23139 @cindex volatile write
23140 @cindex volatile access
23142 The C++ standard differs from the C standard in its treatment of
23143 volatile objects. It fails to specify what constitutes a volatile
23144 access, except to say that C++ should behave in a similar manner to C
23145 with respect to volatiles, where possible. However, the different
23146 lvalueness of expressions between C and C++ complicate the behavior.
23147 G++ behaves the same as GCC for volatile access, @xref{C
23148 Extensions,,Volatiles}, for a description of GCC's behavior.
23150 The C and C++ language specifications differ when an object is
23151 accessed in a void context:
23154 volatile int *src = @var{somevalue};
23158 The C++ standard specifies that such expressions do not undergo lvalue
23159 to rvalue conversion, and that the type of the dereferenced object may
23160 be incomplete. The C++ standard does not specify explicitly that it
23161 is lvalue to rvalue conversion that is responsible for causing an
23162 access. There is reason to believe that it is, because otherwise
23163 certain simple expressions become undefined. However, because it
23164 would surprise most programmers, G++ treats dereferencing a pointer to
23165 volatile object of complete type as GCC would do for an equivalent
23166 type in C@. When the object has incomplete type, G++ issues a
23167 warning; if you wish to force an error, you must force a conversion to
23168 rvalue with, for instance, a static cast.
23170 When using a reference to volatile, G++ does not treat equivalent
23171 expressions as accesses to volatiles, but instead issues a warning that
23172 no volatile is accessed. The rationale for this is that otherwise it
23173 becomes difficult to determine where volatile access occur, and not
23174 possible to ignore the return value from functions returning volatile
23175 references. Again, if you wish to force a read, cast the reference to
23178 G++ implements the same behavior as GCC does when assigning to a
23179 volatile object---there is no reread of the assigned-to object, the
23180 assigned rvalue is reused. Note that in C++ assignment expressions
23181 are lvalues, and if used as an lvalue, the volatile object is
23182 referred to. For instance, @var{vref} refers to @var{vobj}, as
23183 expected, in the following example:
23187 volatile int &vref = vobj = @var{something};
23190 @node Restricted Pointers
23191 @section Restricting Pointer Aliasing
23192 @cindex restricted pointers
23193 @cindex restricted references
23194 @cindex restricted this pointer
23196 As with the C front end, G++ understands the C99 feature of restricted pointers,
23197 specified with the @code{__restrict__}, or @code{__restrict} type
23198 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
23199 language flag, @code{restrict} is not a keyword in C++.
23201 In addition to allowing restricted pointers, you can specify restricted
23202 references, which indicate that the reference is not aliased in the local
23206 void fn (int *__restrict__ rptr, int &__restrict__ rref)
23213 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
23214 @var{rref} refers to a (different) unaliased integer.
23216 You may also specify whether a member function's @var{this} pointer is
23217 unaliased by using @code{__restrict__} as a member function qualifier.
23220 void T::fn () __restrict__
23227 Within the body of @code{T::fn}, @var{this} has the effective
23228 definition @code{T *__restrict__ const this}. Notice that the
23229 interpretation of a @code{__restrict__} member function qualifier is
23230 different to that of @code{const} or @code{volatile} qualifier, in that it
23231 is applied to the pointer rather than the object. This is consistent with
23232 other compilers that implement restricted pointers.
23234 As with all outermost parameter qualifiers, @code{__restrict__} is
23235 ignored in function definition matching. This means you only need to
23236 specify @code{__restrict__} in a function definition, rather than
23237 in a function prototype as well.
23239 @node Vague Linkage
23240 @section Vague Linkage
23241 @cindex vague linkage
23243 There are several constructs in C++ that require space in the object
23244 file but are not clearly tied to a single translation unit. We say that
23245 these constructs have ``vague linkage''. Typically such constructs are
23246 emitted wherever they are needed, though sometimes we can be more
23250 @item Inline Functions
23251 Inline functions are typically defined in a header file which can be
23252 included in many different compilations. Hopefully they can usually be
23253 inlined, but sometimes an out-of-line copy is necessary, if the address
23254 of the function is taken or if inlining fails. In general, we emit an
23255 out-of-line copy in all translation units where one is needed. As an
23256 exception, we only emit inline virtual functions with the vtable, since
23257 it always requires a copy.
23259 Local static variables and string constants used in an inline function
23260 are also considered to have vague linkage, since they must be shared
23261 between all inlined and out-of-line instances of the function.
23265 C++ virtual functions are implemented in most compilers using a lookup
23266 table, known as a vtable. The vtable contains pointers to the virtual
23267 functions provided by a class, and each object of the class contains a
23268 pointer to its vtable (or vtables, in some multiple-inheritance
23269 situations). If the class declares any non-inline, non-pure virtual
23270 functions, the first one is chosen as the ``key method'' for the class,
23271 and the vtable is only emitted in the translation unit where the key
23274 @emph{Note:} If the chosen key method is later defined as inline, the
23275 vtable is still emitted in every translation unit that defines it.
23276 Make sure that any inline virtuals are declared inline in the class
23277 body, even if they are not defined there.
23279 @item @code{type_info} objects
23280 @cindex @code{type_info}
23282 C++ requires information about types to be written out in order to
23283 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
23284 For polymorphic classes (classes with virtual functions), the @samp{type_info}
23285 object is written out along with the vtable so that @samp{dynamic_cast}
23286 can determine the dynamic type of a class object at run time. For all
23287 other types, we write out the @samp{type_info} object when it is used: when
23288 applying @samp{typeid} to an expression, throwing an object, or
23289 referring to a type in a catch clause or exception specification.
23291 @item Template Instantiations
23292 Most everything in this section also applies to template instantiations,
23293 but there are other options as well.
23294 @xref{Template Instantiation,,Where's the Template?}.
23298 When used with GNU ld version 2.8 or later on an ELF system such as
23299 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
23300 these constructs will be discarded at link time. This is known as
23303 On targets that don't support COMDAT, but do support weak symbols, GCC
23304 uses them. This way one copy overrides all the others, but
23305 the unused copies still take up space in the executable.
23307 For targets that do not support either COMDAT or weak symbols,
23308 most entities with vague linkage are emitted as local symbols to
23309 avoid duplicate definition errors from the linker. This does not happen
23310 for local statics in inlines, however, as having multiple copies
23311 almost certainly breaks things.
23313 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
23314 another way to control placement of these constructs.
23316 @node C++ Interface
23317 @section C++ Interface and Implementation Pragmas
23319 @cindex interface and implementation headers, C++
23320 @cindex C++ interface and implementation headers
23321 @cindex pragmas, interface and implementation
23323 @code{#pragma interface} and @code{#pragma implementation} provide the
23324 user with a way of explicitly directing the compiler to emit entities
23325 with vague linkage (and debugging information) in a particular
23328 @emph{Note:} These @code{#pragma}s have been superceded as of GCC 2.7.2
23329 by COMDAT support and the ``key method'' heuristic
23330 mentioned in @ref{Vague Linkage}. Using them can actually cause your
23331 program to grow due to unnecessary out-of-line copies of inline
23335 @item #pragma interface
23336 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
23337 @kindex #pragma interface
23338 Use this directive in @emph{header files} that define object classes, to save
23339 space in most of the object files that use those classes. Normally,
23340 local copies of certain information (backup copies of inline member
23341 functions, debugging information, and the internal tables that implement
23342 virtual functions) must be kept in each object file that includes class
23343 definitions. You can use this pragma to avoid such duplication. When a
23344 header file containing @samp{#pragma interface} is included in a
23345 compilation, this auxiliary information is not generated (unless
23346 the main input source file itself uses @samp{#pragma implementation}).
23347 Instead, the object files contain references to be resolved at link
23350 The second form of this directive is useful for the case where you have
23351 multiple headers with the same name in different directories. If you
23352 use this form, you must specify the same string to @samp{#pragma
23355 @item #pragma implementation
23356 @itemx #pragma implementation "@var{objects}.h"
23357 @kindex #pragma implementation
23358 Use this pragma in a @emph{main input file}, when you want full output from
23359 included header files to be generated (and made globally visible). The
23360 included header file, in turn, should use @samp{#pragma interface}.
23361 Backup copies of inline member functions, debugging information, and the
23362 internal tables used to implement virtual functions are all generated in
23363 implementation files.
23365 @cindex implied @code{#pragma implementation}
23366 @cindex @code{#pragma implementation}, implied
23367 @cindex naming convention, implementation headers
23368 If you use @samp{#pragma implementation} with no argument, it applies to
23369 an include file with the same basename@footnote{A file's @dfn{basename}
23370 is the name stripped of all leading path information and of trailing
23371 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
23372 file. For example, in @file{allclass.cc}, giving just
23373 @samp{#pragma implementation}
23374 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
23376 Use the string argument if you want a single implementation file to
23377 include code from multiple header files. (You must also use
23378 @samp{#include} to include the header file; @samp{#pragma
23379 implementation} only specifies how to use the file---it doesn't actually
23382 There is no way to split up the contents of a single header file into
23383 multiple implementation files.
23386 @cindex inlining and C++ pragmas
23387 @cindex C++ pragmas, effect on inlining
23388 @cindex pragmas in C++, effect on inlining
23389 @samp{#pragma implementation} and @samp{#pragma interface} also have an
23390 effect on function inlining.
23392 If you define a class in a header file marked with @samp{#pragma
23393 interface}, the effect on an inline function defined in that class is
23394 similar to an explicit @code{extern} declaration---the compiler emits
23395 no code at all to define an independent version of the function. Its
23396 definition is used only for inlining with its callers.
23398 @opindex fno-implement-inlines
23399 Conversely, when you include the same header file in a main source file
23400 that declares it as @samp{#pragma implementation}, the compiler emits
23401 code for the function itself; this defines a version of the function
23402 that can be found via pointers (or by callers compiled without
23403 inlining). If all calls to the function can be inlined, you can avoid
23404 emitting the function by compiling with @option{-fno-implement-inlines}.
23405 If any calls are not inlined, you will get linker errors.
23407 @node Template Instantiation
23408 @section Where's the Template?
23409 @cindex template instantiation
23411 C++ templates were the first language feature to require more
23412 intelligence from the environment than was traditionally found on a UNIX
23413 system. Somehow the compiler and linker have to make sure that each
23414 template instance occurs exactly once in the executable if it is needed,
23415 and not at all otherwise. There are two basic approaches to this
23416 problem, which are referred to as the Borland model and the Cfront model.
23419 @item Borland model
23420 Borland C++ solved the template instantiation problem by adding the code
23421 equivalent of common blocks to their linker; the compiler emits template
23422 instances in each translation unit that uses them, and the linker
23423 collapses them together. The advantage of this model is that the linker
23424 only has to consider the object files themselves; there is no external
23425 complexity to worry about. The disadvantage is that compilation time
23426 is increased because the template code is being compiled repeatedly.
23427 Code written for this model tends to include definitions of all
23428 templates in the header file, since they must be seen to be
23432 The AT&T C++ translator, Cfront, solved the template instantiation
23433 problem by creating the notion of a template repository, an
23434 automatically maintained place where template instances are stored. A
23435 more modern version of the repository works as follows: As individual
23436 object files are built, the compiler places any template definitions and
23437 instantiations encountered in the repository. At link time, the link
23438 wrapper adds in the objects in the repository and compiles any needed
23439 instances that were not previously emitted. The advantages of this
23440 model are more optimal compilation speed and the ability to use the
23441 system linker; to implement the Borland model a compiler vendor also
23442 needs to replace the linker. The disadvantages are vastly increased
23443 complexity, and thus potential for error; for some code this can be
23444 just as transparent, but in practice it can been very difficult to build
23445 multiple programs in one directory and one program in multiple
23446 directories. Code written for this model tends to separate definitions
23447 of non-inline member templates into a separate file, which should be
23448 compiled separately.
23451 G++ implements the Borland model on targets where the linker supports it,
23452 including ELF targets (such as GNU/Linux), Mac OS X and Microsoft Windows.
23453 Otherwise G++ implements neither automatic model.
23455 You have the following options for dealing with template instantiations:
23459 Do nothing. Code written for the Borland model works fine, but
23460 each translation unit contains instances of each of the templates it
23461 uses. The duplicate instances will be discarded by the linker, but in
23462 a large program, this can lead to an unacceptable amount of code
23463 duplication in object files or shared libraries.
23465 Duplicate instances of a template can be avoided by defining an explicit
23466 instantiation in one object file, and preventing the compiler from doing
23467 implicit instantiations in any other object files by using an explicit
23468 instantiation declaration, using the @code{extern template} syntax:
23471 extern template int max (int, int);
23474 This syntax is defined in the C++ 2011 standard, but has been supported by
23475 G++ and other compilers since well before 2011.
23477 Explicit instantiations can be used for the largest or most frequently
23478 duplicated instances, without having to know exactly which other instances
23479 are used in the rest of the program. You can scatter the explicit
23480 instantiations throughout your program, perhaps putting them in the
23481 translation units where the instances are used or the translation units
23482 that define the templates themselves; you can put all of the explicit
23483 instantiations you need into one big file; or you can create small files
23490 template class Foo<int>;
23491 template ostream& operator <<
23492 (ostream&, const Foo<int>&);
23496 for each of the instances you need, and create a template instantiation
23497 library from those.
23499 This is the simplest option, but also offers flexibility and
23500 fine-grained control when necessary. It is also the most portable
23501 alternative and programs using this approach will work with most modern
23506 Compile your template-using code with @option{-frepo}. The compiler
23507 generates files with the extension @samp{.rpo} listing all of the
23508 template instantiations used in the corresponding object files that
23509 could be instantiated there; the link wrapper, @samp{collect2},
23510 then updates the @samp{.rpo} files to tell the compiler where to place
23511 those instantiations and rebuild any affected object files. The
23512 link-time overhead is negligible after the first pass, as the compiler
23513 continues to place the instantiations in the same files.
23515 This can be a suitable option for application code written for the Borland
23516 model, as it usually just works. Code written for the Cfront model
23517 needs to be modified so that the template definitions are available at
23518 one or more points of instantiation; usually this is as simple as adding
23519 @code{#include <tmethods.cc>} to the end of each template header.
23521 For library code, if you want the library to provide all of the template
23522 instantiations it needs, just try to link all of its object files
23523 together; the link will fail, but cause the instantiations to be
23524 generated as a side effect. Be warned, however, that this may cause
23525 conflicts if multiple libraries try to provide the same instantiations.
23526 For greater control, use explicit instantiation as described in the next
23530 @opindex fno-implicit-templates
23531 Compile your code with @option{-fno-implicit-templates} to disable the
23532 implicit generation of template instances, and explicitly instantiate
23533 all the ones you use. This approach requires more knowledge of exactly
23534 which instances you need than do the others, but it's less
23535 mysterious and allows greater control if you want to ensure that only
23536 the intended instances are used.
23538 If you are using Cfront-model code, you can probably get away with not
23539 using @option{-fno-implicit-templates} when compiling files that don't
23540 @samp{#include} the member template definitions.
23542 If you use one big file to do the instantiations, you may want to
23543 compile it without @option{-fno-implicit-templates} so you get all of the
23544 instances required by your explicit instantiations (but not by any
23545 other files) without having to specify them as well.
23547 In addition to forward declaration of explicit instantiations
23548 (with @code{extern}), G++ has extended the template instantiation
23549 syntax to support instantiation of the compiler support data for a
23550 template class (i.e.@: the vtable) without instantiating any of its
23551 members (with @code{inline}), and instantiation of only the static data
23552 members of a template class, without the support data or member
23553 functions (with @code{static}):
23556 inline template class Foo<int>;
23557 static template class Foo<int>;
23561 @node Bound member functions
23562 @section Extracting the Function Pointer from a Bound Pointer to Member Function
23564 @cindex pointer to member function
23565 @cindex bound pointer to member function
23567 In C++, pointer to member functions (PMFs) are implemented using a wide
23568 pointer of sorts to handle all the possible call mechanisms; the PMF
23569 needs to store information about how to adjust the @samp{this} pointer,
23570 and if the function pointed to is virtual, where to find the vtable, and
23571 where in the vtable to look for the member function. If you are using
23572 PMFs in an inner loop, you should really reconsider that decision. If
23573 that is not an option, you can extract the pointer to the function that
23574 would be called for a given object/PMF pair and call it directly inside
23575 the inner loop, to save a bit of time.
23577 Note that you still pay the penalty for the call through a
23578 function pointer; on most modern architectures, such a call defeats the
23579 branch prediction features of the CPU@. This is also true of normal
23580 virtual function calls.
23582 The syntax for this extension is
23586 extern int (A::*fp)();
23587 typedef int (*fptr)(A *);
23589 fptr p = (fptr)(a.*fp);
23592 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
23593 no object is needed to obtain the address of the function. They can be
23594 converted to function pointers directly:
23597 fptr p1 = (fptr)(&A::foo);
23600 @opindex Wno-pmf-conversions
23601 You must specify @option{-Wno-pmf-conversions} to use this extension.
23603 @node C++ Attributes
23604 @section C++-Specific Variable, Function, and Type Attributes
23606 Some attributes only make sense for C++ programs.
23609 @item abi_tag ("@var{tag}", ...)
23610 @cindex @code{abi_tag} function attribute
23611 @cindex @code{abi_tag} variable attribute
23612 @cindex @code{abi_tag} type attribute
23613 The @code{abi_tag} attribute can be applied to a function, variable, or class
23614 declaration. It modifies the mangled name of the entity to
23615 incorporate the tag name, in order to distinguish the function or
23616 class from an earlier version with a different ABI; perhaps the class
23617 has changed size, or the function has a different return type that is
23618 not encoded in the mangled name.
23620 The attribute can also be applied to an inline namespace, but does not
23621 affect the mangled name of the namespace; in this case it is only used
23622 for @option{-Wabi-tag} warnings and automatic tagging of functions and
23623 variables. Tagging inline namespaces is generally preferable to
23624 tagging individual declarations, but the latter is sometimes
23625 necessary, such as when only certain members of a class need to be
23628 The argument can be a list of strings of arbitrary length. The
23629 strings are sorted on output, so the order of the list is
23632 A redeclaration of an entity must not add new ABI tags,
23633 since doing so would change the mangled name.
23635 The ABI tags apply to a name, so all instantiations and
23636 specializations of a template have the same tags. The attribute will
23637 be ignored if applied to an explicit specialization or instantiation.
23639 The @option{-Wabi-tag} flag enables a warning about a class which does
23640 not have all the ABI tags used by its subobjects and virtual functions; for users with code
23641 that needs to coexist with an earlier ABI, using this option can help
23642 to find all affected types that need to be tagged.
23644 When a type involving an ABI tag is used as the type of a variable or
23645 return type of a function where that tag is not already present in the
23646 signature of the function, the tag is automatically applied to the
23647 variable or function. @option{-Wabi-tag} also warns about this
23648 situation; this warning can be avoided by explicitly tagging the
23649 variable or function or moving it into a tagged inline namespace.
23651 @item init_priority (@var{priority})
23652 @cindex @code{init_priority} variable attribute
23654 In Standard C++, objects defined at namespace scope are guaranteed to be
23655 initialized in an order in strict accordance with that of their definitions
23656 @emph{in a given translation unit}. No guarantee is made for initializations
23657 across translation units. However, GNU C++ allows users to control the
23658 order of initialization of objects defined at namespace scope with the
23659 @code{init_priority} attribute by specifying a relative @var{priority},
23660 a constant integral expression currently bounded between 101 and 65535
23661 inclusive. Lower numbers indicate a higher priority.
23663 In the following example, @code{A} would normally be created before
23664 @code{B}, but the @code{init_priority} attribute reverses that order:
23667 Some_Class A __attribute__ ((init_priority (2000)));
23668 Some_Class B __attribute__ ((init_priority (543)));
23672 Note that the particular values of @var{priority} do not matter; only their
23676 @cindex @code{warn_unused} type attribute
23678 For C++ types with non-trivial constructors and/or destructors it is
23679 impossible for the compiler to determine whether a variable of this
23680 type is truly unused if it is not referenced. This type attribute
23681 informs the compiler that variables of this type should be warned
23682 about if they appear to be unused, just like variables of fundamental
23685 This attribute is appropriate for types which just represent a value,
23686 such as @code{std::string}; it is not appropriate for types which
23687 control a resource, such as @code{std::lock_guard}.
23689 This attribute is also accepted in C, but it is unnecessary because C
23690 does not have constructors or destructors.
23694 @node Function Multiversioning
23695 @section Function Multiversioning
23696 @cindex function versions
23698 With the GNU C++ front end, for x86 targets, you may specify multiple
23699 versions of a function, where each function is specialized for a
23700 specific target feature. At runtime, the appropriate version of the
23701 function is automatically executed depending on the characteristics of
23702 the execution platform. Here is an example.
23705 __attribute__ ((target ("default")))
23708 // The default version of foo.
23712 __attribute__ ((target ("sse4.2")))
23715 // foo version for SSE4.2
23719 __attribute__ ((target ("arch=atom")))
23722 // foo version for the Intel ATOM processor
23726 __attribute__ ((target ("arch=amdfam10")))
23729 // foo version for the AMD Family 0x10 processors.
23736 assert ((*p) () == foo ());
23741 In the above example, four versions of function foo are created. The
23742 first version of foo with the target attribute "default" is the default
23743 version. This version gets executed when no other target specific
23744 version qualifies for execution on a particular platform. A new version
23745 of foo is created by using the same function signature but with a
23746 different target string. Function foo is called or a pointer to it is
23747 taken just like a regular function. GCC takes care of doing the
23748 dispatching to call the right version at runtime. Refer to the
23749 @uref{http://gcc.gnu.org/wiki/FunctionMultiVersioning, GCC wiki on
23750 Function Multiversioning} for more details.
23753 @section Type Traits
23755 The C++ front end implements syntactic extensions that allow
23756 compile-time determination of
23757 various characteristics of a type (or of a
23761 @item __has_nothrow_assign (type)
23762 If @code{type} is const qualified or is a reference type then the trait is
23763 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
23764 is true, else if @code{type} is a cv class or union type with copy assignment
23765 operators that are known not to throw an exception then the trait is true,
23766 else it is false. Requires: @code{type} shall be a complete type,
23767 (possibly cv-qualified) @code{void}, or an array of unknown bound.
23769 @item __has_nothrow_copy (type)
23770 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
23771 @code{type} is a cv class or union type with copy constructors that
23772 are known not to throw an exception then the trait is true, else it is false.
23773 Requires: @code{type} shall be a complete type, (possibly cv-qualified)
23774 @code{void}, or an array of unknown bound.
23776 @item __has_nothrow_constructor (type)
23777 If @code{__has_trivial_constructor (type)} is true then the trait is
23778 true, else if @code{type} is a cv class or union type (or array
23779 thereof) with a default constructor that is known not to throw an
23780 exception then the trait is true, else it is false. Requires:
23781 @code{type} shall be a complete type, (possibly cv-qualified)
23782 @code{void}, or an array of unknown bound.
23784 @item __has_trivial_assign (type)
23785 If @code{type} is const qualified or is a reference type then the trait is
23786 false. Otherwise if @code{__is_pod (type)} is true then the trait is
23787 true, else if @code{type} is a cv class or union type with a trivial
23788 copy assignment ([class.copy]) then the trait is true, else it is
23789 false. Requires: @code{type} shall be a complete type, (possibly
23790 cv-qualified) @code{void}, or an array of unknown bound.
23792 @item __has_trivial_copy (type)
23793 If @code{__is_pod (type)} is true or @code{type} is a reference type
23794 then the trait is true, else if @code{type} is a cv class or union type
23795 with a trivial copy constructor ([class.copy]) then the trait
23796 is true, else it is false. Requires: @code{type} shall be a complete
23797 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23799 @item __has_trivial_constructor (type)
23800 If @code{__is_pod (type)} is true then the trait is true, else if
23801 @code{type} is a cv class or union type (or array thereof) with a
23802 trivial default constructor ([class.ctor]) then the trait is true,
23803 else it is false. Requires: @code{type} shall be a complete
23804 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23806 @item __has_trivial_destructor (type)
23807 If @code{__is_pod (type)} is true or @code{type} is a reference type then
23808 the trait is true, else if @code{type} is a cv class or union type (or
23809 array thereof) with a trivial destructor ([class.dtor]) then the trait
23810 is true, else it is false. Requires: @code{type} shall be a complete
23811 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23813 @item __has_virtual_destructor (type)
23814 If @code{type} is a class type with a virtual destructor
23815 ([class.dtor]) then the trait is true, else it is false. Requires:
23816 @code{type} shall be a complete type, (possibly cv-qualified)
23817 @code{void}, or an array of unknown bound.
23819 @item __is_abstract (type)
23820 If @code{type} is an abstract class ([class.abstract]) then the trait
23821 is true, else it is false. Requires: @code{type} shall be a complete
23822 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23824 @item __is_base_of (base_type, derived_type)
23825 If @code{base_type} is a base class of @code{derived_type}
23826 ([class.derived]) then the trait is true, otherwise it is false.
23827 Top-level cv qualifications of @code{base_type} and
23828 @code{derived_type} are ignored. For the purposes of this trait, a
23829 class type is considered is own base. Requires: if @code{__is_class
23830 (base_type)} and @code{__is_class (derived_type)} are true and
23831 @code{base_type} and @code{derived_type} are not the same type
23832 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
23833 type. A diagnostic is produced if this requirement is not met.
23835 @item __is_class (type)
23836 If @code{type} is a cv class type, and not a union type
23837 ([basic.compound]) the trait is true, else it is false.
23839 @item __is_empty (type)
23840 If @code{__is_class (type)} is false then the trait is false.
23841 Otherwise @code{type} is considered empty if and only if: @code{type}
23842 has no non-static data members, or all non-static data members, if
23843 any, are bit-fields of length 0, and @code{type} has no virtual
23844 members, and @code{type} has no virtual base classes, and @code{type}
23845 has no base classes @code{base_type} for which
23846 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
23847 be a complete type, (possibly cv-qualified) @code{void}, or an array
23850 @item __is_enum (type)
23851 If @code{type} is a cv enumeration type ([basic.compound]) the trait is
23852 true, else it is false.
23854 @item __is_literal_type (type)
23855 If @code{type} is a literal type ([basic.types]) the trait is
23856 true, else it is false. Requires: @code{type} shall be a complete type,
23857 (possibly cv-qualified) @code{void}, or an array of unknown bound.
23859 @item __is_pod (type)
23860 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
23861 else it is false. Requires: @code{type} shall be a complete type,
23862 (possibly cv-qualified) @code{void}, or an array of unknown bound.
23864 @item __is_polymorphic (type)
23865 If @code{type} is a polymorphic class ([class.virtual]) then the trait
23866 is true, else it is false. Requires: @code{type} shall be a complete
23867 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23869 @item __is_standard_layout (type)
23870 If @code{type} is a standard-layout type ([basic.types]) the trait is
23871 true, else it is false. Requires: @code{type} shall be a complete
23872 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23874 @item __is_trivial (type)
23875 If @code{type} is a trivial type ([basic.types]) the trait is
23876 true, else it is false. Requires: @code{type} shall be a complete
23877 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
23879 @item __is_union (type)
23880 If @code{type} is a cv union type ([basic.compound]) the trait is
23881 true, else it is false.
23883 @item __underlying_type (type)
23884 The underlying type of @code{type}. Requires: @code{type} shall be
23885 an enumeration type ([dcl.enum]).
23887 @item __integer_pack (length)
23888 When used as the pattern of a pack expansion within a template
23889 definition, expands to a template argument pack containing integers
23890 from @code{0} to @code{length-1}. This is provided for efficient
23891 implementation of @code{std::make_integer_sequence}.
23897 @section C++ Concepts
23899 C++ concepts provide much-improved support for generic programming. In
23900 particular, they allow the specification of constraints on template arguments.
23901 The constraints are used to extend the usual overloading and partial
23902 specialization capabilities of the language, allowing generic data structures
23903 and algorithms to be ``refined'' based on their properties rather than their
23906 The following keywords are reserved for concepts.
23910 States an expression as an assumption, and if possible, verifies that the
23911 assumption is valid. For example, @code{assume(n > 0)}.
23914 Introduces an axiom definition. Axioms introduce requirements on values.
23917 Introduces a universally quantified object in an axiom. For example,
23918 @code{forall (int n) n + 0 == n}).
23921 Introduces a concept definition. Concepts are sets of syntactic and semantic
23922 requirements on types and their values.
23925 Introduces constraints on template arguments or requirements for a member
23926 function of a class template.
23930 The front end also exposes a number of internal mechanism that can be used
23931 to simplify the writing of type traits. Note that some of these traits are
23932 likely to be removed in the future.
23935 @item __is_same (type1, type2)
23936 A binary type trait: true whenever the type arguments are the same.
23941 @node Deprecated Features
23942 @section Deprecated Features
23944 In the past, the GNU C++ compiler was extended to experiment with new
23945 features, at a time when the C++ language was still evolving. Now that
23946 the C++ standard is complete, some of those features are superseded by
23947 superior alternatives. Using the old features might cause a warning in
23948 some cases that the feature will be dropped in the future. In other
23949 cases, the feature might be gone already.
23951 G++ allows a virtual function returning @samp{void *} to be overridden
23952 by one returning a different pointer type. This extension to the
23953 covariant return type rules is now deprecated and will be removed from a
23956 The use of default arguments in function pointers, function typedefs
23957 and other places where they are not permitted by the standard is
23958 deprecated and will be removed from a future version of G++.
23960 G++ allows floating-point literals to appear in integral constant expressions,
23961 e.g.@: @samp{ enum E @{ e = int(2.2 * 3.7) @} }
23962 This extension is deprecated and will be removed from a future version.
23964 G++ allows static data members of const floating-point type to be declared
23965 with an initializer in a class definition. The standard only allows
23966 initializers for static members of const integral types and const
23967 enumeration types so this extension has been deprecated and will be removed
23968 from a future version.
23970 G++ allows attributes to follow a parenthesized direct initializer,
23971 e.g.@: @samp{ int f (0) __attribute__ ((something)); } This extension
23972 has been ignored since G++ 3.3 and is deprecated.
23974 G++ allows anonymous structs and unions to have members that are not
23975 public non-static data members (i.e.@: fields). These extensions are
23978 @node Backwards Compatibility
23979 @section Backwards Compatibility
23980 @cindex Backwards Compatibility
23981 @cindex ARM [Annotated C++ Reference Manual]
23983 Now that there is a definitive ISO standard C++, G++ has a specification
23984 to adhere to. The C++ language evolved over time, and features that
23985 used to be acceptable in previous drafts of the standard, such as the ARM
23986 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
23987 compilation of C++ written to such drafts, G++ contains some backwards
23988 compatibilities. @emph{All such backwards compatibility features are
23989 liable to disappear in future versions of G++.} They should be considered
23990 deprecated. @xref{Deprecated Features}.
23994 @item Implicit C language
23995 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
23996 scope to set the language. On such systems, all header files are
23997 implicitly scoped inside a C language scope. Also, an empty prototype
23998 @code{()} is treated as an unspecified number of arguments, rather
23999 than no arguments, as C++ demands.
24003 @c LocalWords: emph deftypefn builtin ARCv2EM SIMD builtins msimd
24004 @c LocalWords: typedef v4si v8hi DMA dma vdiwr vdowr