1 @c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1996, 1998, 1999, 2000,
2 @c 2001, 2002, 2003, 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
4 @c This is part of the GCC manual.
5 @c For copying conditions, see the file gcc.texi.
8 @chapter Extensions to the C Language Family
9 @cindex extensions, C language
10 @cindex C language extensions
13 GNU C provides several language features not found in ISO standard C@.
14 (The @option{-pedantic} option directs GCC to print a warning message if
15 any of these features is used.) To test for the availability of these
16 features in conditional compilation, check for a predefined macro
17 @code{__GNUC__}, which is always defined under GCC@.
19 These extensions are available in C and Objective-C@. Most of them are
20 also available in C++. @xref{C++ Extensions,,Extensions to the
21 C++ Language}, for extensions that apply @emph{only} to C++.
23 Some features that are in ISO C99 but not C89 or C++ are also, as
24 extensions, accepted by GCC in C89 mode and in C++.
27 * Statement Exprs:: Putting statements and declarations inside expressions.
28 * Local Labels:: Labels local to a block.
29 * Labels as Values:: Getting pointers to labels, and computed gotos.
30 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
31 * Constructing Calls:: Dispatching a call to another function.
32 * Typeof:: @code{typeof}: referring to the type of an expression.
33 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
34 * Long Long:: Double-word integers---@code{long long int}.
35 * Complex:: Data types for complex numbers.
36 * Floating Types:: Additional Floating Types.
37 * Decimal Float:: Decimal Floating Types.
38 * Hex Floats:: Hexadecimal floating-point constants.
39 * Zero Length:: Zero-length arrays.
40 * Variable Length:: Arrays whose length is computed at run time.
41 * Empty Structures:: Structures with no members.
42 * Variadic Macros:: Macros with a variable number of arguments.
43 * Escaped Newlines:: Slightly looser rules for escaped newlines.
44 * Subscripting:: Any array can be subscripted, even if not an lvalue.
45 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
46 * Initializers:: Non-constant initializers.
47 * Compound Literals:: Compound literals give structures, unions
49 * Designated Inits:: Labeling elements of initializers.
50 * Cast to Union:: Casting to union type from any member of the union.
51 * Case Ranges:: `case 1 ... 9' and such.
52 * Mixed Declarations:: Mixing declarations and code.
53 * Function Attributes:: Declaring that functions have no side effects,
54 or that they can never return.
55 * Attribute Syntax:: Formal syntax for attributes.
56 * Function Prototypes:: Prototype declarations and old-style definitions.
57 * C++ Comments:: C++ comments are recognized.
58 * Dollar Signs:: Dollar sign is allowed in identifiers.
59 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
60 * Variable Attributes:: Specifying attributes of variables.
61 * Type Attributes:: Specifying attributes of types.
62 * Alignment:: Inquiring about the alignment of a type or variable.
63 * Inline:: Defining inline functions (as fast as macros).
64 * Extended Asm:: Assembler instructions with C expressions as operands.
65 (With them you can define ``built-in'' functions.)
66 * Constraints:: Constraints for asm operands
67 * Asm Labels:: Specifying the assembler name to use for a C symbol.
68 * Explicit Reg Vars:: Defining variables residing in specified registers.
69 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
70 * Incomplete Enums:: @code{enum foo;}, with details to follow.
71 * Function Names:: Printable strings which are the name of the current
73 * Return Address:: Getting the return or frame address of a function.
74 * Vector Extensions:: Using vector instructions through built-in functions.
75 * Offsetof:: Special syntax for implementing @code{offsetof}.
76 * Atomic Builtins:: Built-in functions for atomic memory access.
77 * Object Size Checking:: Built-in functions for limited buffer overflow
79 * Other Builtins:: Other built-in functions.
80 * Target Builtins:: Built-in functions specific to particular targets.
81 * Target Format Checks:: Format checks specific to particular targets.
82 * Pragmas:: Pragmas accepted by GCC.
83 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
84 * Thread-Local:: Per-thread variables.
85 * Binary constants:: Binary constants using the @samp{0b} prefix.
89 @section Statements and Declarations in Expressions
90 @cindex statements inside expressions
91 @cindex declarations inside expressions
92 @cindex expressions containing statements
93 @cindex macros, statements in expressions
95 @c the above section title wrapped and causes an underfull hbox.. i
96 @c changed it from "within" to "in". --mew 4feb93
97 A compound statement enclosed in parentheses may appear as an expression
98 in GNU C@. This allows you to use loops, switches, and local variables
101 Recall that a compound statement is a sequence of statements surrounded
102 by braces; in this construct, parentheses go around the braces. For
106 (@{ int y = foo (); int z;
113 is a valid (though slightly more complex than necessary) expression
114 for the absolute value of @code{foo ()}.
116 The last thing in the compound statement should be an expression
117 followed by a semicolon; the value of this subexpression serves as the
118 value of the entire construct. (If you use some other kind of statement
119 last within the braces, the construct has type @code{void}, and thus
120 effectively no value.)
122 This feature is especially useful in making macro definitions ``safe'' (so
123 that they evaluate each operand exactly once). For example, the
124 ``maximum'' function is commonly defined as a macro in standard C as
128 #define max(a,b) ((a) > (b) ? (a) : (b))
132 @cindex side effects, macro argument
133 But this definition computes either @var{a} or @var{b} twice, with bad
134 results if the operand has side effects. In GNU C, if you know the
135 type of the operands (here taken as @code{int}), you can define
136 the macro safely as follows:
139 #define maxint(a,b) \
140 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
143 Embedded statements are not allowed in constant expressions, such as
144 the value of an enumeration constant, the width of a bit-field, or
145 the initial value of a static variable.
147 If you don't know the type of the operand, you can still do this, but you
148 must use @code{typeof} (@pxref{Typeof}).
150 In G++, the result value of a statement expression undergoes array and
151 function pointer decay, and is returned by value to the enclosing
152 expression. For instance, if @code{A} is a class, then
161 will construct a temporary @code{A} object to hold the result of the
162 statement expression, and that will be used to invoke @code{Foo}.
163 Therefore the @code{this} pointer observed by @code{Foo} will not be the
166 Any temporaries created within a statement within a statement expression
167 will be destroyed at the statement's end. This makes statement
168 expressions inside macros slightly different from function calls. In
169 the latter case temporaries introduced during argument evaluation will
170 be destroyed at the end of the statement that includes the function
171 call. In the statement expression case they will be destroyed during
172 the statement expression. For instance,
175 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
176 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
186 will have different places where temporaries are destroyed. For the
187 @code{macro} case, the temporary @code{X} will be destroyed just after
188 the initialization of @code{b}. In the @code{function} case that
189 temporary will be destroyed when the function returns.
191 These considerations mean that it is probably a bad idea to use
192 statement-expressions of this form in header files that are designed to
193 work with C++. (Note that some versions of the GNU C Library contained
194 header files using statement-expression that lead to precisely this
197 Jumping into a statement expression with @code{goto} or using a
198 @code{switch} statement outside the statement expression with a
199 @code{case} or @code{default} label inside the statement expression is
200 not permitted. Jumping into a statement expression with a computed
201 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
202 Jumping out of a statement expression is permitted, but if the
203 statement expression is part of a larger expression then it is
204 unspecified which other subexpressions of that expression have been
205 evaluated except where the language definition requires certain
206 subexpressions to be evaluated before or after the statement
207 expression. In any case, as with a function call the evaluation of a
208 statement expression is not interleaved with the evaluation of other
209 parts of the containing expression. For example,
212 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
216 will call @code{foo} and @code{bar1} and will not call @code{baz} but
217 may or may not call @code{bar2}. If @code{bar2} is called, it will be
218 called after @code{foo} and before @code{bar1}
221 @section Locally Declared Labels
223 @cindex macros, local labels
225 GCC allows you to declare @dfn{local labels} in any nested block
226 scope. A local label is just like an ordinary label, but you can
227 only reference it (with a @code{goto} statement, or by taking its
228 address) within the block in which it was declared.
230 A local label declaration looks like this:
233 __label__ @var{label};
240 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
243 Local label declarations must come at the beginning of the block,
244 before any ordinary declarations or statements.
246 The label declaration defines the label @emph{name}, but does not define
247 the label itself. You must do this in the usual way, with
248 @code{@var{label}:}, within the statements of the statement expression.
250 The local label feature is useful for complex macros. If a macro
251 contains nested loops, a @code{goto} can be useful for breaking out of
252 them. However, an ordinary label whose scope is the whole function
253 cannot be used: if the macro can be expanded several times in one
254 function, the label will be multiply defined in that function. A
255 local label avoids this problem. For example:
258 #define SEARCH(value, array, target) \
261 typeof (target) _SEARCH_target = (target); \
262 typeof (*(array)) *_SEARCH_array = (array); \
265 for (i = 0; i < max; i++) \
266 for (j = 0; j < max; j++) \
267 if (_SEARCH_array[i][j] == _SEARCH_target) \
268 @{ (value) = i; goto found; @} \
274 This could also be written using a statement-expression:
277 #define SEARCH(array, target) \
280 typeof (target) _SEARCH_target = (target); \
281 typeof (*(array)) *_SEARCH_array = (array); \
284 for (i = 0; i < max; i++) \
285 for (j = 0; j < max; j++) \
286 if (_SEARCH_array[i][j] == _SEARCH_target) \
287 @{ value = i; goto found; @} \
294 Local label declarations also make the labels they declare visible to
295 nested functions, if there are any. @xref{Nested Functions}, for details.
297 @node Labels as Values
298 @section Labels as Values
299 @cindex labels as values
300 @cindex computed gotos
301 @cindex goto with computed label
302 @cindex address of a label
304 You can get the address of a label defined in the current function
305 (or a containing function) with the unary operator @samp{&&}. The
306 value has type @code{void *}. This value is a constant and can be used
307 wherever a constant of that type is valid. For example:
315 To use these values, you need to be able to jump to one. This is done
316 with the computed goto statement@footnote{The analogous feature in
317 Fortran is called an assigned goto, but that name seems inappropriate in
318 C, where one can do more than simply store label addresses in label
319 variables.}, @code{goto *@var{exp};}. For example,
326 Any expression of type @code{void *} is allowed.
328 One way of using these constants is in initializing a static array that
329 will serve as a jump table:
332 static void *array[] = @{ &&foo, &&bar, &&hack @};
335 Then you can select a label with indexing, like this:
342 Note that this does not check whether the subscript is in bounds---array
343 indexing in C never does that.
345 Such an array of label values serves a purpose much like that of the
346 @code{switch} statement. The @code{switch} statement is cleaner, so
347 use that rather than an array unless the problem does not fit a
348 @code{switch} statement very well.
350 Another use of label values is in an interpreter for threaded code.
351 The labels within the interpreter function can be stored in the
352 threaded code for super-fast dispatching.
354 You may not use this mechanism to jump to code in a different function.
355 If you do that, totally unpredictable things will happen. The best way to
356 avoid this is to store the label address only in automatic variables and
357 never pass it as an argument.
359 An alternate way to write the above example is
362 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
364 goto *(&&foo + array[i]);
368 This is more friendly to code living in shared libraries, as it reduces
369 the number of dynamic relocations that are needed, and by consequence,
370 allows the data to be read-only.
372 @node Nested Functions
373 @section Nested Functions
374 @cindex nested functions
375 @cindex downward funargs
378 A @dfn{nested function} is a function defined inside another function.
379 (Nested functions are not supported for GNU C++.) The nested function's
380 name is local to the block where it is defined. For example, here we
381 define a nested function named @code{square}, and call it twice:
385 foo (double a, double b)
387 double square (double z) @{ return z * z; @}
389 return square (a) + square (b);
394 The nested function can access all the variables of the containing
395 function that are visible at the point of its definition. This is
396 called @dfn{lexical scoping}. For example, here we show a nested
397 function which uses an inherited variable named @code{offset}:
401 bar (int *array, int offset, int size)
403 int access (int *array, int index)
404 @{ return array[index + offset]; @}
407 for (i = 0; i < size; i++)
408 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
413 Nested function definitions are permitted within functions in the places
414 where variable definitions are allowed; that is, in any block, mixed
415 with the other declarations and statements in the block.
417 It is possible to call the nested function from outside the scope of its
418 name by storing its address or passing the address to another function:
421 hack (int *array, int size)
423 void store (int index, int value)
424 @{ array[index] = value; @}
426 intermediate (store, size);
430 Here, the function @code{intermediate} receives the address of
431 @code{store} as an argument. If @code{intermediate} calls @code{store},
432 the arguments given to @code{store} are used to store into @code{array}.
433 But this technique works only so long as the containing function
434 (@code{hack}, in this example) does not exit.
436 If you try to call the nested function through its address after the
437 containing function has exited, all hell will break loose. If you try
438 to call it after a containing scope level has exited, and if it refers
439 to some of the variables that are no longer in scope, you may be lucky,
440 but it's not wise to take the risk. If, however, the nested function
441 does not refer to anything that has gone out of scope, you should be
444 GCC implements taking the address of a nested function using a technique
445 called @dfn{trampolines}. A paper describing them is available as
448 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
450 A nested function can jump to a label inherited from a containing
451 function, provided the label was explicitly declared in the containing
452 function (@pxref{Local Labels}). Such a jump returns instantly to the
453 containing function, exiting the nested function which did the
454 @code{goto} and any intermediate functions as well. Here is an example:
458 bar (int *array, int offset, int size)
461 int access (int *array, int index)
465 return array[index + offset];
469 for (i = 0; i < size; i++)
470 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
474 /* @r{Control comes here from @code{access}
475 if it detects an error.} */
482 A nested function always has no linkage. Declaring one with
483 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
484 before its definition, use @code{auto} (which is otherwise meaningless
485 for function declarations).
488 bar (int *array, int offset, int size)
491 auto int access (int *, int);
493 int access (int *array, int index)
497 return array[index + offset];
503 @node Constructing Calls
504 @section Constructing Function Calls
505 @cindex constructing calls
506 @cindex forwarding calls
508 Using the built-in functions described below, you can record
509 the arguments a function received, and call another function
510 with the same arguments, without knowing the number or types
513 You can also record the return value of that function call,
514 and later return that value, without knowing what data type
515 the function tried to return (as long as your caller expects
518 However, these built-in functions may interact badly with some
519 sophisticated features or other extensions of the language. It
520 is, therefore, not recommended to use them outside very simple
521 functions acting as mere forwarders for their arguments.
523 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
524 This built-in function returns a pointer to data
525 describing how to perform a call with the same arguments as were passed
526 to the current function.
528 The function saves the arg pointer register, structure value address,
529 and all registers that might be used to pass arguments to a function
530 into a block of memory allocated on the stack. Then it returns the
531 address of that block.
534 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
535 This built-in function invokes @var{function}
536 with a copy of the parameters described by @var{arguments}
539 The value of @var{arguments} should be the value returned by
540 @code{__builtin_apply_args}. The argument @var{size} specifies the size
541 of the stack argument data, in bytes.
543 This function returns a pointer to data describing
544 how to return whatever value was returned by @var{function}. The data
545 is saved in a block of memory allocated on the stack.
547 It is not always simple to compute the proper value for @var{size}. The
548 value is used by @code{__builtin_apply} to compute the amount of data
549 that should be pushed on the stack and copied from the incoming argument
553 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
554 This built-in function returns the value described by @var{result} from
555 the containing function. You should specify, for @var{result}, a value
556 returned by @code{__builtin_apply}.
560 @section Referring to a Type with @code{typeof}
563 @cindex macros, types of arguments
565 Another way to refer to the type of an expression is with @code{typeof}.
566 The syntax of using of this keyword looks like @code{sizeof}, but the
567 construct acts semantically like a type name defined with @code{typedef}.
569 There are two ways of writing the argument to @code{typeof}: with an
570 expression or with a type. Here is an example with an expression:
577 This assumes that @code{x} is an array of pointers to functions;
578 the type described is that of the values of the functions.
580 Here is an example with a typename as the argument:
587 Here the type described is that of pointers to @code{int}.
589 If you are writing a header file that must work when included in ISO C
590 programs, write @code{__typeof__} instead of @code{typeof}.
591 @xref{Alternate Keywords}.
593 A @code{typeof}-construct can be used anywhere a typedef name could be
594 used. For example, you can use it in a declaration, in a cast, or inside
595 of @code{sizeof} or @code{typeof}.
597 @code{typeof} is often useful in conjunction with the
598 statements-within-expressions feature. Here is how the two together can
599 be used to define a safe ``maximum'' macro that operates on any
600 arithmetic type and evaluates each of its arguments exactly once:
604 (@{ typeof (a) _a = (a); \
605 typeof (b) _b = (b); \
606 _a > _b ? _a : _b; @})
609 @cindex underscores in variables in macros
610 @cindex @samp{_} in variables in macros
611 @cindex local variables in macros
612 @cindex variables, local, in macros
613 @cindex macros, local variables in
615 The reason for using names that start with underscores for the local
616 variables is to avoid conflicts with variable names that occur within the
617 expressions that are substituted for @code{a} and @code{b}. Eventually we
618 hope to design a new form of declaration syntax that allows you to declare
619 variables whose scopes start only after their initializers; this will be a
620 more reliable way to prevent such conflicts.
623 Some more examples of the use of @code{typeof}:
627 This declares @code{y} with the type of what @code{x} points to.
634 This declares @code{y} as an array of such values.
641 This declares @code{y} as an array of pointers to characters:
644 typeof (typeof (char *)[4]) y;
648 It is equivalent to the following traditional C declaration:
654 To see the meaning of the declaration using @code{typeof}, and why it
655 might be a useful way to write, rewrite it with these macros:
658 #define pointer(T) typeof(T *)
659 #define array(T, N) typeof(T [N])
663 Now the declaration can be rewritten this way:
666 array (pointer (char), 4) y;
670 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
671 pointers to @code{char}.
674 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
675 a more limited extension which permitted one to write
678 typedef @var{T} = @var{expr};
682 with the effect of declaring @var{T} to have the type of the expression
683 @var{expr}. This extension does not work with GCC 3 (versions between
684 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
685 relies on it should be rewritten to use @code{typeof}:
688 typedef typeof(@var{expr}) @var{T};
692 This will work with all versions of GCC@.
695 @section Conditionals with Omitted Operands
696 @cindex conditional expressions, extensions
697 @cindex omitted middle-operands
698 @cindex middle-operands, omitted
699 @cindex extensions, @code{?:}
700 @cindex @code{?:} extensions
702 The middle operand in a conditional expression may be omitted. Then
703 if the first operand is nonzero, its value is the value of the conditional
706 Therefore, the expression
713 has the value of @code{x} if that is nonzero; otherwise, the value of
716 This example is perfectly equivalent to
722 @cindex side effect in ?:
723 @cindex ?: side effect
725 In this simple case, the ability to omit the middle operand is not
726 especially useful. When it becomes useful is when the first operand does,
727 or may (if it is a macro argument), contain a side effect. Then repeating
728 the operand in the middle would perform the side effect twice. Omitting
729 the middle operand uses the value already computed without the undesirable
730 effects of recomputing it.
733 @section Double-Word Integers
734 @cindex @code{long long} data types
735 @cindex double-word arithmetic
736 @cindex multiprecision arithmetic
737 @cindex @code{LL} integer suffix
738 @cindex @code{ULL} integer suffix
740 ISO C99 supports data types for integers that are at least 64 bits wide,
741 and as an extension GCC supports them in C89 mode and in C++.
742 Simply write @code{long long int} for a signed integer, or
743 @code{unsigned long long int} for an unsigned integer. To make an
744 integer constant of type @code{long long int}, add the suffix @samp{LL}
745 to the integer. To make an integer constant of type @code{unsigned long
746 long int}, add the suffix @samp{ULL} to the integer.
748 You can use these types in arithmetic like any other integer types.
749 Addition, subtraction, and bitwise boolean operations on these types
750 are open-coded on all types of machines. Multiplication is open-coded
751 if the machine supports fullword-to-doubleword a widening multiply
752 instruction. Division and shifts are open-coded only on machines that
753 provide special support. The operations that are not open-coded use
754 special library routines that come with GCC@.
756 There may be pitfalls when you use @code{long long} types for function
757 arguments, unless you declare function prototypes. If a function
758 expects type @code{int} for its argument, and you pass a value of type
759 @code{long long int}, confusion will result because the caller and the
760 subroutine will disagree about the number of bytes for the argument.
761 Likewise, if the function expects @code{long long int} and you pass
762 @code{int}. The best way to avoid such problems is to use prototypes.
765 @section Complex Numbers
766 @cindex complex numbers
767 @cindex @code{_Complex} keyword
768 @cindex @code{__complex__} keyword
770 ISO C99 supports complex floating data types, and as an extension GCC
771 supports them in C89 mode and in C++, and supports complex integer data
772 types which are not part of ISO C99. You can declare complex types
773 using the keyword @code{_Complex}. As an extension, the older GNU
774 keyword @code{__complex__} is also supported.
776 For example, @samp{_Complex double x;} declares @code{x} as a
777 variable whose real part and imaginary part are both of type
778 @code{double}. @samp{_Complex short int y;} declares @code{y} to
779 have real and imaginary parts of type @code{short int}; this is not
780 likely to be useful, but it shows that the set of complex types is
783 To write a constant with a complex data type, use the suffix @samp{i} or
784 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
785 has type @code{_Complex float} and @code{3i} has type
786 @code{_Complex int}. Such a constant always has a pure imaginary
787 value, but you can form any complex value you like by adding one to a
788 real constant. This is a GNU extension; if you have an ISO C99
789 conforming C library (such as GNU libc), and want to construct complex
790 constants of floating type, you should include @code{<complex.h>} and
791 use the macros @code{I} or @code{_Complex_I} instead.
793 @cindex @code{__real__} keyword
794 @cindex @code{__imag__} keyword
795 To extract the real part of a complex-valued expression @var{exp}, write
796 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
797 extract the imaginary part. This is a GNU extension; for values of
798 floating type, you should use the ISO C99 functions @code{crealf},
799 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
800 @code{cimagl}, declared in @code{<complex.h>} and also provided as
801 built-in functions by GCC@.
803 @cindex complex conjugation
804 The operator @samp{~} performs complex conjugation when used on a value
805 with a complex type. This is a GNU extension; for values of
806 floating type, you should use the ISO C99 functions @code{conjf},
807 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
808 provided as built-in functions by GCC@.
810 GCC can allocate complex automatic variables in a noncontiguous
811 fashion; it's even possible for the real part to be in a register while
812 the imaginary part is on the stack (or vice-versa). Only the DWARF2
813 debug info format can represent this, so use of DWARF2 is recommended.
814 If you are using the stabs debug info format, GCC describes a noncontiguous
815 complex variable as if it were two separate variables of noncomplex type.
816 If the variable's actual name is @code{foo}, the two fictitious
817 variables are named @code{foo$real} and @code{foo$imag}. You can
818 examine and set these two fictitious variables with your debugger.
821 @section Additional Floating Types
822 @cindex additional floating types
823 @cindex @code{__float80} data type
824 @cindex @code{__float128} data type
825 @cindex @code{w} floating point suffix
826 @cindex @code{q} floating point suffix
827 @cindex @code{W} floating point suffix
828 @cindex @code{Q} floating point suffix
830 As an extension, the GNU C compiler supports additional floating
831 types, @code{__float80} and @code{__float128} to support 80bit
832 (@code{XFmode}) and 128 bit (@code{TFmode}) floating types.
833 Support for additional types includes the arithmetic operators:
834 add, subtract, multiply, divide; unary arithmetic operators;
835 relational operators; equality operators; and conversions to and from
836 integer and other floating types. Use a suffix @samp{w} or @samp{W}
837 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
838 for @code{_float128}. You can declare complex types using the
839 corresponding internal complex type, @code{XCmode} for @code{__float80}
840 type and @code{TCmode} for @code{__float128} type:
843 typedef _Complex float __attribute__((mode(TC))) _Complex128;
844 typedef _Complex float __attribute__((mode(XC))) _Complex80;
847 Not all targets support additional floating point types. @code{__float80}
848 is supported on i386, x86_64 and ia64 targets and target @code{__float128}
849 is supported on x86_64 and ia64 targets.
852 @section Decimal Floating Types
853 @cindex decimal floating types
854 @cindex @code{_Decimal32} data type
855 @cindex @code{_Decimal64} data type
856 @cindex @code{_Decimal128} data type
857 @cindex @code{df} integer suffix
858 @cindex @code{dd} integer suffix
859 @cindex @code{dl} integer suffix
860 @cindex @code{DF} integer suffix
861 @cindex @code{DD} integer suffix
862 @cindex @code{DL} integer suffix
864 As an extension, the GNU C compiler supports decimal floating types as
865 defined in the N1176 draft of ISO/IEC WDTR24732. Support for decimal
866 floating types in GCC will evolve as the draft technical report changes.
867 Calling conventions for any target might also change. Not all targets
868 support decimal floating types.
870 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
871 @code{_Decimal128}. They use a radix of ten, unlike the floating types
872 @code{float}, @code{double}, and @code{long double} whose radix is not
873 specified by the C standard but is usually two.
875 Support for decimal floating types includes the arithmetic operators
876 add, subtract, multiply, divide; unary arithmetic operators;
877 relational operators; equality operators; and conversions to and from
878 integer and other floating types. Use a suffix @samp{df} or
879 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
880 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
883 GCC support of decimal float as specified by the draft technical report
888 Translation time data type (TTDT) is not supported.
891 When the value of a decimal floating type cannot be represented in the
892 integer type to which it is being converted, the result is undefined
893 rather than the result value specified by the draft technical report.
896 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
897 are supported by the DWARF2 debug information format.
903 ISO C99 supports floating-point numbers written not only in the usual
904 decimal notation, such as @code{1.55e1}, but also numbers such as
905 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
906 supports this in C89 mode (except in some cases when strictly
907 conforming) and in C++. In that format the
908 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
909 mandatory. The exponent is a decimal number that indicates the power of
910 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
917 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
918 is the same as @code{1.55e1}.
920 Unlike for floating-point numbers in the decimal notation the exponent
921 is always required in the hexadecimal notation. Otherwise the compiler
922 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
923 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
924 extension for floating-point constants of type @code{float}.
927 @section Arrays of Length Zero
928 @cindex arrays of length zero
929 @cindex zero-length arrays
930 @cindex length-zero arrays
931 @cindex flexible array members
933 Zero-length arrays are allowed in GNU C@. They are very useful as the
934 last element of a structure which is really a header for a variable-length
943 struct line *thisline = (struct line *)
944 malloc (sizeof (struct line) + this_length);
945 thisline->length = this_length;
948 In ISO C90, you would have to give @code{contents} a length of 1, which
949 means either you waste space or complicate the argument to @code{malloc}.
951 In ISO C99, you would use a @dfn{flexible array member}, which is
952 slightly different in syntax and semantics:
956 Flexible array members are written as @code{contents[]} without
960 Flexible array members have incomplete type, and so the @code{sizeof}
961 operator may not be applied. As a quirk of the original implementation
962 of zero-length arrays, @code{sizeof} evaluates to zero.
965 Flexible array members may only appear as the last member of a
966 @code{struct} that is otherwise non-empty.
969 A structure containing a flexible array member, or a union containing
970 such a structure (possibly recursively), may not be a member of a
971 structure or an element of an array. (However, these uses are
972 permitted by GCC as extensions.)
975 GCC versions before 3.0 allowed zero-length arrays to be statically
976 initialized, as if they were flexible arrays. In addition to those
977 cases that were useful, it also allowed initializations in situations
978 that would corrupt later data. Non-empty initialization of zero-length
979 arrays is now treated like any case where there are more initializer
980 elements than the array holds, in that a suitable warning about "excess
981 elements in array" is given, and the excess elements (all of them, in
982 this case) are ignored.
984 Instead GCC allows static initialization of flexible array members.
985 This is equivalent to defining a new structure containing the original
986 structure followed by an array of sufficient size to contain the data.
987 I.e.@: in the following, @code{f1} is constructed as if it were declared
993 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
996 struct f1 f1; int data[3];
997 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1001 The convenience of this extension is that @code{f1} has the desired
1002 type, eliminating the need to consistently refer to @code{f2.f1}.
1004 This has symmetry with normal static arrays, in that an array of
1005 unknown size is also written with @code{[]}.
1007 Of course, this extension only makes sense if the extra data comes at
1008 the end of a top-level object, as otherwise we would be overwriting
1009 data at subsequent offsets. To avoid undue complication and confusion
1010 with initialization of deeply nested arrays, we simply disallow any
1011 non-empty initialization except when the structure is the top-level
1012 object. For example:
1015 struct foo @{ int x; int y[]; @};
1016 struct bar @{ struct foo z; @};
1018 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1019 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1020 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1021 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1024 @node Empty Structures
1025 @section Structures With No Members
1026 @cindex empty structures
1027 @cindex zero-size structures
1029 GCC permits a C structure to have no members:
1036 The structure will have size zero. In C++, empty structures are part
1037 of the language. G++ treats empty structures as if they had a single
1038 member of type @code{char}.
1040 @node Variable Length
1041 @section Arrays of Variable Length
1042 @cindex variable-length arrays
1043 @cindex arrays of variable length
1046 Variable-length automatic arrays are allowed in ISO C99, and as an
1047 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1048 implementation of variable-length arrays does not yet conform in detail
1049 to the ISO C99 standard.) These arrays are
1050 declared like any other automatic arrays, but with a length that is not
1051 a constant expression. The storage is allocated at the point of
1052 declaration and deallocated when the brace-level is exited. For
1057 concat_fopen (char *s1, char *s2, char *mode)
1059 char str[strlen (s1) + strlen (s2) + 1];
1062 return fopen (str, mode);
1066 @cindex scope of a variable length array
1067 @cindex variable-length array scope
1068 @cindex deallocating variable length arrays
1069 Jumping or breaking out of the scope of the array name deallocates the
1070 storage. Jumping into the scope is not allowed; you get an error
1073 @cindex @code{alloca} vs variable-length arrays
1074 You can use the function @code{alloca} to get an effect much like
1075 variable-length arrays. The function @code{alloca} is available in
1076 many other C implementations (but not in all). On the other hand,
1077 variable-length arrays are more elegant.
1079 There are other differences between these two methods. Space allocated
1080 with @code{alloca} exists until the containing @emph{function} returns.
1081 The space for a variable-length array is deallocated as soon as the array
1082 name's scope ends. (If you use both variable-length arrays and
1083 @code{alloca} in the same function, deallocation of a variable-length array
1084 will also deallocate anything more recently allocated with @code{alloca}.)
1086 You can also use variable-length arrays as arguments to functions:
1090 tester (int len, char data[len][len])
1096 The length of an array is computed once when the storage is allocated
1097 and is remembered for the scope of the array in case you access it with
1100 If you want to pass the array first and the length afterward, you can
1101 use a forward declaration in the parameter list---another GNU extension.
1105 tester (int len; char data[len][len], int len)
1111 @cindex parameter forward declaration
1112 The @samp{int len} before the semicolon is a @dfn{parameter forward
1113 declaration}, and it serves the purpose of making the name @code{len}
1114 known when the declaration of @code{data} is parsed.
1116 You can write any number of such parameter forward declarations in the
1117 parameter list. They can be separated by commas or semicolons, but the
1118 last one must end with a semicolon, which is followed by the ``real''
1119 parameter declarations. Each forward declaration must match a ``real''
1120 declaration in parameter name and data type. ISO C99 does not support
1121 parameter forward declarations.
1123 @node Variadic Macros
1124 @section Macros with a Variable Number of Arguments.
1125 @cindex variable number of arguments
1126 @cindex macro with variable arguments
1127 @cindex rest argument (in macro)
1128 @cindex variadic macros
1130 In the ISO C standard of 1999, a macro can be declared to accept a
1131 variable number of arguments much as a function can. The syntax for
1132 defining the macro is similar to that of a function. Here is an
1136 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1139 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1140 such a macro, it represents the zero or more tokens until the closing
1141 parenthesis that ends the invocation, including any commas. This set of
1142 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1143 wherever it appears. See the CPP manual for more information.
1145 GCC has long supported variadic macros, and used a different syntax that
1146 allowed you to give a name to the variable arguments just like any other
1147 argument. Here is an example:
1150 #define debug(format, args...) fprintf (stderr, format, args)
1153 This is in all ways equivalent to the ISO C example above, but arguably
1154 more readable and descriptive.
1156 GNU CPP has two further variadic macro extensions, and permits them to
1157 be used with either of the above forms of macro definition.
1159 In standard C, you are not allowed to leave the variable argument out
1160 entirely; but you are allowed to pass an empty argument. For example,
1161 this invocation is invalid in ISO C, because there is no comma after
1168 GNU CPP permits you to completely omit the variable arguments in this
1169 way. In the above examples, the compiler would complain, though since
1170 the expansion of the macro still has the extra comma after the format
1173 To help solve this problem, CPP behaves specially for variable arguments
1174 used with the token paste operator, @samp{##}. If instead you write
1177 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1180 and if the variable arguments are omitted or empty, the @samp{##}
1181 operator causes the preprocessor to remove the comma before it. If you
1182 do provide some variable arguments in your macro invocation, GNU CPP
1183 does not complain about the paste operation and instead places the
1184 variable arguments after the comma. Just like any other pasted macro
1185 argument, these arguments are not macro expanded.
1187 @node Escaped Newlines
1188 @section Slightly Looser Rules for Escaped Newlines
1189 @cindex escaped newlines
1190 @cindex newlines (escaped)
1192 Recently, the preprocessor has relaxed its treatment of escaped
1193 newlines. Previously, the newline had to immediately follow a
1194 backslash. The current implementation allows whitespace in the form
1195 of spaces, horizontal and vertical tabs, and form feeds between the
1196 backslash and the subsequent newline. The preprocessor issues a
1197 warning, but treats it as a valid escaped newline and combines the two
1198 lines to form a single logical line. This works within comments and
1199 tokens, as well as between tokens. Comments are @emph{not} treated as
1200 whitespace for the purposes of this relaxation, since they have not
1201 yet been replaced with spaces.
1204 @section Non-Lvalue Arrays May Have Subscripts
1205 @cindex subscripting
1206 @cindex arrays, non-lvalue
1208 @cindex subscripting and function values
1209 In ISO C99, arrays that are not lvalues still decay to pointers, and
1210 may be subscripted, although they may not be modified or used after
1211 the next sequence point and the unary @samp{&} operator may not be
1212 applied to them. As an extension, GCC allows such arrays to be
1213 subscripted in C89 mode, though otherwise they do not decay to
1214 pointers outside C99 mode. For example,
1215 this is valid in GNU C though not valid in C89:
1219 struct foo @{int a[4];@};
1225 return f().a[index];
1231 @section Arithmetic on @code{void}- and Function-Pointers
1232 @cindex void pointers, arithmetic
1233 @cindex void, size of pointer to
1234 @cindex function pointers, arithmetic
1235 @cindex function, size of pointer to
1237 In GNU C, addition and subtraction operations are supported on pointers to
1238 @code{void} and on pointers to functions. This is done by treating the
1239 size of a @code{void} or of a function as 1.
1241 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1242 and on function types, and returns 1.
1244 @opindex Wpointer-arith
1245 The option @option{-Wpointer-arith} requests a warning if these extensions
1249 @section Non-Constant Initializers
1250 @cindex initializers, non-constant
1251 @cindex non-constant initializers
1253 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1254 automatic variable are not required to be constant expressions in GNU C@.
1255 Here is an example of an initializer with run-time varying elements:
1258 foo (float f, float g)
1260 float beat_freqs[2] = @{ f-g, f+g @};
1265 @node Compound Literals
1266 @section Compound Literals
1267 @cindex constructor expressions
1268 @cindex initializations in expressions
1269 @cindex structures, constructor expression
1270 @cindex expressions, constructor
1271 @cindex compound literals
1272 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1274 ISO C99 supports compound literals. A compound literal looks like
1275 a cast containing an initializer. Its value is an object of the
1276 type specified in the cast, containing the elements specified in
1277 the initializer; it is an lvalue. As an extension, GCC supports
1278 compound literals in C89 mode and in C++.
1280 Usually, the specified type is a structure. Assume that
1281 @code{struct foo} and @code{structure} are declared as shown:
1284 struct foo @{int a; char b[2];@} structure;
1288 Here is an example of constructing a @code{struct foo} with a compound literal:
1291 structure = ((struct foo) @{x + y, 'a', 0@});
1295 This is equivalent to writing the following:
1299 struct foo temp = @{x + y, 'a', 0@};
1304 You can also construct an array. If all the elements of the compound literal
1305 are (made up of) simple constant expressions, suitable for use in
1306 initializers of objects of static storage duration, then the compound
1307 literal can be coerced to a pointer to its first element and used in
1308 such an initializer, as shown here:
1311 char **foo = (char *[]) @{ "x", "y", "z" @};
1314 Compound literals for scalar types and union types are is
1315 also allowed, but then the compound literal is equivalent
1318 As a GNU extension, GCC allows initialization of objects with static storage
1319 duration by compound literals (which is not possible in ISO C99, because
1320 the initializer is not a constant).
1321 It is handled as if the object was initialized only with the bracket
1322 enclosed list if the types of the compound literal and the object match.
1323 The initializer list of the compound literal must be constant.
1324 If the object being initialized has array type of unknown size, the size is
1325 determined by compound literal size.
1328 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1329 static int y[] = (int []) @{1, 2, 3@};
1330 static int z[] = (int [3]) @{1@};
1334 The above lines are equivalent to the following:
1336 static struct foo x = @{1, 'a', 'b'@};
1337 static int y[] = @{1, 2, 3@};
1338 static int z[] = @{1, 0, 0@};
1341 @node Designated Inits
1342 @section Designated Initializers
1343 @cindex initializers with labeled elements
1344 @cindex labeled elements in initializers
1345 @cindex case labels in initializers
1346 @cindex designated initializers
1348 Standard C89 requires the elements of an initializer to appear in a fixed
1349 order, the same as the order of the elements in the array or structure
1352 In ISO C99 you can give the elements in any order, specifying the array
1353 indices or structure field names they apply to, and GNU C allows this as
1354 an extension in C89 mode as well. This extension is not
1355 implemented in GNU C++.
1357 To specify an array index, write
1358 @samp{[@var{index}] =} before the element value. For example,
1361 int a[6] = @{ [4] = 29, [2] = 15 @};
1368 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1372 The index values must be constant expressions, even if the array being
1373 initialized is automatic.
1375 An alternative syntax for this which has been obsolete since GCC 2.5 but
1376 GCC still accepts is to write @samp{[@var{index}]} before the element
1377 value, with no @samp{=}.
1379 To initialize a range of elements to the same value, write
1380 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1381 extension. For example,
1384 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1388 If the value in it has side-effects, the side-effects will happen only once,
1389 not for each initialized field by the range initializer.
1392 Note that the length of the array is the highest value specified
1395 In a structure initializer, specify the name of a field to initialize
1396 with @samp{.@var{fieldname} =} before the element value. For example,
1397 given the following structure,
1400 struct point @{ int x, y; @};
1404 the following initialization
1407 struct point p = @{ .y = yvalue, .x = xvalue @};
1414 struct point p = @{ xvalue, yvalue @};
1417 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1418 @samp{@var{fieldname}:}, as shown here:
1421 struct point p = @{ y: yvalue, x: xvalue @};
1425 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1426 @dfn{designator}. You can also use a designator (or the obsolete colon
1427 syntax) when initializing a union, to specify which element of the union
1428 should be used. For example,
1431 union foo @{ int i; double d; @};
1433 union foo f = @{ .d = 4 @};
1437 will convert 4 to a @code{double} to store it in the union using
1438 the second element. By contrast, casting 4 to type @code{union foo}
1439 would store it into the union as the integer @code{i}, since it is
1440 an integer. (@xref{Cast to Union}.)
1442 You can combine this technique of naming elements with ordinary C
1443 initialization of successive elements. Each initializer element that
1444 does not have a designator applies to the next consecutive element of the
1445 array or structure. For example,
1448 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1455 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1458 Labeling the elements of an array initializer is especially useful
1459 when the indices are characters or belong to an @code{enum} type.
1464 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1465 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1468 @cindex designator lists
1469 You can also write a series of @samp{.@var{fieldname}} and
1470 @samp{[@var{index}]} designators before an @samp{=} to specify a
1471 nested subobject to initialize; the list is taken relative to the
1472 subobject corresponding to the closest surrounding brace pair. For
1473 example, with the @samp{struct point} declaration above:
1476 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1480 If the same field is initialized multiple times, it will have value from
1481 the last initialization. If any such overridden initialization has
1482 side-effect, it is unspecified whether the side-effect happens or not.
1483 Currently, GCC will discard them and issue a warning.
1486 @section Case Ranges
1488 @cindex ranges in case statements
1490 You can specify a range of consecutive values in a single @code{case} label,
1494 case @var{low} ... @var{high}:
1498 This has the same effect as the proper number of individual @code{case}
1499 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1501 This feature is especially useful for ranges of ASCII character codes:
1507 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1508 it may be parsed wrong when you use it with integer values. For example,
1523 @section Cast to a Union Type
1524 @cindex cast to a union
1525 @cindex union, casting to a
1527 A cast to union type is similar to other casts, except that the type
1528 specified is a union type. You can specify the type either with
1529 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1530 a constructor though, not a cast, and hence does not yield an lvalue like
1531 normal casts. (@xref{Compound Literals}.)
1533 The types that may be cast to the union type are those of the members
1534 of the union. Thus, given the following union and variables:
1537 union foo @{ int i; double d; @};
1543 both @code{x} and @code{y} can be cast to type @code{union foo}.
1545 Using the cast as the right-hand side of an assignment to a variable of
1546 union type is equivalent to storing in a member of the union:
1551 u = (union foo) x @equiv{} u.i = x
1552 u = (union foo) y @equiv{} u.d = y
1555 You can also use the union cast as a function argument:
1558 void hack (union foo);
1560 hack ((union foo) x);
1563 @node Mixed Declarations
1564 @section Mixed Declarations and Code
1565 @cindex mixed declarations and code
1566 @cindex declarations, mixed with code
1567 @cindex code, mixed with declarations
1569 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1570 within compound statements. As an extension, GCC also allows this in
1571 C89 mode. For example, you could do:
1580 Each identifier is visible from where it is declared until the end of
1581 the enclosing block.
1583 @node Function Attributes
1584 @section Declaring Attributes of Functions
1585 @cindex function attributes
1586 @cindex declaring attributes of functions
1587 @cindex functions that never return
1588 @cindex functions that return more than once
1589 @cindex functions that have no side effects
1590 @cindex functions in arbitrary sections
1591 @cindex functions that behave like malloc
1592 @cindex @code{volatile} applied to function
1593 @cindex @code{const} applied to function
1594 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1595 @cindex functions with non-null pointer arguments
1596 @cindex functions that are passed arguments in registers on the 386
1597 @cindex functions that pop the argument stack on the 386
1598 @cindex functions that do not pop the argument stack on the 386
1600 In GNU C, you declare certain things about functions called in your program
1601 which help the compiler optimize function calls and check your code more
1604 The keyword @code{__attribute__} allows you to specify special
1605 attributes when making a declaration. This keyword is followed by an
1606 attribute specification inside double parentheses. The following
1607 attributes are currently defined for functions on all targets:
1608 @code{aligned}, @code{alloc_size}, @code{noreturn},
1609 @code{returns_twice}, @code{noinline}, @code{always_inline},
1610 @code{flatten}, @code{pure}, @code{const}, @code{nothrow},
1611 @code{sentinel}, @code{format}, @code{format_arg},
1612 @code{no_instrument_function}, @code{section}, @code{constructor},
1613 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
1614 @code{weak}, @code{malloc}, @code{alias}, @code{warn_unused_result},
1615 @code{nonnull}, @code{gnu_inline} and @code{externally_visible},
1616 @code{hot}, @code{cold}.
1617 Several other attributes are defined for functions on particular
1618 target systems. Other attributes, including @code{section} are
1619 supported for variables declarations (@pxref{Variable Attributes}) and
1620 for types (@pxref{Type Attributes}).
1622 You may also specify attributes with @samp{__} preceding and following
1623 each keyword. This allows you to use them in header files without
1624 being concerned about a possible macro of the same name. For example,
1625 you may use @code{__noreturn__} instead of @code{noreturn}.
1627 @xref{Attribute Syntax}, for details of the exact syntax for using
1631 @c Keep this table alphabetized by attribute name. Treat _ as space.
1633 @item alias ("@var{target}")
1634 @cindex @code{alias} attribute
1635 The @code{alias} attribute causes the declaration to be emitted as an
1636 alias for another symbol, which must be specified. For instance,
1639 void __f () @{ /* @r{Do something.} */; @}
1640 void f () __attribute__ ((weak, alias ("__f")));
1643 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
1644 mangled name for the target must be used. It is an error if @samp{__f}
1645 is not defined in the same translation unit.
1647 Not all target machines support this attribute.
1649 @item aligned (@var{alignment})
1650 @cindex @code{aligned} attribute
1651 This attribute specifies a minimum alignment for the function,
1654 You cannot use this attribute to decrease the alignment of a function,
1655 only to increase it. However, when you explicitly specify a function
1656 alignment this will override the effect of the
1657 @option{-falign-functions} (@pxref{Optimize Options}) option for this
1660 Note that the effectiveness of @code{aligned} attributes may be
1661 limited by inherent limitations in your linker. On many systems, the
1662 linker is only able to arrange for functions to be aligned up to a
1663 certain maximum alignment. (For some linkers, the maximum supported
1664 alignment may be very very small.) See your linker documentation for
1665 further information.
1667 The @code{aligned} attribute can also be used for variables and fields
1668 (@pxref{Variable Attributes}.)
1671 @cindex @code{alloc_size} attribute
1672 The @code{alloc_size} attribute is used to tell the compiler that the
1673 function return value points to memory, where the size is given by
1674 one or two of the functions parameters. GCC uses this
1675 information to improve the correctness of @code{__builtin_object_size}.
1677 The function parameter(s) denoting the allocated size are specified by
1678 one or two integer arguments supplied to the attribute. The allocated size
1679 is either the value of the single function argument specified or the product
1680 of the two function arguments specified. Argument numbering starts at
1686 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
1687 void my_realloc(void* size_t) __attribute__((alloc_size(2)))
1690 declares that my_calloc will return memory of the size given by
1691 the product of parameter 1 and 2 and that my_realloc will return memory
1692 of the size given by parameter 2.
1695 @cindex @code{always_inline} function attribute
1696 Generally, functions are not inlined unless optimization is specified.
1697 For functions declared inline, this attribute inlines the function even
1698 if no optimization level was specified.
1701 @cindex @code{gnu_inline} function attribute
1702 This attribute should be used with a function which is also declared
1703 with the @code{inline} keyword. It directs GCC to treat the function
1704 as if it were defined in gnu89 mode even when compiling in C99 or
1707 If the function is declared @code{extern}, then this definition of the
1708 function is used only for inlining. In no case is the function
1709 compiled as a standalone function, not even if you take its address
1710 explicitly. Such an address becomes an external reference, as if you
1711 had only declared the function, and had not defined it. This has
1712 almost the effect of a macro. The way to use this is to put a
1713 function definition in a header file with this attribute, and put
1714 another copy of the function, without @code{extern}, in a library
1715 file. The definition in the header file will cause most calls to the
1716 function to be inlined. If any uses of the function remain, they will
1717 refer to the single copy in the library. Note that the two
1718 definitions of the functions need not be precisely the same, although
1719 if they do not have the same effect your program may behave oddly.
1721 In C, if the function is neither @code{extern} nor @code{static}, then
1722 the function is compiled as a standalone function, as well as being
1723 inlined where possible.
1725 This is how GCC traditionally handled functions declared
1726 @code{inline}. Since ISO C99 specifies a different semantics for
1727 @code{inline}, this function attribute is provided as a transition
1728 measure and as a useful feature in its own right. This attribute is
1729 available in GCC 4.1.3 and later. It is available if either of the
1730 preprocessor macros @code{__GNUC_GNU_INLINE__} or
1731 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
1732 Function is As Fast As a Macro}.
1734 In C++, this attribute does not depend on @code{extern} in any way,
1735 but it still requires the @code{inline} keyword to enable its special
1738 @cindex @code{flatten} function attribute
1740 Generally, inlining into a function is limited. For a function marked with
1741 this attribute, every call inside this function will be inlined, if possible.
1742 Whether the function itself is considered for inlining depends on its size and
1743 the current inlining parameters. The @code{flatten} attribute only works
1744 reliably in unit-at-a-time mode.
1747 @cindex functions that do pop the argument stack on the 386
1749 On the Intel 386, the @code{cdecl} attribute causes the compiler to
1750 assume that the calling function will pop off the stack space used to
1751 pass arguments. This is
1752 useful to override the effects of the @option{-mrtd} switch.
1755 @cindex @code{const} function attribute
1756 Many functions do not examine any values except their arguments, and
1757 have no effects except the return value. Basically this is just slightly
1758 more strict class than the @code{pure} attribute below, since function is not
1759 allowed to read global memory.
1761 @cindex pointer arguments
1762 Note that a function that has pointer arguments and examines the data
1763 pointed to must @emph{not} be declared @code{const}. Likewise, a
1764 function that calls a non-@code{const} function usually must not be
1765 @code{const}. It does not make sense for a @code{const} function to
1768 The attribute @code{const} is not implemented in GCC versions earlier
1769 than 2.5. An alternative way to declare that a function has no side
1770 effects, which works in the current version and in some older versions,
1774 typedef int intfn ();
1776 extern const intfn square;
1779 This approach does not work in GNU C++ from 2.6.0 on, since the language
1780 specifies that the @samp{const} must be attached to the return value.
1784 @itemx constructor (@var{priority})
1785 @itemx destructor (@var{priority})
1786 @cindex @code{constructor} function attribute
1787 @cindex @code{destructor} function attribute
1788 The @code{constructor} attribute causes the function to be called
1789 automatically before execution enters @code{main ()}. Similarly, the
1790 @code{destructor} attribute causes the function to be called
1791 automatically after @code{main ()} has completed or @code{exit ()} has
1792 been called. Functions with these attributes are useful for
1793 initializing data that will be used implicitly during the execution of
1796 You may provide an optional integer priority to control the order in
1797 which constructor and destructor functions are run. A constructor
1798 with a smaller priority number runs before a constructor with a larger
1799 priority number; the opposite relationship holds for destructors. So,
1800 if you have a constructor that allocates a resource and a destructor
1801 that deallocates the same resource, both functions typically have the
1802 same priority. The priorities for constructor and destructor
1803 functions are the same as those specified for namespace-scope C++
1804 objects (@pxref{C++ Attributes}).
1806 These attributes are not currently implemented for Objective-C@.
1809 @cindex @code{deprecated} attribute.
1810 The @code{deprecated} attribute results in a warning if the function
1811 is used anywhere in the source file. This is useful when identifying
1812 functions that are expected to be removed in a future version of a
1813 program. The warning also includes the location of the declaration
1814 of the deprecated function, to enable users to easily find further
1815 information about why the function is deprecated, or what they should
1816 do instead. Note that the warnings only occurs for uses:
1819 int old_fn () __attribute__ ((deprecated));
1821 int (*fn_ptr)() = old_fn;
1824 results in a warning on line 3 but not line 2.
1826 The @code{deprecated} attribute can also be used for variables and
1827 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
1830 @cindex @code{__declspec(dllexport)}
1831 On Microsoft Windows targets and Symbian OS targets the
1832 @code{dllexport} attribute causes the compiler to provide a global
1833 pointer to a pointer in a DLL, so that it can be referenced with the
1834 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
1835 name is formed by combining @code{_imp__} and the function or variable
1838 You can use @code{__declspec(dllexport)} as a synonym for
1839 @code{__attribute__ ((dllexport))} for compatibility with other
1842 On systems that support the @code{visibility} attribute, this
1843 attribute also implies ``default'' visibility. It is an error to
1844 explicitly specify any other visibility.
1846 Currently, the @code{dllexport} attribute is ignored for inlined
1847 functions, unless the @option{-fkeep-inline-functions} flag has been
1848 used. The attribute is also ignored for undefined symbols.
1850 When applied to C++ classes, the attribute marks defined non-inlined
1851 member functions and static data members as exports. Static consts
1852 initialized in-class are not marked unless they are also defined
1855 For Microsoft Windows targets there are alternative methods for
1856 including the symbol in the DLL's export table such as using a
1857 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
1858 the @option{--export-all} linker flag.
1861 @cindex @code{__declspec(dllimport)}
1862 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
1863 attribute causes the compiler to reference a function or variable via
1864 a global pointer to a pointer that is set up by the DLL exporting the
1865 symbol. The attribute implies @code{extern}. On Microsoft Windows
1866 targets, the pointer name is formed by combining @code{_imp__} and the
1867 function or variable name.
1869 You can use @code{__declspec(dllimport)} as a synonym for
1870 @code{__attribute__ ((dllimport))} for compatibility with other
1873 On systems that support the @code{visibility} attribute, this
1874 attribute also implies ``default'' visibility. It is an error to
1875 explicitly specify any other visibility.
1877 Currently, the attribute is ignored for inlined functions. If the
1878 attribute is applied to a symbol @emph{definition}, an error is reported.
1879 If a symbol previously declared @code{dllimport} is later defined, the
1880 attribute is ignored in subsequent references, and a warning is emitted.
1881 The attribute is also overridden by a subsequent declaration as
1884 When applied to C++ classes, the attribute marks non-inlined
1885 member functions and static data members as imports. However, the
1886 attribute is ignored for virtual methods to allow creation of vtables
1889 On the SH Symbian OS target the @code{dllimport} attribute also has
1890 another affect---it can cause the vtable and run-time type information
1891 for a class to be exported. This happens when the class has a
1892 dllimport'ed constructor or a non-inline, non-pure virtual function
1893 and, for either of those two conditions, the class also has a inline
1894 constructor or destructor and has a key function that is defined in
1895 the current translation unit.
1897 For Microsoft Windows based targets the use of the @code{dllimport}
1898 attribute on functions is not necessary, but provides a small
1899 performance benefit by eliminating a thunk in the DLL@. The use of the
1900 @code{dllimport} attribute on imported variables was required on older
1901 versions of the GNU linker, but can now be avoided by passing the
1902 @option{--enable-auto-import} switch to the GNU linker. As with
1903 functions, using the attribute for a variable eliminates a thunk in
1906 One drawback to using this attribute is that a pointer to a function
1907 or variable marked as @code{dllimport} cannot be used as a constant
1908 address. On Microsoft Windows targets, the attribute can be disabled
1909 for functions by setting the @option{-mnop-fun-dllimport} flag.
1912 @cindex eight bit data on the H8/300, H8/300H, and H8S
1913 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1914 variable should be placed into the eight bit data section.
1915 The compiler will generate more efficient code for certain operations
1916 on data in the eight bit data area. Note the eight bit data area is limited to
1919 You must use GAS and GLD from GNU binutils version 2.7 or later for
1920 this attribute to work correctly.
1922 @item exception_handler
1923 @cindex exception handler functions on the Blackfin processor
1924 Use this attribute on the Blackfin to indicate that the specified function
1925 is an exception handler. The compiler will generate function entry and
1926 exit sequences suitable for use in an exception handler when this
1927 attribute is present.
1930 @cindex functions which handle memory bank switching
1931 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
1932 use a calling convention that takes care of switching memory banks when
1933 entering and leaving a function. This calling convention is also the
1934 default when using the @option{-mlong-calls} option.
1936 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
1937 to call and return from a function.
1939 On 68HC11 the compiler will generate a sequence of instructions
1940 to invoke a board-specific routine to switch the memory bank and call the
1941 real function. The board-specific routine simulates a @code{call}.
1942 At the end of a function, it will jump to a board-specific routine
1943 instead of using @code{rts}. The board-specific return routine simulates
1947 @cindex functions that pop the argument stack on the 386
1948 On the Intel 386, the @code{fastcall} attribute causes the compiler to
1949 pass the first argument (if of integral type) in the register ECX and
1950 the second argument (if of integral type) in the register EDX@. Subsequent
1951 and other typed arguments are passed on the stack. The called function will
1952 pop the arguments off the stack. If the number of arguments is variable all
1953 arguments are pushed on the stack.
1955 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
1956 @cindex @code{format} function attribute
1958 The @code{format} attribute specifies that a function takes @code{printf},
1959 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
1960 should be type-checked against a format string. For example, the
1965 my_printf (void *my_object, const char *my_format, ...)
1966 __attribute__ ((format (printf, 2, 3)));
1970 causes the compiler to check the arguments in calls to @code{my_printf}
1971 for consistency with the @code{printf} style format string argument
1974 The parameter @var{archetype} determines how the format string is
1975 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
1976 or @code{strfmon}. (You can also use @code{__printf__},
1977 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
1978 parameter @var{string-index} specifies which argument is the format
1979 string argument (starting from 1), while @var{first-to-check} is the
1980 number of the first argument to check against the format string. For
1981 functions where the arguments are not available to be checked (such as
1982 @code{vprintf}), specify the third parameter as zero. In this case the
1983 compiler only checks the format string for consistency. For
1984 @code{strftime} formats, the third parameter is required to be zero.
1985 Since non-static C++ methods have an implicit @code{this} argument, the
1986 arguments of such methods should be counted from two, not one, when
1987 giving values for @var{string-index} and @var{first-to-check}.
1989 In the example above, the format string (@code{my_format}) is the second
1990 argument of the function @code{my_print}, and the arguments to check
1991 start with the third argument, so the correct parameters for the format
1992 attribute are 2 and 3.
1994 @opindex ffreestanding
1995 @opindex fno-builtin
1996 The @code{format} attribute allows you to identify your own functions
1997 which take format strings as arguments, so that GCC can check the
1998 calls to these functions for errors. The compiler always (unless
1999 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2000 for the standard library functions @code{printf}, @code{fprintf},
2001 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2002 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2003 warnings are requested (using @option{-Wformat}), so there is no need to
2004 modify the header file @file{stdio.h}. In C99 mode, the functions
2005 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2006 @code{vsscanf} are also checked. Except in strictly conforming C
2007 standard modes, the X/Open function @code{strfmon} is also checked as
2008 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2009 @xref{C Dialect Options,,Options Controlling C Dialect}.
2011 The target may provide additional types of format checks.
2012 @xref{Target Format Checks,,Format Checks Specific to Particular
2015 @item format_arg (@var{string-index})
2016 @cindex @code{format_arg} function attribute
2017 @opindex Wformat-nonliteral
2018 The @code{format_arg} attribute specifies that a function takes a format
2019 string for a @code{printf}, @code{scanf}, @code{strftime} or
2020 @code{strfmon} style function and modifies it (for example, to translate
2021 it into another language), so the result can be passed to a
2022 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2023 function (with the remaining arguments to the format function the same
2024 as they would have been for the unmodified string). For example, the
2029 my_dgettext (char *my_domain, const char *my_format)
2030 __attribute__ ((format_arg (2)));
2034 causes the compiler to check the arguments in calls to a @code{printf},
2035 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2036 format string argument is a call to the @code{my_dgettext} function, for
2037 consistency with the format string argument @code{my_format}. If the
2038 @code{format_arg} attribute had not been specified, all the compiler
2039 could tell in such calls to format functions would be that the format
2040 string argument is not constant; this would generate a warning when
2041 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2042 without the attribute.
2044 The parameter @var{string-index} specifies which argument is the format
2045 string argument (starting from one). Since non-static C++ methods have
2046 an implicit @code{this} argument, the arguments of such methods should
2047 be counted from two.
2049 The @code{format-arg} attribute allows you to identify your own
2050 functions which modify format strings, so that GCC can check the
2051 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2052 type function whose operands are a call to one of your own function.
2053 The compiler always treats @code{gettext}, @code{dgettext}, and
2054 @code{dcgettext} in this manner except when strict ISO C support is
2055 requested by @option{-ansi} or an appropriate @option{-std} option, or
2056 @option{-ffreestanding} or @option{-fno-builtin}
2057 is used. @xref{C Dialect Options,,Options
2058 Controlling C Dialect}.
2060 @item function_vector
2061 @cindex calling functions through the function vector on H8/300, M16C, and M32C processors
2062 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2063 function should be called through the function vector. Calling a
2064 function through the function vector will reduce code size, however;
2065 the function vector has a limited size (maximum 128 entries on the H8/300
2066 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2068 You must use GAS and GLD from GNU binutils version 2.7 or later for
2069 this attribute to work correctly.
2071 On M16C/M32C targets, the @code{function_vector} attribute declares a
2072 special page subroutine call function. Use of this attribute reduces
2073 the code size by 2 bytes for each call generated to the
2074 subroutine. The argument to the attribute is the vector number entry
2075 from the special page vector table which contains the 16 low-order
2076 bits of the subroutine's entry address. Each vector table has special
2077 page number (18 to 255) which are used in @code{jsrs} instruction.
2078 Jump addresses of the routines are generated by adding 0x0F0000 (in
2079 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the 2
2080 byte addresses set in the vector table. Therefore you need to ensure
2081 that all the special page vector routines should get mapped within the
2082 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2085 In the following example 2 bytes will be saved for each call to
2086 function @code{foo}.
2089 void foo (void) __attribute__((function_vector(0x18)));
2100 If functions are defined in one file and are called in another file,
2101 then be sure to write this declaration in both files.
2103 This attribute is ignored for R8C target.
2106 @cindex interrupt handler functions
2107 Use this attribute on the ARM, AVR, C4x, CRX, M32C, M32R/D, m68k, MS1,
2108 and Xstormy16 ports to indicate that the specified function is an
2109 interrupt handler. The compiler will generate function entry and exit
2110 sequences suitable for use in an interrupt handler when this attribute
2113 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, and
2114 SH processors can be specified via the @code{interrupt_handler} attribute.
2116 Note, on the AVR, interrupts will be enabled inside the function.
2118 Note, for the ARM, you can specify the kind of interrupt to be handled by
2119 adding an optional parameter to the interrupt attribute like this:
2122 void f () __attribute__ ((interrupt ("IRQ")));
2125 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2127 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2128 may be called with a word aligned stack pointer.
2130 @item interrupt_handler
2131 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2132 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2133 indicate that the specified function is an interrupt handler. The compiler
2134 will generate function entry and exit sequences suitable for use in an
2135 interrupt handler when this attribute is present.
2137 @item interrupt_thread
2138 @cindex interrupt thread functions on fido
2139 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2140 that the specified function is an interrupt handler that is designed
2141 to run as a thread. The compiler omits generate prologue/epilogue
2142 sequences and replaces the return instruction with a @code{sleep}
2143 instruction. This attribute is available only on fido.
2146 @cindex User stack pointer in interrupts on the Blackfin
2147 When used together with @code{interrupt_handler}, @code{exception_handler}
2148 or @code{nmi_handler}, code will be generated to load the stack pointer
2149 from the USP register in the function prologue.
2151 @item long_call/short_call
2152 @cindex indirect calls on ARM
2153 This attribute specifies how a particular function is called on
2154 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2155 command line switch and @code{#pragma long_calls} settings. The
2156 @code{long_call} attribute indicates that the function might be far
2157 away from the call site and require a different (more expensive)
2158 calling sequence. The @code{short_call} attribute always places
2159 the offset to the function from the call site into the @samp{BL}
2160 instruction directly.
2162 @item longcall/shortcall
2163 @cindex functions called via pointer on the RS/6000 and PowerPC
2164 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
2165 indicates that the function might be far away from the call site and
2166 require a different (more expensive) calling sequence. The
2167 @code{shortcall} attribute indicates that the function is always close
2168 enough for the shorter calling sequence to be used. These attributes
2169 override both the @option{-mlongcall} switch and, on the RS/6000 and
2170 PowerPC, the @code{#pragma longcall} setting.
2172 @xref{RS/6000 and PowerPC Options}, for more information on whether long
2173 calls are necessary.
2175 @item long_call/near/far
2176 @cindex indirect calls on MIPS
2177 These attributes specify how a particular function is called on MIPS@.
2178 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
2179 command-line switch. The @code{long_call} and @code{far} attributes are
2180 synonyms, and cause the compiler to always call
2181 the function by first loading its address into a register, and then using
2182 the contents of that register. The @code{near} attribute has the opposite
2183 effect; it specifies that non-PIC calls should be made using the more
2184 efficient @code{jal} instruction.
2187 @cindex @code{malloc} attribute
2188 The @code{malloc} attribute is used to tell the compiler that a function
2189 may be treated as if any non-@code{NULL} pointer it returns cannot
2190 alias any other pointer valid when the function returns.
2191 This will often improve optimization.
2192 Standard functions with this property include @code{malloc} and
2193 @code{calloc}. @code{realloc}-like functions have this property as
2194 long as the old pointer is never referred to (including comparing it
2195 to the new pointer) after the function returns a non-@code{NULL}
2198 @item model (@var{model-name})
2199 @cindex function addressability on the M32R/D
2200 @cindex variable addressability on the IA-64
2202 On the M32R/D, use this attribute to set the addressability of an
2203 object, and of the code generated for a function. The identifier
2204 @var{model-name} is one of @code{small}, @code{medium}, or
2205 @code{large}, representing each of the code models.
2207 Small model objects live in the lower 16MB of memory (so that their
2208 addresses can be loaded with the @code{ld24} instruction), and are
2209 callable with the @code{bl} instruction.
2211 Medium model objects may live anywhere in the 32-bit address space (the
2212 compiler will generate @code{seth/add3} instructions to load their addresses),
2213 and are callable with the @code{bl} instruction.
2215 Large model objects may live anywhere in the 32-bit address space (the
2216 compiler will generate @code{seth/add3} instructions to load their addresses),
2217 and may not be reachable with the @code{bl} instruction (the compiler will
2218 generate the much slower @code{seth/add3/jl} instruction sequence).
2220 On IA-64, use this attribute to set the addressability of an object.
2221 At present, the only supported identifier for @var{model-name} is
2222 @code{small}, indicating addressability via ``small'' (22-bit)
2223 addresses (so that their addresses can be loaded with the @code{addl}
2224 instruction). Caveat: such addressing is by definition not position
2225 independent and hence this attribute must not be used for objects
2226 defined by shared libraries.
2229 @cindex function without a prologue/epilogue code
2230 Use this attribute on the ARM, AVR, C4x, IP2K and SPU ports to indicate that
2231 the specified function does not need prologue/epilogue sequences generated by
2232 the compiler. It is up to the programmer to provide these sequences.
2235 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2236 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2237 use the normal calling convention based on @code{jsr} and @code{rts}.
2238 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2242 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2243 Use this attribute together with @code{interrupt_handler},
2244 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2245 entry code should enable nested interrupts or exceptions.
2248 @cindex NMI handler functions on the Blackfin processor
2249 Use this attribute on the Blackfin to indicate that the specified function
2250 is an NMI handler. The compiler will generate function entry and
2251 exit sequences suitable for use in an NMI handler when this
2252 attribute is present.
2254 @item no_instrument_function
2255 @cindex @code{no_instrument_function} function attribute
2256 @opindex finstrument-functions
2257 If @option{-finstrument-functions} is given, profiling function calls will
2258 be generated at entry and exit of most user-compiled functions.
2259 Functions with this attribute will not be so instrumented.
2262 @cindex @code{noinline} function attribute
2263 This function attribute prevents a function from being considered for
2266 @item nonnull (@var{arg-index}, @dots{})
2267 @cindex @code{nonnull} function attribute
2268 The @code{nonnull} attribute specifies that some function parameters should
2269 be non-null pointers. For instance, the declaration:
2273 my_memcpy (void *dest, const void *src, size_t len)
2274 __attribute__((nonnull (1, 2)));
2278 causes the compiler to check that, in calls to @code{my_memcpy},
2279 arguments @var{dest} and @var{src} are non-null. If the compiler
2280 determines that a null pointer is passed in an argument slot marked
2281 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2282 is issued. The compiler may also choose to make optimizations based
2283 on the knowledge that certain function arguments will not be null.
2285 If no argument index list is given to the @code{nonnull} attribute,
2286 all pointer arguments are marked as non-null. To illustrate, the
2287 following declaration is equivalent to the previous example:
2291 my_memcpy (void *dest, const void *src, size_t len)
2292 __attribute__((nonnull));
2296 @cindex @code{noreturn} function attribute
2297 A few standard library functions, such as @code{abort} and @code{exit},
2298 cannot return. GCC knows this automatically. Some programs define
2299 their own functions that never return. You can declare them
2300 @code{noreturn} to tell the compiler this fact. For example,
2304 void fatal () __attribute__ ((noreturn));
2307 fatal (/* @r{@dots{}} */)
2309 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2315 The @code{noreturn} keyword tells the compiler to assume that
2316 @code{fatal} cannot return. It can then optimize without regard to what
2317 would happen if @code{fatal} ever did return. This makes slightly
2318 better code. More importantly, it helps avoid spurious warnings of
2319 uninitialized variables.
2321 The @code{noreturn} keyword does not affect the exceptional path when that
2322 applies: a @code{noreturn}-marked function may still return to the caller
2323 by throwing an exception or calling @code{longjmp}.
2325 Do not assume that registers saved by the calling function are
2326 restored before calling the @code{noreturn} function.
2328 It does not make sense for a @code{noreturn} function to have a return
2329 type other than @code{void}.
2331 The attribute @code{noreturn} is not implemented in GCC versions
2332 earlier than 2.5. An alternative way to declare that a function does
2333 not return, which works in the current version and in some older
2334 versions, is as follows:
2337 typedef void voidfn ();
2339 volatile voidfn fatal;
2342 This approach does not work in GNU C++.
2345 @cindex @code{nothrow} function attribute
2346 The @code{nothrow} attribute is used to inform the compiler that a
2347 function cannot throw an exception. For example, most functions in
2348 the standard C library can be guaranteed not to throw an exception
2349 with the notable exceptions of @code{qsort} and @code{bsearch} that
2350 take function pointer arguments. The @code{nothrow} attribute is not
2351 implemented in GCC versions earlier than 3.3.
2354 @cindex @code{pure} function attribute
2355 Many functions have no effects except the return value and their
2356 return value depends only on the parameters and/or global variables.
2357 Such a function can be subject
2358 to common subexpression elimination and loop optimization just as an
2359 arithmetic operator would be. These functions should be declared
2360 with the attribute @code{pure}. For example,
2363 int square (int) __attribute__ ((pure));
2367 says that the hypothetical function @code{square} is safe to call
2368 fewer times than the program says.
2370 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2371 Interesting non-pure functions are functions with infinite loops or those
2372 depending on volatile memory or other system resource, that may change between
2373 two consecutive calls (such as @code{feof} in a multithreading environment).
2375 The attribute @code{pure} is not implemented in GCC versions earlier
2379 @cindex @code{hot} function attribute
2380 The @code{hot} attribute is used to inform the compiler that a function is a
2381 hot spot of the compiled program. The function is optimized more aggressively
2382 and on many target it is placed into special subsection of the text section so
2383 all hot functions appears close together improving locality.
2385 When profile feedback is available, via @option{-fprofile-use}, hot functions
2386 are automatically detected and this attribute is ignored.
2388 The @code{hot} attribute is not implemented in GCC versions earlier than 4.3.
2391 @cindex @code{cold} function attribute
2392 The @code{cold} attribute is used to inform the compiler that a function is
2393 unlikely executed. The function is optimized for size rather than speed and on
2394 many targets it is placed into special subsection of the text section so all
2395 cold functions appears close together improving code locality of non-cold parts
2396 of program. The paths leading to call of cold functions within code are marked
2397 as unlikely by the branch prediction mechanism. It is thus useful to mark
2398 functions used to handle unlikely conditions, such as @code{perror}, as cold to
2399 improve optimization of hot functions that do call marked functions in rare
2402 When profile feedback is available, via @option{-fprofile-use}, hot functions
2403 are automatically detected and this attribute is ignored.
2405 The @code{hot} attribute is not implemented in GCC versions earlier than 4.3.
2407 @item regparm (@var{number})
2408 @cindex @code{regparm} attribute
2409 @cindex functions that are passed arguments in registers on the 386
2410 On the Intel 386, the @code{regparm} attribute causes the compiler to
2411 pass arguments number one to @var{number} if they are of integral type
2412 in registers EAX, EDX, and ECX instead of on the stack. Functions that
2413 take a variable number of arguments will continue to be passed all of their
2414 arguments on the stack.
2416 Beware that on some ELF systems this attribute is unsuitable for
2417 global functions in shared libraries with lazy binding (which is the
2418 default). Lazy binding will send the first call via resolving code in
2419 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2420 per the standard calling conventions. Solaris 8 is affected by this.
2421 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2422 safe since the loaders there save all registers. (Lazy binding can be
2423 disabled with the linker or the loader if desired, to avoid the
2427 @cindex @code{sseregparm} attribute
2428 On the Intel 386 with SSE support, the @code{sseregparm} attribute
2429 causes the compiler to pass up to 3 floating point arguments in
2430 SSE registers instead of on the stack. Functions that take a
2431 variable number of arguments will continue to pass all of their
2432 floating point arguments on the stack.
2434 @item force_align_arg_pointer
2435 @cindex @code{force_align_arg_pointer} attribute
2436 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
2437 applied to individual function definitions, generating an alternate
2438 prologue and epilogue that realigns the runtime stack. This supports
2439 mixing legacy codes that run with a 4-byte aligned stack with modern
2440 codes that keep a 16-byte stack for SSE compatibility. The alternate
2441 prologue and epilogue are slower and bigger than the regular ones, and
2442 the alternate prologue requires a scratch register; this lowers the
2443 number of registers available if used in conjunction with the
2444 @code{regparm} attribute. The @code{force_align_arg_pointer}
2445 attribute is incompatible with nested functions; this is considered a
2449 @cindex @code{returns_twice} attribute
2450 The @code{returns_twice} attribute tells the compiler that a function may
2451 return more than one time. The compiler will ensure that all registers
2452 are dead before calling such a function and will emit a warning about
2453 the variables that may be clobbered after the second return from the
2454 function. Examples of such functions are @code{setjmp} and @code{vfork}.
2455 The @code{longjmp}-like counterpart of such function, if any, might need
2456 to be marked with the @code{noreturn} attribute.
2459 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
2460 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
2461 all registers except the stack pointer should be saved in the prologue
2462 regardless of whether they are used or not.
2464 @item section ("@var{section-name}")
2465 @cindex @code{section} function attribute
2466 Normally, the compiler places the code it generates in the @code{text} section.
2467 Sometimes, however, you need additional sections, or you need certain
2468 particular functions to appear in special sections. The @code{section}
2469 attribute specifies that a function lives in a particular section.
2470 For example, the declaration:
2473 extern void foobar (void) __attribute__ ((section ("bar")));
2477 puts the function @code{foobar} in the @code{bar} section.
2479 Some file formats do not support arbitrary sections so the @code{section}
2480 attribute is not available on all platforms.
2481 If you need to map the entire contents of a module to a particular
2482 section, consider using the facilities of the linker instead.
2485 @cindex @code{sentinel} function attribute
2486 This function attribute ensures that a parameter in a function call is
2487 an explicit @code{NULL}. The attribute is only valid on variadic
2488 functions. By default, the sentinel is located at position zero, the
2489 last parameter of the function call. If an optional integer position
2490 argument P is supplied to the attribute, the sentinel must be located at
2491 position P counting backwards from the end of the argument list.
2494 __attribute__ ((sentinel))
2496 __attribute__ ((sentinel(0)))
2499 The attribute is automatically set with a position of 0 for the built-in
2500 functions @code{execl} and @code{execlp}. The built-in function
2501 @code{execle} has the attribute set with a position of 1.
2503 A valid @code{NULL} in this context is defined as zero with any pointer
2504 type. If your system defines the @code{NULL} macro with an integer type
2505 then you need to add an explicit cast. GCC replaces @code{stddef.h}
2506 with a copy that redefines NULL appropriately.
2508 The warnings for missing or incorrect sentinels are enabled with
2512 See long_call/short_call.
2515 See longcall/shortcall.
2518 @cindex signal handler functions on the AVR processors
2519 Use this attribute on the AVR to indicate that the specified
2520 function is a signal handler. The compiler will generate function
2521 entry and exit sequences suitable for use in a signal handler when this
2522 attribute is present. Interrupts will be disabled inside the function.
2525 Use this attribute on the SH to indicate an @code{interrupt_handler}
2526 function should switch to an alternate stack. It expects a string
2527 argument that names a global variable holding the address of the
2532 void f () __attribute__ ((interrupt_handler,
2533 sp_switch ("alt_stack")));
2537 @cindex functions that pop the argument stack on the 386
2538 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2539 assume that the called function will pop off the stack space used to
2540 pass arguments, unless it takes a variable number of arguments.
2543 @cindex tiny data section on the H8/300H and H8S
2544 Use this attribute on the H8/300H and H8S to indicate that the specified
2545 variable should be placed into the tiny data section.
2546 The compiler will generate more efficient code for loads and stores
2547 on data in the tiny data section. Note the tiny data area is limited to
2548 slightly under 32kbytes of data.
2551 Use this attribute on the SH for an @code{interrupt_handler} to return using
2552 @code{trapa} instead of @code{rte}. This attribute expects an integer
2553 argument specifying the trap number to be used.
2556 @cindex @code{unused} attribute.
2557 This attribute, attached to a function, means that the function is meant
2558 to be possibly unused. GCC will not produce a warning for this
2562 @cindex @code{used} attribute.
2563 This attribute, attached to a function, means that code must be emitted
2564 for the function even if it appears that the function is not referenced.
2565 This is useful, for example, when the function is referenced only in
2569 @cindex @code{version_id} attribute on IA64 HP-UX
2570 This attribute, attached to a global variable or function, renames a
2571 symbol to contain a version string, thus allowing for function level
2572 versioning. HP-UX system header files may use version level functioning
2573 for some system calls.
2576 extern int foo () __attribute__((version_id ("20040821")));
2579 Calls to @var{foo} will be mapped to calls to @var{foo@{20040821@}}.
2581 @item visibility ("@var{visibility_type}")
2582 @cindex @code{visibility} attribute
2583 This attribute affects the linkage of the declaration to which it is attached.
2584 There are four supported @var{visibility_type} values: default,
2585 hidden, protected or internal visibility.
2588 void __attribute__ ((visibility ("protected")))
2589 f () @{ /* @r{Do something.} */; @}
2590 int i __attribute__ ((visibility ("hidden")));
2593 The possible values of @var{visibility_type} correspond to the
2594 visibility settings in the ELF gABI.
2597 @c keep this list of visibilities in alphabetical order.
2600 Default visibility is the normal case for the object file format.
2601 This value is available for the visibility attribute to override other
2602 options that may change the assumed visibility of entities.
2604 On ELF, default visibility means that the declaration is visible to other
2605 modules and, in shared libraries, means that the declared entity may be
2608 On Darwin, default visibility means that the declaration is visible to
2611 Default visibility corresponds to ``external linkage'' in the language.
2614 Hidden visibility indicates that the entity declared will have a new
2615 form of linkage, which we'll call ``hidden linkage''. Two
2616 declarations of an object with hidden linkage refer to the same object
2617 if they are in the same shared object.
2620 Internal visibility is like hidden visibility, but with additional
2621 processor specific semantics. Unless otherwise specified by the
2622 psABI, GCC defines internal visibility to mean that a function is
2623 @emph{never} called from another module. Compare this with hidden
2624 functions which, while they cannot be referenced directly by other
2625 modules, can be referenced indirectly via function pointers. By
2626 indicating that a function cannot be called from outside the module,
2627 GCC may for instance omit the load of a PIC register since it is known
2628 that the calling function loaded the correct value.
2631 Protected visibility is like default visibility except that it
2632 indicates that references within the defining module will bind to the
2633 definition in that module. That is, the declared entity cannot be
2634 overridden by another module.
2638 All visibilities are supported on many, but not all, ELF targets
2639 (supported when the assembler supports the @samp{.visibility}
2640 pseudo-op). Default visibility is supported everywhere. Hidden
2641 visibility is supported on Darwin targets.
2643 The visibility attribute should be applied only to declarations which
2644 would otherwise have external linkage. The attribute should be applied
2645 consistently, so that the same entity should not be declared with
2646 different settings of the attribute.
2648 In C++, the visibility attribute applies to types as well as functions
2649 and objects, because in C++ types have linkage. A class must not have
2650 greater visibility than its non-static data member types and bases,
2651 and class members default to the visibility of their class. Also, a
2652 declaration without explicit visibility is limited to the visibility
2655 In C++, you can mark member functions and static member variables of a
2656 class with the visibility attribute. This is useful if if you know a
2657 particular method or static member variable should only be used from
2658 one shared object; then you can mark it hidden while the rest of the
2659 class has default visibility. Care must be taken to avoid breaking
2660 the One Definition Rule; for example, it is usually not useful to mark
2661 an inline method as hidden without marking the whole class as hidden.
2663 A C++ namespace declaration can also have the visibility attribute.
2664 This attribute applies only to the particular namespace body, not to
2665 other definitions of the same namespace; it is equivalent to using
2666 @samp{#pragma GCC visibility} before and after the namespace
2667 definition (@pxref{Visibility Pragmas}).
2669 In C++, if a template argument has limited visibility, this
2670 restriction is implicitly propagated to the template instantiation.
2671 Otherwise, template instantiations and specializations default to the
2672 visibility of their template.
2674 If both the template and enclosing class have explicit visibility, the
2675 visibility from the template is used.
2677 @item warn_unused_result
2678 @cindex @code{warn_unused_result} attribute
2679 The @code{warn_unused_result} attribute causes a warning to be emitted
2680 if a caller of the function with this attribute does not use its
2681 return value. This is useful for functions where not checking
2682 the result is either a security problem or always a bug, such as
2686 int fn () __attribute__ ((warn_unused_result));
2689 if (fn () < 0) return -1;
2695 results in warning on line 5.
2698 @cindex @code{weak} attribute
2699 The @code{weak} attribute causes the declaration to be emitted as a weak
2700 symbol rather than a global. This is primarily useful in defining
2701 library functions which can be overridden in user code, though it can
2702 also be used with non-function declarations. Weak symbols are supported
2703 for ELF targets, and also for a.out targets when using the GNU assembler
2707 @itemx weakref ("@var{target}")
2708 @cindex @code{weakref} attribute
2709 The @code{weakref} attribute marks a declaration as a weak reference.
2710 Without arguments, it should be accompanied by an @code{alias} attribute
2711 naming the target symbol. Optionally, the @var{target} may be given as
2712 an argument to @code{weakref} itself. In either case, @code{weakref}
2713 implicitly marks the declaration as @code{weak}. Without a
2714 @var{target}, given as an argument to @code{weakref} or to @code{alias},
2715 @code{weakref} is equivalent to @code{weak}.
2718 static int x() __attribute__ ((weakref ("y")));
2719 /* is equivalent to... */
2720 static int x() __attribute__ ((weak, weakref, alias ("y")));
2722 static int x() __attribute__ ((weakref));
2723 static int x() __attribute__ ((alias ("y")));
2726 A weak reference is an alias that does not by itself require a
2727 definition to be given for the target symbol. If the target symbol is
2728 only referenced through weak references, then the becomes a @code{weak}
2729 undefined symbol. If it is directly referenced, however, then such
2730 strong references prevail, and a definition will be required for the
2731 symbol, not necessarily in the same translation unit.
2733 The effect is equivalent to moving all references to the alias to a
2734 separate translation unit, renaming the alias to the aliased symbol,
2735 declaring it as weak, compiling the two separate translation units and
2736 performing a reloadable link on them.
2738 At present, a declaration to which @code{weakref} is attached can
2739 only be @code{static}.
2741 @item externally_visible
2742 @cindex @code{externally_visible} attribute.
2743 This attribute, attached to a global variable or function nullify
2744 effect of @option{-fwhole-program} command line option, so the object
2745 remain visible outside the current compilation unit
2749 You can specify multiple attributes in a declaration by separating them
2750 by commas within the double parentheses or by immediately following an
2751 attribute declaration with another attribute declaration.
2753 @cindex @code{#pragma}, reason for not using
2754 @cindex pragma, reason for not using
2755 Some people object to the @code{__attribute__} feature, suggesting that
2756 ISO C's @code{#pragma} should be used instead. At the time
2757 @code{__attribute__} was designed, there were two reasons for not doing
2762 It is impossible to generate @code{#pragma} commands from a macro.
2765 There is no telling what the same @code{#pragma} might mean in another
2769 These two reasons applied to almost any application that might have been
2770 proposed for @code{#pragma}. It was basically a mistake to use
2771 @code{#pragma} for @emph{anything}.
2773 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2774 to be generated from macros. In addition, a @code{#pragma GCC}
2775 namespace is now in use for GCC-specific pragmas. However, it has been
2776 found convenient to use @code{__attribute__} to achieve a natural
2777 attachment of attributes to their corresponding declarations, whereas
2778 @code{#pragma GCC} is of use for constructs that do not naturally form
2779 part of the grammar. @xref{Other Directives,,Miscellaneous
2780 Preprocessing Directives, cpp, The GNU C Preprocessor}.
2782 @node Attribute Syntax
2783 @section Attribute Syntax
2784 @cindex attribute syntax
2786 This section describes the syntax with which @code{__attribute__} may be
2787 used, and the constructs to which attribute specifiers bind, for the C
2788 language. Some details may vary for C++ and Objective-C@. Because of
2789 infelicities in the grammar for attributes, some forms described here
2790 may not be successfully parsed in all cases.
2792 There are some problems with the semantics of attributes in C++. For
2793 example, there are no manglings for attributes, although they may affect
2794 code generation, so problems may arise when attributed types are used in
2795 conjunction with templates or overloading. Similarly, @code{typeid}
2796 does not distinguish between types with different attributes. Support
2797 for attributes in C++ may be restricted in future to attributes on
2798 declarations only, but not on nested declarators.
2800 @xref{Function Attributes}, for details of the semantics of attributes
2801 applying to functions. @xref{Variable Attributes}, for details of the
2802 semantics of attributes applying to variables. @xref{Type Attributes},
2803 for details of the semantics of attributes applying to structure, union
2804 and enumerated types.
2806 An @dfn{attribute specifier} is of the form
2807 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2808 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2809 each attribute is one of the following:
2813 Empty. Empty attributes are ignored.
2816 A word (which may be an identifier such as @code{unused}, or a reserved
2817 word such as @code{const}).
2820 A word, followed by, in parentheses, parameters for the attribute.
2821 These parameters take one of the following forms:
2825 An identifier. For example, @code{mode} attributes use this form.
2828 An identifier followed by a comma and a non-empty comma-separated list
2829 of expressions. For example, @code{format} attributes use this form.
2832 A possibly empty comma-separated list of expressions. For example,
2833 @code{format_arg} attributes use this form with the list being a single
2834 integer constant expression, and @code{alias} attributes use this form
2835 with the list being a single string constant.
2839 An @dfn{attribute specifier list} is a sequence of one or more attribute
2840 specifiers, not separated by any other tokens.
2842 In GNU C, an attribute specifier list may appear after the colon following a
2843 label, other than a @code{case} or @code{default} label. The only
2844 attribute it makes sense to use after a label is @code{unused}. This
2845 feature is intended for code generated by programs which contains labels
2846 that may be unused but which is compiled with @option{-Wall}. It would
2847 not normally be appropriate to use in it human-written code, though it
2848 could be useful in cases where the code that jumps to the label is
2849 contained within an @code{#ifdef} conditional. GNU C++ does not permit
2850 such placement of attribute lists, as it is permissible for a
2851 declaration, which could begin with an attribute list, to be labelled in
2852 C++. Declarations cannot be labelled in C90 or C99, so the ambiguity
2853 does not arise there.
2855 An attribute specifier list may appear as part of a @code{struct},
2856 @code{union} or @code{enum} specifier. It may go either immediately
2857 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2858 the closing brace. The former syntax is preferred.
2859 Where attribute specifiers follow the closing brace, they are considered
2860 to relate to the structure, union or enumerated type defined, not to any
2861 enclosing declaration the type specifier appears in, and the type
2862 defined is not complete until after the attribute specifiers.
2863 @c Otherwise, there would be the following problems: a shift/reduce
2864 @c conflict between attributes binding the struct/union/enum and
2865 @c binding to the list of specifiers/qualifiers; and "aligned"
2866 @c attributes could use sizeof for the structure, but the size could be
2867 @c changed later by "packed" attributes.
2869 Otherwise, an attribute specifier appears as part of a declaration,
2870 counting declarations of unnamed parameters and type names, and relates
2871 to that declaration (which may be nested in another declaration, for
2872 example in the case of a parameter declaration), or to a particular declarator
2873 within a declaration. Where an
2874 attribute specifier is applied to a parameter declared as a function or
2875 an array, it should apply to the function or array rather than the
2876 pointer to which the parameter is implicitly converted, but this is not
2877 yet correctly implemented.
2879 Any list of specifiers and qualifiers at the start of a declaration may
2880 contain attribute specifiers, whether or not such a list may in that
2881 context contain storage class specifiers. (Some attributes, however,
2882 are essentially in the nature of storage class specifiers, and only make
2883 sense where storage class specifiers may be used; for example,
2884 @code{section}.) There is one necessary limitation to this syntax: the
2885 first old-style parameter declaration in a function definition cannot
2886 begin with an attribute specifier, because such an attribute applies to
2887 the function instead by syntax described below (which, however, is not
2888 yet implemented in this case). In some other cases, attribute
2889 specifiers are permitted by this grammar but not yet supported by the
2890 compiler. All attribute specifiers in this place relate to the
2891 declaration as a whole. In the obsolescent usage where a type of
2892 @code{int} is implied by the absence of type specifiers, such a list of
2893 specifiers and qualifiers may be an attribute specifier list with no
2894 other specifiers or qualifiers.
2896 At present, the first parameter in a function prototype must have some
2897 type specifier which is not an attribute specifier; this resolves an
2898 ambiguity in the interpretation of @code{void f(int
2899 (__attribute__((foo)) x))}, but is subject to change. At present, if
2900 the parentheses of a function declarator contain only attributes then
2901 those attributes are ignored, rather than yielding an error or warning
2902 or implying a single parameter of type int, but this is subject to
2905 An attribute specifier list may appear immediately before a declarator
2906 (other than the first) in a comma-separated list of declarators in a
2907 declaration of more than one identifier using a single list of
2908 specifiers and qualifiers. Such attribute specifiers apply
2909 only to the identifier before whose declarator they appear. For
2913 __attribute__((noreturn)) void d0 (void),
2914 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2919 the @code{noreturn} attribute applies to all the functions
2920 declared; the @code{format} attribute only applies to @code{d1}.
2922 An attribute specifier list may appear immediately before the comma,
2923 @code{=} or semicolon terminating the declaration of an identifier other
2924 than a function definition. At present, such attribute specifiers apply
2925 to the declared object or function, but in future they may attach to the
2926 outermost adjacent declarator. In simple cases there is no difference,
2927 but, for example, in
2930 void (****f)(void) __attribute__((noreturn));
2934 at present the @code{noreturn} attribute applies to @code{f}, which
2935 causes a warning since @code{f} is not a function, but in future it may
2936 apply to the function @code{****f}. The precise semantics of what
2937 attributes in such cases will apply to are not yet specified. Where an
2938 assembler name for an object or function is specified (@pxref{Asm
2939 Labels}), at present the attribute must follow the @code{asm}
2940 specification; in future, attributes before the @code{asm} specification
2941 may apply to the adjacent declarator, and those after it to the declared
2944 An attribute specifier list may, in future, be permitted to appear after
2945 the declarator in a function definition (before any old-style parameter
2946 declarations or the function body).
2948 Attribute specifiers may be mixed with type qualifiers appearing inside
2949 the @code{[]} of a parameter array declarator, in the C99 construct by
2950 which such qualifiers are applied to the pointer to which the array is
2951 implicitly converted. Such attribute specifiers apply to the pointer,
2952 not to the array, but at present this is not implemented and they are
2955 An attribute specifier list may appear at the start of a nested
2956 declarator. At present, there are some limitations in this usage: the
2957 attributes correctly apply to the declarator, but for most individual
2958 attributes the semantics this implies are not implemented.
2959 When attribute specifiers follow the @code{*} of a pointer
2960 declarator, they may be mixed with any type qualifiers present.
2961 The following describes the formal semantics of this syntax. It will make the
2962 most sense if you are familiar with the formal specification of
2963 declarators in the ISO C standard.
2965 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2966 D1}, where @code{T} contains declaration specifiers that specify a type
2967 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2968 contains an identifier @var{ident}. The type specified for @var{ident}
2969 for derived declarators whose type does not include an attribute
2970 specifier is as in the ISO C standard.
2972 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2973 and the declaration @code{T D} specifies the type
2974 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2975 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2976 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2978 If @code{D1} has the form @code{*
2979 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2980 declaration @code{T D} specifies the type
2981 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2982 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2983 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2989 void (__attribute__((noreturn)) ****f) (void);
2993 specifies the type ``pointer to pointer to pointer to pointer to
2994 non-returning function returning @code{void}''. As another example,
2997 char *__attribute__((aligned(8))) *f;
3001 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
3002 Note again that this does not work with most attributes; for example,
3003 the usage of @samp{aligned} and @samp{noreturn} attributes given above
3004 is not yet supported.
3006 For compatibility with existing code written for compiler versions that
3007 did not implement attributes on nested declarators, some laxity is
3008 allowed in the placing of attributes. If an attribute that only applies
3009 to types is applied to a declaration, it will be treated as applying to
3010 the type of that declaration. If an attribute that only applies to
3011 declarations is applied to the type of a declaration, it will be treated
3012 as applying to that declaration; and, for compatibility with code
3013 placing the attributes immediately before the identifier declared, such
3014 an attribute applied to a function return type will be treated as
3015 applying to the function type, and such an attribute applied to an array
3016 element type will be treated as applying to the array type. If an
3017 attribute that only applies to function types is applied to a
3018 pointer-to-function type, it will be treated as applying to the pointer
3019 target type; if such an attribute is applied to a function return type
3020 that is not a pointer-to-function type, it will be treated as applying
3021 to the function type.
3023 @node Function Prototypes
3024 @section Prototypes and Old-Style Function Definitions
3025 @cindex function prototype declarations
3026 @cindex old-style function definitions
3027 @cindex promotion of formal parameters
3029 GNU C extends ISO C to allow a function prototype to override a later
3030 old-style non-prototype definition. Consider the following example:
3033 /* @r{Use prototypes unless the compiler is old-fashioned.} */
3040 /* @r{Prototype function declaration.} */
3041 int isroot P((uid_t));
3043 /* @r{Old-style function definition.} */
3045 isroot (x) /* @r{??? lossage here ???} */
3052 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
3053 not allow this example, because subword arguments in old-style
3054 non-prototype definitions are promoted. Therefore in this example the
3055 function definition's argument is really an @code{int}, which does not
3056 match the prototype argument type of @code{short}.
3058 This restriction of ISO C makes it hard to write code that is portable
3059 to traditional C compilers, because the programmer does not know
3060 whether the @code{uid_t} type is @code{short}, @code{int}, or
3061 @code{long}. Therefore, in cases like these GNU C allows a prototype
3062 to override a later old-style definition. More precisely, in GNU C, a
3063 function prototype argument type overrides the argument type specified
3064 by a later old-style definition if the former type is the same as the
3065 latter type before promotion. Thus in GNU C the above example is
3066 equivalent to the following:
3079 GNU C++ does not support old-style function definitions, so this
3080 extension is irrelevant.
3083 @section C++ Style Comments
3085 @cindex C++ comments
3086 @cindex comments, C++ style
3088 In GNU C, you may use C++ style comments, which start with @samp{//} and
3089 continue until the end of the line. Many other C implementations allow
3090 such comments, and they are included in the 1999 C standard. However,
3091 C++ style comments are not recognized if you specify an @option{-std}
3092 option specifying a version of ISO C before C99, or @option{-ansi}
3093 (equivalent to @option{-std=c89}).
3096 @section Dollar Signs in Identifier Names
3098 @cindex dollar signs in identifier names
3099 @cindex identifier names, dollar signs in
3101 In GNU C, you may normally use dollar signs in identifier names.
3102 This is because many traditional C implementations allow such identifiers.
3103 However, dollar signs in identifiers are not supported on a few target
3104 machines, typically because the target assembler does not allow them.
3106 @node Character Escapes
3107 @section The Character @key{ESC} in Constants
3109 You can use the sequence @samp{\e} in a string or character constant to
3110 stand for the ASCII character @key{ESC}.
3113 @section Inquiring on Alignment of Types or Variables
3115 @cindex type alignment
3116 @cindex variable alignment
3118 The keyword @code{__alignof__} allows you to inquire about how an object
3119 is aligned, or the minimum alignment usually required by a type. Its
3120 syntax is just like @code{sizeof}.
3122 For example, if the target machine requires a @code{double} value to be
3123 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
3124 This is true on many RISC machines. On more traditional machine
3125 designs, @code{__alignof__ (double)} is 4 or even 2.
3127 Some machines never actually require alignment; they allow reference to any
3128 data type even at an odd address. For these machines, @code{__alignof__}
3129 reports the @emph{recommended} alignment of a type.
3131 If the operand of @code{__alignof__} is an lvalue rather than a type,
3132 its value is the required alignment for its type, taking into account
3133 any minimum alignment specified with GCC's @code{__attribute__}
3134 extension (@pxref{Variable Attributes}). For example, after this
3138 struct foo @{ int x; char y; @} foo1;
3142 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
3143 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
3145 It is an error to ask for the alignment of an incomplete type.
3147 @node Variable Attributes
3148 @section Specifying Attributes of Variables
3149 @cindex attribute of variables
3150 @cindex variable attributes
3152 The keyword @code{__attribute__} allows you to specify special
3153 attributes of variables or structure fields. This keyword is followed
3154 by an attribute specification inside double parentheses. Some
3155 attributes are currently defined generically for variables.
3156 Other attributes are defined for variables on particular target
3157 systems. Other attributes are available for functions
3158 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
3159 Other front ends might define more attributes
3160 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
3162 You may also specify attributes with @samp{__} preceding and following
3163 each keyword. This allows you to use them in header files without
3164 being concerned about a possible macro of the same name. For example,
3165 you may use @code{__aligned__} instead of @code{aligned}.
3167 @xref{Attribute Syntax}, for details of the exact syntax for using
3171 @cindex @code{aligned} attribute
3172 @item aligned (@var{alignment})
3173 This attribute specifies a minimum alignment for the variable or
3174 structure field, measured in bytes. For example, the declaration:
3177 int x __attribute__ ((aligned (16))) = 0;
3181 causes the compiler to allocate the global variable @code{x} on a
3182 16-byte boundary. On a 68040, this could be used in conjunction with
3183 an @code{asm} expression to access the @code{move16} instruction which
3184 requires 16-byte aligned operands.
3186 You can also specify the alignment of structure fields. For example, to
3187 create a double-word aligned @code{int} pair, you could write:
3190 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
3194 This is an alternative to creating a union with a @code{double} member
3195 that forces the union to be double-word aligned.
3197 As in the preceding examples, you can explicitly specify the alignment
3198 (in bytes) that you wish the compiler to use for a given variable or
3199 structure field. Alternatively, you can leave out the alignment factor
3200 and just ask the compiler to align a variable or field to the maximum
3201 useful alignment for the target machine you are compiling for. For
3202 example, you could write:
3205 short array[3] __attribute__ ((aligned));
3208 Whenever you leave out the alignment factor in an @code{aligned} attribute
3209 specification, the compiler automatically sets the alignment for the declared
3210 variable or field to the largest alignment which is ever used for any data
3211 type on the target machine you are compiling for. Doing this can often make
3212 copy operations more efficient, because the compiler can use whatever
3213 instructions copy the biggest chunks of memory when performing copies to
3214 or from the variables or fields that you have aligned this way.
3216 When used on a struct, or struct member, the @code{aligned} attribute can
3217 only increase the alignment; in order to decrease it, the @code{packed}
3218 attribute must be specified as well. When used as part of a typedef, the
3219 @code{aligned} attribute can both increase and decrease alignment, and
3220 specifying the @code{packed} attribute will generate a warning.
3222 Note that the effectiveness of @code{aligned} attributes may be limited
3223 by inherent limitations in your linker. On many systems, the linker is
3224 only able to arrange for variables to be aligned up to a certain maximum
3225 alignment. (For some linkers, the maximum supported alignment may
3226 be very very small.) If your linker is only able to align variables
3227 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3228 in an @code{__attribute__} will still only provide you with 8 byte
3229 alignment. See your linker documentation for further information.
3231 The @code{aligned} attribute can also be used for functions
3232 (@pxref{Function Attributes}.)
3234 @item cleanup (@var{cleanup_function})
3235 @cindex @code{cleanup} attribute
3236 The @code{cleanup} attribute runs a function when the variable goes
3237 out of scope. This attribute can only be applied to auto function
3238 scope variables; it may not be applied to parameters or variables
3239 with static storage duration. The function must take one parameter,
3240 a pointer to a type compatible with the variable. The return value
3241 of the function (if any) is ignored.
3243 If @option{-fexceptions} is enabled, then @var{cleanup_function}
3244 will be run during the stack unwinding that happens during the
3245 processing of the exception. Note that the @code{cleanup} attribute
3246 does not allow the exception to be caught, only to perform an action.
3247 It is undefined what happens if @var{cleanup_function} does not
3252 @cindex @code{common} attribute
3253 @cindex @code{nocommon} attribute
3256 The @code{common} attribute requests GCC to place a variable in
3257 ``common'' storage. The @code{nocommon} attribute requests the
3258 opposite---to allocate space for it directly.
3260 These attributes override the default chosen by the
3261 @option{-fno-common} and @option{-fcommon} flags respectively.
3264 @cindex @code{deprecated} attribute
3265 The @code{deprecated} attribute results in a warning if the variable
3266 is used anywhere in the source file. This is useful when identifying
3267 variables that are expected to be removed in a future version of a
3268 program. The warning also includes the location of the declaration
3269 of the deprecated variable, to enable users to easily find further
3270 information about why the variable is deprecated, or what they should
3271 do instead. Note that the warning only occurs for uses:
3274 extern int old_var __attribute__ ((deprecated));
3276 int new_fn () @{ return old_var; @}
3279 results in a warning on line 3 but not line 2.
3281 The @code{deprecated} attribute can also be used for functions and
3282 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
3284 @item mode (@var{mode})
3285 @cindex @code{mode} attribute
3286 This attribute specifies the data type for the declaration---whichever
3287 type corresponds to the mode @var{mode}. This in effect lets you
3288 request an integer or floating point type according to its width.
3290 You may also specify a mode of @samp{byte} or @samp{__byte__} to
3291 indicate the mode corresponding to a one-byte integer, @samp{word} or
3292 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
3293 or @samp{__pointer__} for the mode used to represent pointers.
3296 @cindex @code{packed} attribute
3297 The @code{packed} attribute specifies that a variable or structure field
3298 should have the smallest possible alignment---one byte for a variable,
3299 and one bit for a field, unless you specify a larger value with the
3300 @code{aligned} attribute.
3302 Here is a structure in which the field @code{x} is packed, so that it
3303 immediately follows @code{a}:
3309 int x[2] __attribute__ ((packed));
3313 @item section ("@var{section-name}")
3314 @cindex @code{section} variable attribute
3315 Normally, the compiler places the objects it generates in sections like
3316 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
3317 or you need certain particular variables to appear in special sections,
3318 for example to map to special hardware. The @code{section}
3319 attribute specifies that a variable (or function) lives in a particular
3320 section. For example, this small program uses several specific section names:
3323 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
3324 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
3325 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
3326 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3330 /* @r{Initialize stack pointer} */
3331 init_sp (stack + sizeof (stack));
3333 /* @r{Initialize initialized data} */
3334 memcpy (&init_data, &data, &edata - &data);
3336 /* @r{Turn on the serial ports} */
3343 Use the @code{section} attribute with an @emph{initialized} definition
3344 of a @emph{global} variable, as shown in the example. GCC issues
3345 a warning and otherwise ignores the @code{section} attribute in
3346 uninitialized variable declarations.
3348 You may only use the @code{section} attribute with a fully initialized
3349 global definition because of the way linkers work. The linker requires
3350 each object be defined once, with the exception that uninitialized
3351 variables tentatively go in the @code{common} (or @code{bss}) section
3352 and can be multiply ``defined''. You can force a variable to be
3353 initialized with the @option{-fno-common} flag or the @code{nocommon}
3356 Some file formats do not support arbitrary sections so the @code{section}
3357 attribute is not available on all platforms.
3358 If you need to map the entire contents of a module to a particular
3359 section, consider using the facilities of the linker instead.
3362 @cindex @code{shared} variable attribute
3363 On Microsoft Windows, in addition to putting variable definitions in a named
3364 section, the section can also be shared among all running copies of an
3365 executable or DLL@. For example, this small program defines shared data
3366 by putting it in a named section @code{shared} and marking the section
3370 int foo __attribute__((section ("shared"), shared)) = 0;
3375 /* @r{Read and write foo. All running
3376 copies see the same value.} */
3382 You may only use the @code{shared} attribute along with @code{section}
3383 attribute with a fully initialized global definition because of the way
3384 linkers work. See @code{section} attribute for more information.
3386 The @code{shared} attribute is only available on Microsoft Windows@.
3388 @item tls_model ("@var{tls_model}")
3389 @cindex @code{tls_model} attribute
3390 The @code{tls_model} attribute sets thread-local storage model
3391 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
3392 overriding @option{-ftls-model=} command line switch on a per-variable
3394 The @var{tls_model} argument should be one of @code{global-dynamic},
3395 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
3397 Not all targets support this attribute.
3400 This attribute, attached to a variable, means that the variable is meant
3401 to be possibly unused. GCC will not produce a warning for this
3405 This attribute, attached to a variable, means that the variable must be
3406 emitted even if it appears that the variable is not referenced.
3408 @item vector_size (@var{bytes})
3409 This attribute specifies the vector size for the variable, measured in
3410 bytes. For example, the declaration:
3413 int foo __attribute__ ((vector_size (16)));
3417 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3418 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3419 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3421 This attribute is only applicable to integral and float scalars,
3422 although arrays, pointers, and function return values are allowed in
3423 conjunction with this construct.
3425 Aggregates with this attribute are invalid, even if they are of the same
3426 size as a corresponding scalar. For example, the declaration:
3429 struct S @{ int a; @};
3430 struct S __attribute__ ((vector_size (16))) foo;
3434 is invalid even if the size of the structure is the same as the size of
3438 The @code{selectany} attribute causes an initialized global variable to
3439 have link-once semantics. When multiple definitions of the variable are
3440 encountered by the linker, the first is selected and the remainder are
3441 discarded. Following usage by the Microsoft compiler, the linker is told
3442 @emph{not} to warn about size or content differences of the multiple
3445 Although the primary usage of this attribute is for POD types, the
3446 attribute can also be applied to global C++ objects that are initialized
3447 by a constructor. In this case, the static initialization and destruction
3448 code for the object is emitted in each translation defining the object,
3449 but the calls to the constructor and destructor are protected by a
3450 link-once guard variable.
3452 The @code{selectany} attribute is only available on Microsoft Windows
3453 targets. You can use @code{__declspec (selectany)} as a synonym for
3454 @code{__attribute__ ((selectany))} for compatibility with other
3458 The @code{weak} attribute is described in @xref{Function Attributes}.
3461 The @code{dllimport} attribute is described in @xref{Function Attributes}.
3464 The @code{dllexport} attribute is described in @xref{Function Attributes}.
3468 @subsection M32R/D Variable Attributes
3470 One attribute is currently defined for the M32R/D@.
3473 @item model (@var{model-name})
3474 @cindex variable addressability on the M32R/D
3475 Use this attribute on the M32R/D to set the addressability of an object.
3476 The identifier @var{model-name} is one of @code{small}, @code{medium},
3477 or @code{large}, representing each of the code models.
3479 Small model objects live in the lower 16MB of memory (so that their
3480 addresses can be loaded with the @code{ld24} instruction).
3482 Medium and large model objects may live anywhere in the 32-bit address space
3483 (the compiler will generate @code{seth/add3} instructions to load their
3487 @anchor{i386 Variable Attributes}
3488 @subsection i386 Variable Attributes
3490 Two attributes are currently defined for i386 configurations:
3491 @code{ms_struct} and @code{gcc_struct}
3496 @cindex @code{ms_struct} attribute
3497 @cindex @code{gcc_struct} attribute
3499 If @code{packed} is used on a structure, or if bit-fields are used
3500 it may be that the Microsoft ABI packs them differently
3501 than GCC would normally pack them. Particularly when moving packed
3502 data between functions compiled with GCC and the native Microsoft compiler
3503 (either via function call or as data in a file), it may be necessary to access
3506 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3507 compilers to match the native Microsoft compiler.
3509 The Microsoft structure layout algorithm is fairly simple with the exception
3510 of the bitfield packing:
3512 The padding and alignment of members of structures and whether a bit field
3513 can straddle a storage-unit boundary
3516 @item Structure members are stored sequentially in the order in which they are
3517 declared: the first member has the lowest memory address and the last member
3520 @item Every data object has an alignment-requirement. The alignment-requirement
3521 for all data except structures, unions, and arrays is either the size of the
3522 object or the current packing size (specified with either the aligned attribute
3523 or the pack pragma), whichever is less. For structures, unions, and arrays,
3524 the alignment-requirement is the largest alignment-requirement of its members.
3525 Every object is allocated an offset so that:
3527 offset % alignment-requirement == 0
3529 @item Adjacent bit fields are packed into the same 1-, 2-, or 4-byte allocation
3530 unit if the integral types are the same size and if the next bit field fits
3531 into the current allocation unit without crossing the boundary imposed by the
3532 common alignment requirements of the bit fields.
3535 Handling of zero-length bitfields:
3537 MSVC interprets zero-length bitfields in the following ways:
3540 @item If a zero-length bitfield is inserted between two bitfields that would
3541 normally be coalesced, the bitfields will not be coalesced.
3548 unsigned long bf_1 : 12;
3550 unsigned long bf_2 : 12;
3554 The size of @code{t1} would be 8 bytes with the zero-length bitfield. If the
3555 zero-length bitfield were removed, @code{t1}'s size would be 4 bytes.
3557 @item If a zero-length bitfield is inserted after a bitfield, @code{foo}, and the
3558 alignment of the zero-length bitfield is greater than the member that follows it,
3559 @code{bar}, @code{bar} will be aligned as the type of the zero-length bitfield.
3579 For @code{t2}, @code{bar} will be placed at offset 2, rather than offset 1.
3580 Accordingly, the size of @code{t2} will be 4. For @code{t3}, the zero-length
3581 bitfield will not affect the alignment of @code{bar} or, as a result, the size
3584 Taking this into account, it is important to note the following:
3587 @item If a zero-length bitfield follows a normal bitfield, the type of the
3588 zero-length bitfield may affect the alignment of the structure as whole. For
3589 example, @code{t2} has a size of 4 bytes, since the zero-length bitfield follows a
3590 normal bitfield, and is of type short.
3592 @item Even if a zero-length bitfield is not followed by a normal bitfield, it may
3593 still affect the alignment of the structure:
3603 Here, @code{t4} will take up 4 bytes.
3606 @item Zero-length bitfields following non-bitfield members are ignored:
3617 Here, @code{t5} will take up 2 bytes.
3621 @subsection PowerPC Variable Attributes
3623 Three attributes currently are defined for PowerPC configurations:
3624 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
3626 For full documentation of the struct attributes please see the
3627 documentation in the @xref{i386 Variable Attributes}, section.
3629 For documentation of @code{altivec} attribute please see the
3630 documentation in the @xref{PowerPC Type Attributes}, section.
3632 @subsection SPU Variable Attributes
3634 The SPU supports the @code{spu_vector} attribute for variables. For
3635 documentation of this attribute please see the documentation in the
3636 @xref{SPU Type Attributes}, section.
3638 @subsection Xstormy16 Variable Attributes
3640 One attribute is currently defined for xstormy16 configurations:
3645 @cindex @code{below100} attribute
3647 If a variable has the @code{below100} attribute (@code{BELOW100} is
3648 allowed also), GCC will place the variable in the first 0x100 bytes of
3649 memory and use special opcodes to access it. Such variables will be
3650 placed in either the @code{.bss_below100} section or the
3651 @code{.data_below100} section.
3655 @node Type Attributes
3656 @section Specifying Attributes of Types
3657 @cindex attribute of types
3658 @cindex type attributes
3660 The keyword @code{__attribute__} allows you to specify special
3661 attributes of @code{struct} and @code{union} types when you define
3662 such types. This keyword is followed by an attribute specification
3663 inside double parentheses. Seven attributes are currently defined for
3664 types: @code{aligned}, @code{packed}, @code{transparent_union},
3665 @code{unused}, @code{deprecated}, @code{visibility}, and
3666 @code{may_alias}. Other attributes are defined for functions
3667 (@pxref{Function Attributes}) and for variables (@pxref{Variable
3670 You may also specify any one of these attributes with @samp{__}
3671 preceding and following its keyword. This allows you to use these
3672 attributes in header files without being concerned about a possible
3673 macro of the same name. For example, you may use @code{__aligned__}
3674 instead of @code{aligned}.
3676 You may specify type attributes either in a @code{typedef} declaration
3677 or in an enum, struct or union type declaration or definition.
3679 For an enum, struct or union type, you may specify attributes either
3680 between the enum, struct or union tag and the name of the type, or
3681 just past the closing curly brace of the @emph{definition}. The
3682 former syntax is preferred.
3684 @xref{Attribute Syntax}, for details of the exact syntax for using
3688 @cindex @code{aligned} attribute
3689 @item aligned (@var{alignment})
3690 This attribute specifies a minimum alignment (in bytes) for variables
3691 of the specified type. For example, the declarations:
3694 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3695 typedef int more_aligned_int __attribute__ ((aligned (8)));
3699 force the compiler to insure (as far as it can) that each variable whose
3700 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3701 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
3702 variables of type @code{struct S} aligned to 8-byte boundaries allows
3703 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3704 store) instructions when copying one variable of type @code{struct S} to
3705 another, thus improving run-time efficiency.
3707 Note that the alignment of any given @code{struct} or @code{union} type
3708 is required by the ISO C standard to be at least a perfect multiple of
3709 the lowest common multiple of the alignments of all of the members of
3710 the @code{struct} or @code{union} in question. This means that you @emph{can}
3711 effectively adjust the alignment of a @code{struct} or @code{union}
3712 type by attaching an @code{aligned} attribute to any one of the members
3713 of such a type, but the notation illustrated in the example above is a
3714 more obvious, intuitive, and readable way to request the compiler to
3715 adjust the alignment of an entire @code{struct} or @code{union} type.
3717 As in the preceding example, you can explicitly specify the alignment
3718 (in bytes) that you wish the compiler to use for a given @code{struct}
3719 or @code{union} type. Alternatively, you can leave out the alignment factor
3720 and just ask the compiler to align a type to the maximum
3721 useful alignment for the target machine you are compiling for. For
3722 example, you could write:
3725 struct S @{ short f[3]; @} __attribute__ ((aligned));
3728 Whenever you leave out the alignment factor in an @code{aligned}
3729 attribute specification, the compiler automatically sets the alignment
3730 for the type to the largest alignment which is ever used for any data
3731 type on the target machine you are compiling for. Doing this can often
3732 make copy operations more efficient, because the compiler can use
3733 whatever instructions copy the biggest chunks of memory when performing
3734 copies to or from the variables which have types that you have aligned
3737 In the example above, if the size of each @code{short} is 2 bytes, then
3738 the size of the entire @code{struct S} type is 6 bytes. The smallest
3739 power of two which is greater than or equal to that is 8, so the
3740 compiler sets the alignment for the entire @code{struct S} type to 8
3743 Note that although you can ask the compiler to select a time-efficient
3744 alignment for a given type and then declare only individual stand-alone
3745 objects of that type, the compiler's ability to select a time-efficient
3746 alignment is primarily useful only when you plan to create arrays of
3747 variables having the relevant (efficiently aligned) type. If you
3748 declare or use arrays of variables of an efficiently-aligned type, then
3749 it is likely that your program will also be doing pointer arithmetic (or
3750 subscripting, which amounts to the same thing) on pointers to the
3751 relevant type, and the code that the compiler generates for these
3752 pointer arithmetic operations will often be more efficient for
3753 efficiently-aligned types than for other types.
3755 The @code{aligned} attribute can only increase the alignment; but you
3756 can decrease it by specifying @code{packed} as well. See below.
3758 Note that the effectiveness of @code{aligned} attributes may be limited
3759 by inherent limitations in your linker. On many systems, the linker is
3760 only able to arrange for variables to be aligned up to a certain maximum
3761 alignment. (For some linkers, the maximum supported alignment may
3762 be very very small.) If your linker is only able to align variables
3763 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3764 in an @code{__attribute__} will still only provide you with 8 byte
3765 alignment. See your linker documentation for further information.
3768 This attribute, attached to @code{struct} or @code{union} type
3769 definition, specifies that each member (other than zero-width bitfields)
3770 of the structure or union is placed to minimize the memory required. When
3771 attached to an @code{enum} definition, it indicates that the smallest
3772 integral type should be used.
3774 @opindex fshort-enums
3775 Specifying this attribute for @code{struct} and @code{union} types is
3776 equivalent to specifying the @code{packed} attribute on each of the
3777 structure or union members. Specifying the @option{-fshort-enums}
3778 flag on the line is equivalent to specifying the @code{packed}
3779 attribute on all @code{enum} definitions.
3781 In the following example @code{struct my_packed_struct}'s members are
3782 packed closely together, but the internal layout of its @code{s} member
3783 is not packed---to do that, @code{struct my_unpacked_struct} would need to
3787 struct my_unpacked_struct
3793 struct __attribute__ ((__packed__)) my_packed_struct
3797 struct my_unpacked_struct s;
3801 You may only specify this attribute on the definition of a @code{enum},
3802 @code{struct} or @code{union}, not on a @code{typedef} which does not
3803 also define the enumerated type, structure or union.
3805 @item transparent_union
3806 This attribute, attached to a @code{union} type definition, indicates
3807 that any function parameter having that union type causes calls to that
3808 function to be treated in a special way.
3810 First, the argument corresponding to a transparent union type can be of
3811 any type in the union; no cast is required. Also, if the union contains
3812 a pointer type, the corresponding argument can be a null pointer
3813 constant or a void pointer expression; and if the union contains a void
3814 pointer type, the corresponding argument can be any pointer expression.
3815 If the union member type is a pointer, qualifiers like @code{const} on
3816 the referenced type must be respected, just as with normal pointer
3819 Second, the argument is passed to the function using the calling
3820 conventions of the first member of the transparent union, not the calling
3821 conventions of the union itself. All members of the union must have the
3822 same machine representation; this is necessary for this argument passing
3825 Transparent unions are designed for library functions that have multiple
3826 interfaces for compatibility reasons. For example, suppose the
3827 @code{wait} function must accept either a value of type @code{int *} to
3828 comply with Posix, or a value of type @code{union wait *} to comply with
3829 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3830 @code{wait} would accept both kinds of arguments, but it would also
3831 accept any other pointer type and this would make argument type checking
3832 less useful. Instead, @code{<sys/wait.h>} might define the interface
3840 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3842 pid_t wait (wait_status_ptr_t);
3845 This interface allows either @code{int *} or @code{union wait *}
3846 arguments to be passed, using the @code{int *} calling convention.
3847 The program can call @code{wait} with arguments of either type:
3850 int w1 () @{ int w; return wait (&w); @}
3851 int w2 () @{ union wait w; return wait (&w); @}
3854 With this interface, @code{wait}'s implementation might look like this:
3857 pid_t wait (wait_status_ptr_t p)
3859 return waitpid (-1, p.__ip, 0);
3864 When attached to a type (including a @code{union} or a @code{struct}),
3865 this attribute means that variables of that type are meant to appear
3866 possibly unused. GCC will not produce a warning for any variables of
3867 that type, even if the variable appears to do nothing. This is often
3868 the case with lock or thread classes, which are usually defined and then
3869 not referenced, but contain constructors and destructors that have
3870 nontrivial bookkeeping functions.
3873 The @code{deprecated} attribute results in a warning if the type
3874 is used anywhere in the source file. This is useful when identifying
3875 types that are expected to be removed in a future version of a program.
3876 If possible, the warning also includes the location of the declaration
3877 of the deprecated type, to enable users to easily find further
3878 information about why the type is deprecated, or what they should do
3879 instead. Note that the warnings only occur for uses and then only
3880 if the type is being applied to an identifier that itself is not being
3881 declared as deprecated.
3884 typedef int T1 __attribute__ ((deprecated));
3888 typedef T1 T3 __attribute__ ((deprecated));
3889 T3 z __attribute__ ((deprecated));
3892 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3893 warning is issued for line 4 because T2 is not explicitly
3894 deprecated. Line 5 has no warning because T3 is explicitly
3895 deprecated. Similarly for line 6.
3897 The @code{deprecated} attribute can also be used for functions and
3898 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3901 Accesses to objects with types with this attribute are not subjected to
3902 type-based alias analysis, but are instead assumed to be able to alias
3903 any other type of objects, just like the @code{char} type. See
3904 @option{-fstrict-aliasing} for more information on aliasing issues.
3909 typedef short __attribute__((__may_alias__)) short_a;
3915 short_a *b = (short_a *) &a;
3919 if (a == 0x12345678)
3926 If you replaced @code{short_a} with @code{short} in the variable
3927 declaration, the above program would abort when compiled with
3928 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3929 above in recent GCC versions.
3932 In C++, attribute visibility (@pxref{Function Attributes}) can also be
3933 applied to class, struct, union and enum types. Unlike other type
3934 attributes, the attribute must appear between the initial keyword and
3935 the name of the type; it cannot appear after the body of the type.
3937 Note that the type visibility is applied to vague linkage entities
3938 associated with the class (vtable, typeinfo node, etc.). In
3939 particular, if a class is thrown as an exception in one shared object
3940 and caught in another, the class must have default visibility.
3941 Otherwise the two shared objects will be unable to use the same
3942 typeinfo node and exception handling will break.
3944 @subsection ARM Type Attributes
3946 On those ARM targets that support @code{dllimport} (such as Symbian
3947 OS), you can use the @code{notshared} attribute to indicate that the
3948 virtual table and other similar data for a class should not be
3949 exported from a DLL@. For example:
3952 class __declspec(notshared) C @{
3954 __declspec(dllimport) C();
3958 __declspec(dllexport)
3962 In this code, @code{C::C} is exported from the current DLL, but the
3963 virtual table for @code{C} is not exported. (You can use
3964 @code{__attribute__} instead of @code{__declspec} if you prefer, but
3965 most Symbian OS code uses @code{__declspec}.)
3967 @anchor{i386 Type Attributes}
3968 @subsection i386 Type Attributes
3970 Two attributes are currently defined for i386 configurations:
3971 @code{ms_struct} and @code{gcc_struct}
3975 @cindex @code{ms_struct}
3976 @cindex @code{gcc_struct}
3978 If @code{packed} is used on a structure, or if bit-fields are used
3979 it may be that the Microsoft ABI packs them differently
3980 than GCC would normally pack them. Particularly when moving packed
3981 data between functions compiled with GCC and the native Microsoft compiler
3982 (either via function call or as data in a file), it may be necessary to access
3985 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3986 compilers to match the native Microsoft compiler.
3989 To specify multiple attributes, separate them by commas within the
3990 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3993 @anchor{PowerPC Type Attributes}
3994 @subsection PowerPC Type Attributes
3996 Three attributes currently are defined for PowerPC configurations:
3997 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
3999 For full documentation of the struct attributes please see the
4000 documentation in the @xref{i386 Type Attributes}, section.
4002 The @code{altivec} attribute allows one to declare AltiVec vector data
4003 types supported by the AltiVec Programming Interface Manual. The
4004 attribute requires an argument to specify one of three vector types:
4005 @code{vector__}, @code{pixel__} (always followed by unsigned short),
4006 and @code{bool__} (always followed by unsigned).
4009 __attribute__((altivec(vector__)))
4010 __attribute__((altivec(pixel__))) unsigned short
4011 __attribute__((altivec(bool__))) unsigned
4014 These attributes mainly are intended to support the @code{__vector},
4015 @code{__pixel}, and @code{__bool} AltiVec keywords.
4017 @anchor{SPU Type Attributes}
4018 @subsection SPU Type Attributes
4020 The SPU supports the @code{spu_vector} attribute for types. This attribute
4021 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
4022 Language Extensions Specification. It is intended to support the
4023 @code{__vector} keyword.
4027 @section An Inline Function is As Fast As a Macro
4028 @cindex inline functions
4029 @cindex integrating function code
4031 @cindex macros, inline alternative
4033 By declaring a function inline, you can direct GCC to make
4034 calls to that function faster. One way GCC can achieve this is to
4035 integrate that function's code into the code for its callers. This
4036 makes execution faster by eliminating the function-call overhead; in
4037 addition, if any of the actual argument values are constant, their
4038 known values may permit simplifications at compile time so that not
4039 all of the inline function's code needs to be included. The effect on
4040 code size is less predictable; object code may be larger or smaller
4041 with function inlining, depending on the particular case. You can
4042 also direct GCC to try to integrate all ``simple enough'' functions
4043 into their callers with the option @option{-finline-functions}.
4045 GCC implements three different semantics of declaring a function
4046 inline. One is available with @option{-std=gnu89} or
4047 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
4048 on all inline declarations, another when @option{-std=c99} or
4049 @option{-std=gnu99} (without @option{-fgnu89-inline}), and the third
4050 is used when compiling C++.
4052 To declare a function inline, use the @code{inline} keyword in its
4053 declaration, like this:
4063 If you are writing a header file to be included in ISO C89 programs, write
4064 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
4066 The three types of inlining behave similarly in two important cases:
4067 when the @code{inline} keyword is used on a @code{static} function,
4068 like the example above, and when a function is first declared without
4069 using the @code{inline} keyword and then is defined with
4070 @code{inline}, like this:
4073 extern int inc (int *a);
4081 In both of these common cases, the program behaves the same as if you
4082 had not used the @code{inline} keyword, except for its speed.
4084 @cindex inline functions, omission of
4085 @opindex fkeep-inline-functions
4086 When a function is both inline and @code{static}, if all calls to the
4087 function are integrated into the caller, and the function's address is
4088 never used, then the function's own assembler code is never referenced.
4089 In this case, GCC does not actually output assembler code for the
4090 function, unless you specify the option @option{-fkeep-inline-functions}.
4091 Some calls cannot be integrated for various reasons (in particular,
4092 calls that precede the function's definition cannot be integrated, and
4093 neither can recursive calls within the definition). If there is a
4094 nonintegrated call, then the function is compiled to assembler code as
4095 usual. The function must also be compiled as usual if the program
4096 refers to its address, because that can't be inlined.
4099 Note that certain usages in a function definition can make it unsuitable
4100 for inline substitution. Among these usages are: use of varargs, use of
4101 alloca, use of variable sized data types (@pxref{Variable Length}),
4102 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
4103 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
4104 will warn when a function marked @code{inline} could not be substituted,
4105 and will give the reason for the failure.
4107 @cindex automatic @code{inline} for C++ member fns
4108 @cindex @code{inline} automatic for C++ member fns
4109 @cindex member fns, automatically @code{inline}
4110 @cindex C++ member fns, automatically @code{inline}
4111 @opindex fno-default-inline
4112 As required by ISO C++, GCC considers member functions defined within
4113 the body of a class to be marked inline even if they are
4114 not explicitly declared with the @code{inline} keyword. You can
4115 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
4116 Options,,Options Controlling C++ Dialect}.
4118 GCC does not inline any functions when not optimizing unless you specify
4119 the @samp{always_inline} attribute for the function, like this:
4122 /* @r{Prototype.} */
4123 inline void foo (const char) __attribute__((always_inline));
4126 The remainder of this section is specific to GNU C89 inlining.
4128 @cindex non-static inline function
4129 When an inline function is not @code{static}, then the compiler must assume
4130 that there may be calls from other source files; since a global symbol can
4131 be defined only once in any program, the function must not be defined in
4132 the other source files, so the calls therein cannot be integrated.
4133 Therefore, a non-@code{static} inline function is always compiled on its
4134 own in the usual fashion.
4136 If you specify both @code{inline} and @code{extern} in the function
4137 definition, then the definition is used only for inlining. In no case
4138 is the function compiled on its own, not even if you refer to its
4139 address explicitly. Such an address becomes an external reference, as
4140 if you had only declared the function, and had not defined it.
4142 This combination of @code{inline} and @code{extern} has almost the
4143 effect of a macro. The way to use it is to put a function definition in
4144 a header file with these keywords, and put another copy of the
4145 definition (lacking @code{inline} and @code{extern}) in a library file.
4146 The definition in the header file will cause most calls to the function
4147 to be inlined. If any uses of the function remain, they will refer to
4148 the single copy in the library.
4151 @section Assembler Instructions with C Expression Operands
4152 @cindex extended @code{asm}
4153 @cindex @code{asm} expressions
4154 @cindex assembler instructions
4157 In an assembler instruction using @code{asm}, you can specify the
4158 operands of the instruction using C expressions. This means you need not
4159 guess which registers or memory locations will contain the data you want
4162 You must specify an assembler instruction template much like what
4163 appears in a machine description, plus an operand constraint string for
4166 For example, here is how to use the 68881's @code{fsinx} instruction:
4169 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
4173 Here @code{angle} is the C expression for the input operand while
4174 @code{result} is that of the output operand. Each has @samp{"f"} as its
4175 operand constraint, saying that a floating point register is required.
4176 The @samp{=} in @samp{=f} indicates that the operand is an output; all
4177 output operands' constraints must use @samp{=}. The constraints use the
4178 same language used in the machine description (@pxref{Constraints}).
4180 Each operand is described by an operand-constraint string followed by
4181 the C expression in parentheses. A colon separates the assembler
4182 template from the first output operand and another separates the last
4183 output operand from the first input, if any. Commas separate the
4184 operands within each group. The total number of operands is currently
4185 limited to 30; this limitation may be lifted in some future version of
4188 If there are no output operands but there are input operands, you must
4189 place two consecutive colons surrounding the place where the output
4192 As of GCC version 3.1, it is also possible to specify input and output
4193 operands using symbolic names which can be referenced within the
4194 assembler code. These names are specified inside square brackets
4195 preceding the constraint string, and can be referenced inside the
4196 assembler code using @code{%[@var{name}]} instead of a percentage sign
4197 followed by the operand number. Using named operands the above example
4201 asm ("fsinx %[angle],%[output]"
4202 : [output] "=f" (result)
4203 : [angle] "f" (angle));
4207 Note that the symbolic operand names have no relation whatsoever to
4208 other C identifiers. You may use any name you like, even those of
4209 existing C symbols, but you must ensure that no two operands within the same
4210 assembler construct use the same symbolic name.
4212 Output operand expressions must be lvalues; the compiler can check this.
4213 The input operands need not be lvalues. The compiler cannot check
4214 whether the operands have data types that are reasonable for the
4215 instruction being executed. It does not parse the assembler instruction
4216 template and does not know what it means or even whether it is valid
4217 assembler input. The extended @code{asm} feature is most often used for
4218 machine instructions the compiler itself does not know exist. If
4219 the output expression cannot be directly addressed (for example, it is a
4220 bit-field), your constraint must allow a register. In that case, GCC
4221 will use the register as the output of the @code{asm}, and then store
4222 that register into the output.
4224 The ordinary output operands must be write-only; GCC will assume that
4225 the values in these operands before the instruction are dead and need
4226 not be generated. Extended asm supports input-output or read-write
4227 operands. Use the constraint character @samp{+} to indicate such an
4228 operand and list it with the output operands. You should only use
4229 read-write operands when the constraints for the operand (or the
4230 operand in which only some of the bits are to be changed) allow a
4233 You may, as an alternative, logically split its function into two
4234 separate operands, one input operand and one write-only output
4235 operand. The connection between them is expressed by constraints
4236 which say they need to be in the same location when the instruction
4237 executes. You can use the same C expression for both operands, or
4238 different expressions. For example, here we write the (fictitious)
4239 @samp{combine} instruction with @code{bar} as its read-only source
4240 operand and @code{foo} as its read-write destination:
4243 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
4247 The constraint @samp{"0"} for operand 1 says that it must occupy the
4248 same location as operand 0. A number in constraint is allowed only in
4249 an input operand and it must refer to an output operand.
4251 Only a number in the constraint can guarantee that one operand will be in
4252 the same place as another. The mere fact that @code{foo} is the value
4253 of both operands is not enough to guarantee that they will be in the
4254 same place in the generated assembler code. The following would not
4258 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
4261 Various optimizations or reloading could cause operands 0 and 1 to be in
4262 different registers; GCC knows no reason not to do so. For example, the
4263 compiler might find a copy of the value of @code{foo} in one register and
4264 use it for operand 1, but generate the output operand 0 in a different
4265 register (copying it afterward to @code{foo}'s own address). Of course,
4266 since the register for operand 1 is not even mentioned in the assembler
4267 code, the result will not work, but GCC can't tell that.
4269 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
4270 the operand number for a matching constraint. For example:
4273 asm ("cmoveq %1,%2,%[result]"
4274 : [result] "=r"(result)
4275 : "r" (test), "r"(new), "[result]"(old));
4278 Sometimes you need to make an @code{asm} operand be a specific register,
4279 but there's no matching constraint letter for that register @emph{by
4280 itself}. To force the operand into that register, use a local variable
4281 for the operand and specify the register in the variable declaration.
4282 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
4283 register constraint letter that matches the register:
4286 register int *p1 asm ("r0") = @dots{};
4287 register int *p2 asm ("r1") = @dots{};
4288 register int *result asm ("r0");
4289 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
4292 @anchor{Example of asm with clobbered asm reg}
4293 In the above example, beware that a register that is call-clobbered by
4294 the target ABI will be overwritten by any function call in the
4295 assignment, including library calls for arithmetic operators.
4296 Assuming it is a call-clobbered register, this may happen to @code{r0}
4297 above by the assignment to @code{p2}. If you have to use such a
4298 register, use temporary variables for expressions between the register
4303 register int *p1 asm ("r0") = @dots{};
4304 register int *p2 asm ("r1") = t1;
4305 register int *result asm ("r0");
4306 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
4309 Some instructions clobber specific hard registers. To describe this,
4310 write a third colon after the input operands, followed by the names of
4311 the clobbered hard registers (given as strings). Here is a realistic
4312 example for the VAX:
4315 asm volatile ("movc3 %0,%1,%2"
4316 : /* @r{no outputs} */
4317 : "g" (from), "g" (to), "g" (count)
4318 : "r0", "r1", "r2", "r3", "r4", "r5");
4321 You may not write a clobber description in a way that overlaps with an
4322 input or output operand. For example, you may not have an operand
4323 describing a register class with one member if you mention that register
4324 in the clobber list. Variables declared to live in specific registers
4325 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
4326 have no part mentioned in the clobber description.
4327 There is no way for you to specify that an input
4328 operand is modified without also specifying it as an output
4329 operand. Note that if all the output operands you specify are for this
4330 purpose (and hence unused), you will then also need to specify
4331 @code{volatile} for the @code{asm} construct, as described below, to
4332 prevent GCC from deleting the @code{asm} statement as unused.
4334 If you refer to a particular hardware register from the assembler code,
4335 you will probably have to list the register after the third colon to
4336 tell the compiler the register's value is modified. In some assemblers,
4337 the register names begin with @samp{%}; to produce one @samp{%} in the
4338 assembler code, you must write @samp{%%} in the input.
4340 If your assembler instruction can alter the condition code register, add
4341 @samp{cc} to the list of clobbered registers. GCC on some machines
4342 represents the condition codes as a specific hardware register;
4343 @samp{cc} serves to name this register. On other machines, the
4344 condition code is handled differently, and specifying @samp{cc} has no
4345 effect. But it is valid no matter what the machine.
4347 If your assembler instructions access memory in an unpredictable
4348 fashion, add @samp{memory} to the list of clobbered registers. This
4349 will cause GCC to not keep memory values cached in registers across the
4350 assembler instruction and not optimize stores or loads to that memory.
4351 You will also want to add the @code{volatile} keyword if the memory
4352 affected is not listed in the inputs or outputs of the @code{asm}, as
4353 the @samp{memory} clobber does not count as a side-effect of the
4354 @code{asm}. If you know how large the accessed memory is, you can add
4355 it as input or output but if this is not known, you should add
4356 @samp{memory}. As an example, if you access ten bytes of a string, you
4357 can use a memory input like:
4360 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
4363 Note that in the following example the memory input is necessary,
4364 otherwise GCC might optimize the store to @code{x} away:
4371 asm ("magic stuff accessing an 'int' pointed to by '%1'"
4372 "=&d" (r) : "a" (y), "m" (*y));
4377 You can put multiple assembler instructions together in a single
4378 @code{asm} template, separated by the characters normally used in assembly
4379 code for the system. A combination that works in most places is a newline
4380 to break the line, plus a tab character to move to the instruction field
4381 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
4382 assembler allows semicolons as a line-breaking character. Note that some
4383 assembler dialects use semicolons to start a comment.
4384 The input operands are guaranteed not to use any of the clobbered
4385 registers, and neither will the output operands' addresses, so you can
4386 read and write the clobbered registers as many times as you like. Here
4387 is an example of multiple instructions in a template; it assumes the
4388 subroutine @code{_foo} accepts arguments in registers 9 and 10:
4391 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
4393 : "g" (from), "g" (to)
4397 Unless an output operand has the @samp{&} constraint modifier, GCC
4398 may allocate it in the same register as an unrelated input operand, on
4399 the assumption the inputs are consumed before the outputs are produced.
4400 This assumption may be false if the assembler code actually consists of
4401 more than one instruction. In such a case, use @samp{&} for each output
4402 operand that may not overlap an input. @xref{Modifiers}.
4404 If you want to test the condition code produced by an assembler
4405 instruction, you must include a branch and a label in the @code{asm}
4406 construct, as follows:
4409 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
4415 This assumes your assembler supports local labels, as the GNU assembler
4416 and most Unix assemblers do.
4418 Speaking of labels, jumps from one @code{asm} to another are not
4419 supported. The compiler's optimizers do not know about these jumps, and
4420 therefore they cannot take account of them when deciding how to
4423 @cindex macros containing @code{asm}
4424 Usually the most convenient way to use these @code{asm} instructions is to
4425 encapsulate them in macros that look like functions. For example,
4429 (@{ double __value, __arg = (x); \
4430 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
4435 Here the variable @code{__arg} is used to make sure that the instruction
4436 operates on a proper @code{double} value, and to accept only those
4437 arguments @code{x} which can convert automatically to a @code{double}.
4439 Another way to make sure the instruction operates on the correct data
4440 type is to use a cast in the @code{asm}. This is different from using a
4441 variable @code{__arg} in that it converts more different types. For
4442 example, if the desired type were @code{int}, casting the argument to
4443 @code{int} would accept a pointer with no complaint, while assigning the
4444 argument to an @code{int} variable named @code{__arg} would warn about
4445 using a pointer unless the caller explicitly casts it.
4447 If an @code{asm} has output operands, GCC assumes for optimization
4448 purposes the instruction has no side effects except to change the output
4449 operands. This does not mean instructions with a side effect cannot be
4450 used, but you must be careful, because the compiler may eliminate them
4451 if the output operands aren't used, or move them out of loops, or
4452 replace two with one if they constitute a common subexpression. Also,
4453 if your instruction does have a side effect on a variable that otherwise
4454 appears not to change, the old value of the variable may be reused later
4455 if it happens to be found in a register.
4457 You can prevent an @code{asm} instruction from being deleted
4458 by writing the keyword @code{volatile} after
4459 the @code{asm}. For example:
4462 #define get_and_set_priority(new) \
4464 asm volatile ("get_and_set_priority %0, %1" \
4465 : "=g" (__old) : "g" (new)); \
4470 The @code{volatile} keyword indicates that the instruction has
4471 important side-effects. GCC will not delete a volatile @code{asm} if
4472 it is reachable. (The instruction can still be deleted if GCC can
4473 prove that control-flow will never reach the location of the
4474 instruction.) Note that even a volatile @code{asm} instruction
4475 can be moved relative to other code, including across jump
4476 instructions. For example, on many targets there is a system
4477 register which can be set to control the rounding mode of
4478 floating point operations. You might try
4479 setting it with a volatile @code{asm}, like this PowerPC example:
4482 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
4487 This will not work reliably, as the compiler may move the addition back
4488 before the volatile @code{asm}. To make it work you need to add an
4489 artificial dependency to the @code{asm} referencing a variable in the code
4490 you don't want moved, for example:
4493 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
4497 Similarly, you can't expect a
4498 sequence of volatile @code{asm} instructions to remain perfectly
4499 consecutive. If you want consecutive output, use a single @code{asm}.
4500 Also, GCC will perform some optimizations across a volatile @code{asm}
4501 instruction; GCC does not ``forget everything'' when it encounters
4502 a volatile @code{asm} instruction the way some other compilers do.
4504 An @code{asm} instruction without any output operands will be treated
4505 identically to a volatile @code{asm} instruction.
4507 It is a natural idea to look for a way to give access to the condition
4508 code left by the assembler instruction. However, when we attempted to
4509 implement this, we found no way to make it work reliably. The problem
4510 is that output operands might need reloading, which would result in
4511 additional following ``store'' instructions. On most machines, these
4512 instructions would alter the condition code before there was time to
4513 test it. This problem doesn't arise for ordinary ``test'' and
4514 ``compare'' instructions because they don't have any output operands.
4516 For reasons similar to those described above, it is not possible to give
4517 an assembler instruction access to the condition code left by previous
4520 If you are writing a header file that should be includable in ISO C
4521 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
4524 @subsection Size of an @code{asm}
4526 Some targets require that GCC track the size of each instruction used in
4527 order to generate correct code. Because the final length of an
4528 @code{asm} is only known by the assembler, GCC must make an estimate as
4529 to how big it will be. The estimate is formed by counting the number of
4530 statements in the pattern of the @code{asm} and multiplying that by the
4531 length of the longest instruction on that processor. Statements in the
4532 @code{asm} are identified by newline characters and whatever statement
4533 separator characters are supported by the assembler; on most processors
4534 this is the `@code{;}' character.
4536 Normally, GCC's estimate is perfectly adequate to ensure that correct
4537 code is generated, but it is possible to confuse the compiler if you use
4538 pseudo instructions or assembler macros that expand into multiple real
4539 instructions or if you use assembler directives that expand to more
4540 space in the object file than would be needed for a single instruction.
4541 If this happens then the assembler will produce a diagnostic saying that
4542 a label is unreachable.
4544 @subsection i386 floating point asm operands
4546 There are several rules on the usage of stack-like regs in
4547 asm_operands insns. These rules apply only to the operands that are
4552 Given a set of input regs that die in an asm_operands, it is
4553 necessary to know which are implicitly popped by the asm, and
4554 which must be explicitly popped by gcc.
4556 An input reg that is implicitly popped by the asm must be
4557 explicitly clobbered, unless it is constrained to match an
4561 For any input reg that is implicitly popped by an asm, it is
4562 necessary to know how to adjust the stack to compensate for the pop.
4563 If any non-popped input is closer to the top of the reg-stack than
4564 the implicitly popped reg, it would not be possible to know what the
4565 stack looked like---it's not clear how the rest of the stack ``slides
4568 All implicitly popped input regs must be closer to the top of
4569 the reg-stack than any input that is not implicitly popped.
4571 It is possible that if an input dies in an insn, reload might
4572 use the input reg for an output reload. Consider this example:
4575 asm ("foo" : "=t" (a) : "f" (b));
4578 This asm says that input B is not popped by the asm, and that
4579 the asm pushes a result onto the reg-stack, i.e., the stack is one
4580 deeper after the asm than it was before. But, it is possible that
4581 reload will think that it can use the same reg for both the input and
4582 the output, if input B dies in this insn.
4584 If any input operand uses the @code{f} constraint, all output reg
4585 constraints must use the @code{&} earlyclobber.
4587 The asm above would be written as
4590 asm ("foo" : "=&t" (a) : "f" (b));
4594 Some operands need to be in particular places on the stack. All
4595 output operands fall in this category---there is no other way to
4596 know which regs the outputs appear in unless the user indicates
4597 this in the constraints.
4599 Output operands must specifically indicate which reg an output
4600 appears in after an asm. @code{=f} is not allowed: the operand
4601 constraints must select a class with a single reg.
4604 Output operands may not be ``inserted'' between existing stack regs.
4605 Since no 387 opcode uses a read/write operand, all output operands
4606 are dead before the asm_operands, and are pushed by the asm_operands.
4607 It makes no sense to push anywhere but the top of the reg-stack.
4609 Output operands must start at the top of the reg-stack: output
4610 operands may not ``skip'' a reg.
4613 Some asm statements may need extra stack space for internal
4614 calculations. This can be guaranteed by clobbering stack registers
4615 unrelated to the inputs and outputs.
4619 Here are a couple of reasonable asms to want to write. This asm
4620 takes one input, which is internally popped, and produces two outputs.
4623 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
4626 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
4627 and replaces them with one output. The user must code the @code{st(1)}
4628 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
4631 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
4637 @section Controlling Names Used in Assembler Code
4638 @cindex assembler names for identifiers
4639 @cindex names used in assembler code
4640 @cindex identifiers, names in assembler code
4642 You can specify the name to be used in the assembler code for a C
4643 function or variable by writing the @code{asm} (or @code{__asm__})
4644 keyword after the declarator as follows:
4647 int foo asm ("myfoo") = 2;
4651 This specifies that the name to be used for the variable @code{foo} in
4652 the assembler code should be @samp{myfoo} rather than the usual
4655 On systems where an underscore is normally prepended to the name of a C
4656 function or variable, this feature allows you to define names for the
4657 linker that do not start with an underscore.
4659 It does not make sense to use this feature with a non-static local
4660 variable since such variables do not have assembler names. If you are
4661 trying to put the variable in a particular register, see @ref{Explicit
4662 Reg Vars}. GCC presently accepts such code with a warning, but will
4663 probably be changed to issue an error, rather than a warning, in the
4666 You cannot use @code{asm} in this way in a function @emph{definition}; but
4667 you can get the same effect by writing a declaration for the function
4668 before its definition and putting @code{asm} there, like this:
4671 extern func () asm ("FUNC");
4678 It is up to you to make sure that the assembler names you choose do not
4679 conflict with any other assembler symbols. Also, you must not use a
4680 register name; that would produce completely invalid assembler code. GCC
4681 does not as yet have the ability to store static variables in registers.
4682 Perhaps that will be added.
4684 @node Explicit Reg Vars
4685 @section Variables in Specified Registers
4686 @cindex explicit register variables
4687 @cindex variables in specified registers
4688 @cindex specified registers
4689 @cindex registers, global allocation
4691 GNU C allows you to put a few global variables into specified hardware
4692 registers. You can also specify the register in which an ordinary
4693 register variable should be allocated.
4697 Global register variables reserve registers throughout the program.
4698 This may be useful in programs such as programming language
4699 interpreters which have a couple of global variables that are accessed
4703 Local register variables in specific registers do not reserve the
4704 registers, except at the point where they are used as input or output
4705 operands in an @code{asm} statement and the @code{asm} statement itself is
4706 not deleted. The compiler's data flow analysis is capable of determining
4707 where the specified registers contain live values, and where they are
4708 available for other uses. Stores into local register variables may be deleted
4709 when they appear to be dead according to dataflow analysis. References
4710 to local register variables may be deleted or moved or simplified.
4712 These local variables are sometimes convenient for use with the extended
4713 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
4714 output of the assembler instruction directly into a particular register.
4715 (This will work provided the register you specify fits the constraints
4716 specified for that operand in the @code{asm}.)
4724 @node Global Reg Vars
4725 @subsection Defining Global Register Variables
4726 @cindex global register variables
4727 @cindex registers, global variables in
4729 You can define a global register variable in GNU C like this:
4732 register int *foo asm ("a5");
4736 Here @code{a5} is the name of the register which should be used. Choose a
4737 register which is normally saved and restored by function calls on your
4738 machine, so that library routines will not clobber it.
4740 Naturally the register name is cpu-dependent, so you would need to
4741 conditionalize your program according to cpu type. The register
4742 @code{a5} would be a good choice on a 68000 for a variable of pointer
4743 type. On machines with register windows, be sure to choose a ``global''
4744 register that is not affected magically by the function call mechanism.
4746 In addition, operating systems on one type of cpu may differ in how they
4747 name the registers; then you would need additional conditionals. For
4748 example, some 68000 operating systems call this register @code{%a5}.
4750 Eventually there may be a way of asking the compiler to choose a register
4751 automatically, but first we need to figure out how it should choose and
4752 how to enable you to guide the choice. No solution is evident.
4754 Defining a global register variable in a certain register reserves that
4755 register entirely for this use, at least within the current compilation.
4756 The register will not be allocated for any other purpose in the functions
4757 in the current compilation. The register will not be saved and restored by
4758 these functions. Stores into this register are never deleted even if they
4759 would appear to be dead, but references may be deleted or moved or
4762 It is not safe to access the global register variables from signal
4763 handlers, or from more than one thread of control, because the system
4764 library routines may temporarily use the register for other things (unless
4765 you recompile them specially for the task at hand).
4767 @cindex @code{qsort}, and global register variables
4768 It is not safe for one function that uses a global register variable to
4769 call another such function @code{foo} by way of a third function
4770 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4771 different source file in which the variable wasn't declared). This is
4772 because @code{lose} might save the register and put some other value there.
4773 For example, you can't expect a global register variable to be available in
4774 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4775 might have put something else in that register. (If you are prepared to
4776 recompile @code{qsort} with the same global register variable, you can
4777 solve this problem.)
4779 If you want to recompile @code{qsort} or other source files which do not
4780 actually use your global register variable, so that they will not use that
4781 register for any other purpose, then it suffices to specify the compiler
4782 option @option{-ffixed-@var{reg}}. You need not actually add a global
4783 register declaration to their source code.
4785 A function which can alter the value of a global register variable cannot
4786 safely be called from a function compiled without this variable, because it
4787 could clobber the value the caller expects to find there on return.
4788 Therefore, the function which is the entry point into the part of the
4789 program that uses the global register variable must explicitly save and
4790 restore the value which belongs to its caller.
4792 @cindex register variable after @code{longjmp}
4793 @cindex global register after @code{longjmp}
4794 @cindex value after @code{longjmp}
4797 On most machines, @code{longjmp} will restore to each global register
4798 variable the value it had at the time of the @code{setjmp}. On some
4799 machines, however, @code{longjmp} will not change the value of global
4800 register variables. To be portable, the function that called @code{setjmp}
4801 should make other arrangements to save the values of the global register
4802 variables, and to restore them in a @code{longjmp}. This way, the same
4803 thing will happen regardless of what @code{longjmp} does.
4805 All global register variable declarations must precede all function
4806 definitions. If such a declaration could appear after function
4807 definitions, the declaration would be too late to prevent the register from
4808 being used for other purposes in the preceding functions.
4810 Global register variables may not have initial values, because an
4811 executable file has no means to supply initial contents for a register.
4813 On the SPARC, there are reports that g3 @dots{} g7 are suitable
4814 registers, but certain library functions, such as @code{getwd}, as well
4815 as the subroutines for division and remainder, modify g3 and g4. g1 and
4816 g2 are local temporaries.
4818 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4819 Of course, it will not do to use more than a few of those.
4821 @node Local Reg Vars
4822 @subsection Specifying Registers for Local Variables
4823 @cindex local variables, specifying registers
4824 @cindex specifying registers for local variables
4825 @cindex registers for local variables
4827 You can define a local register variable with a specified register
4831 register int *foo asm ("a5");
4835 Here @code{a5} is the name of the register which should be used. Note
4836 that this is the same syntax used for defining global register
4837 variables, but for a local variable it would appear within a function.
4839 Naturally the register name is cpu-dependent, but this is not a
4840 problem, since specific registers are most often useful with explicit
4841 assembler instructions (@pxref{Extended Asm}). Both of these things
4842 generally require that you conditionalize your program according to
4845 In addition, operating systems on one type of cpu may differ in how they
4846 name the registers; then you would need additional conditionals. For
4847 example, some 68000 operating systems call this register @code{%a5}.
4849 Defining such a register variable does not reserve the register; it
4850 remains available for other uses in places where flow control determines
4851 the variable's value is not live.
4853 This option does not guarantee that GCC will generate code that has
4854 this variable in the register you specify at all times. You may not
4855 code an explicit reference to this register in the @emph{assembler
4856 instruction template} part of an @code{asm} statement and assume it will
4857 always refer to this variable. However, using the variable as an
4858 @code{asm} @emph{operand} guarantees that the specified register is used
4861 Stores into local register variables may be deleted when they appear to be dead
4862 according to dataflow analysis. References to local register variables may
4863 be deleted or moved or simplified.
4865 As for global register variables, it's recommended that you choose a
4866 register which is normally saved and restored by function calls on
4867 your machine, so that library routines will not clobber it. A common
4868 pitfall is to initialize multiple call-clobbered registers with
4869 arbitrary expressions, where a function call or library call for an
4870 arithmetic operator will overwrite a register value from a previous
4871 assignment, for example @code{r0} below:
4873 register int *p1 asm ("r0") = @dots{};
4874 register int *p2 asm ("r1") = @dots{};
4876 In those cases, a solution is to use a temporary variable for
4877 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
4879 @node Alternate Keywords
4880 @section Alternate Keywords
4881 @cindex alternate keywords
4882 @cindex keywords, alternate
4884 @option{-ansi} and the various @option{-std} options disable certain
4885 keywords. This causes trouble when you want to use GNU C extensions, or
4886 a general-purpose header file that should be usable by all programs,
4887 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4888 @code{inline} are not available in programs compiled with
4889 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4890 program compiled with @option{-std=c99}). The ISO C99 keyword
4891 @code{restrict} is only available when @option{-std=gnu99} (which will
4892 eventually be the default) or @option{-std=c99} (or the equivalent
4893 @option{-std=iso9899:1999}) is used.
4895 The way to solve these problems is to put @samp{__} at the beginning and
4896 end of each problematical keyword. For example, use @code{__asm__}
4897 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4899 Other C compilers won't accept these alternative keywords; if you want to
4900 compile with another compiler, you can define the alternate keywords as
4901 macros to replace them with the customary keywords. It looks like this:
4909 @findex __extension__
4911 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4913 prevent such warnings within one expression by writing
4914 @code{__extension__} before the expression. @code{__extension__} has no
4915 effect aside from this.
4917 @node Incomplete Enums
4918 @section Incomplete @code{enum} Types
4920 You can define an @code{enum} tag without specifying its possible values.
4921 This results in an incomplete type, much like what you get if you write
4922 @code{struct foo} without describing the elements. A later declaration
4923 which does specify the possible values completes the type.
4925 You can't allocate variables or storage using the type while it is
4926 incomplete. However, you can work with pointers to that type.
4928 This extension may not be very useful, but it makes the handling of
4929 @code{enum} more consistent with the way @code{struct} and @code{union}
4932 This extension is not supported by GNU C++.
4934 @node Function Names
4935 @section Function Names as Strings
4936 @cindex @code{__func__} identifier
4937 @cindex @code{__FUNCTION__} identifier
4938 @cindex @code{__PRETTY_FUNCTION__} identifier
4940 GCC provides three magic variables which hold the name of the current
4941 function, as a string. The first of these is @code{__func__}, which
4942 is part of the C99 standard:
4945 The identifier @code{__func__} is implicitly declared by the translator
4946 as if, immediately following the opening brace of each function
4947 definition, the declaration
4950 static const char __func__[] = "function-name";
4953 appeared, where function-name is the name of the lexically-enclosing
4954 function. This name is the unadorned name of the function.
4957 @code{__FUNCTION__} is another name for @code{__func__}. Older
4958 versions of GCC recognize only this name. However, it is not
4959 standardized. For maximum portability, we recommend you use
4960 @code{__func__}, but provide a fallback definition with the
4964 #if __STDC_VERSION__ < 199901L
4966 # define __func__ __FUNCTION__
4968 # define __func__ "<unknown>"
4973 In C, @code{__PRETTY_FUNCTION__} is yet another name for
4974 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
4975 the type signature of the function as well as its bare name. For
4976 example, this program:
4980 extern int printf (char *, ...);
4987 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4988 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
5006 __PRETTY_FUNCTION__ = void a::sub(int)
5009 These identifiers are not preprocessor macros. In GCC 3.3 and
5010 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
5011 were treated as string literals; they could be used to initialize
5012 @code{char} arrays, and they could be concatenated with other string
5013 literals. GCC 3.4 and later treat them as variables, like
5014 @code{__func__}. In C++, @code{__FUNCTION__} and
5015 @code{__PRETTY_FUNCTION__} have always been variables.
5017 @node Return Address
5018 @section Getting the Return or Frame Address of a Function
5020 These functions may be used to get information about the callers of a
5023 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
5024 This function returns the return address of the current function, or of
5025 one of its callers. The @var{level} argument is number of frames to
5026 scan up the call stack. A value of @code{0} yields the return address
5027 of the current function, a value of @code{1} yields the return address
5028 of the caller of the current function, and so forth. When inlining
5029 the expected behavior is that the function will return the address of
5030 the function that will be returned to. To work around this behavior use
5031 the @code{noinline} function attribute.
5033 The @var{level} argument must be a constant integer.
5035 On some machines it may be impossible to determine the return address of
5036 any function other than the current one; in such cases, or when the top
5037 of the stack has been reached, this function will return @code{0} or a
5038 random value. In addition, @code{__builtin_frame_address} may be used
5039 to determine if the top of the stack has been reached.
5041 This function should only be used with a nonzero argument for debugging
5045 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
5046 This function is similar to @code{__builtin_return_address}, but it
5047 returns the address of the function frame rather than the return address
5048 of the function. Calling @code{__builtin_frame_address} with a value of
5049 @code{0} yields the frame address of the current function, a value of
5050 @code{1} yields the frame address of the caller of the current function,
5053 The frame is the area on the stack which holds local variables and saved
5054 registers. The frame address is normally the address of the first word
5055 pushed on to the stack by the function. However, the exact definition
5056 depends upon the processor and the calling convention. If the processor
5057 has a dedicated frame pointer register, and the function has a frame,
5058 then @code{__builtin_frame_address} will return the value of the frame
5061 On some machines it may be impossible to determine the frame address of
5062 any function other than the current one; in such cases, or when the top
5063 of the stack has been reached, this function will return @code{0} if
5064 the first frame pointer is properly initialized by the startup code.
5066 This function should only be used with a nonzero argument for debugging
5070 @node Vector Extensions
5071 @section Using vector instructions through built-in functions
5073 On some targets, the instruction set contains SIMD vector instructions that
5074 operate on multiple values contained in one large register at the same time.
5075 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
5078 The first step in using these extensions is to provide the necessary data
5079 types. This should be done using an appropriate @code{typedef}:
5082 typedef int v4si __attribute__ ((vector_size (16)));
5085 The @code{int} type specifies the base type, while the attribute specifies
5086 the vector size for the variable, measured in bytes. For example, the
5087 declaration above causes the compiler to set the mode for the @code{v4si}
5088 type to be 16 bytes wide and divided into @code{int} sized units. For
5089 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
5090 corresponding mode of @code{foo} will be @acronym{V4SI}.
5092 The @code{vector_size} attribute is only applicable to integral and
5093 float scalars, although arrays, pointers, and function return values
5094 are allowed in conjunction with this construct.
5096 All the basic integer types can be used as base types, both as signed
5097 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
5098 @code{long long}. In addition, @code{float} and @code{double} can be
5099 used to build floating-point vector types.
5101 Specifying a combination that is not valid for the current architecture
5102 will cause GCC to synthesize the instructions using a narrower mode.
5103 For example, if you specify a variable of type @code{V4SI} and your
5104 architecture does not allow for this specific SIMD type, GCC will
5105 produce code that uses 4 @code{SIs}.
5107 The types defined in this manner can be used with a subset of normal C
5108 operations. Currently, GCC will allow using the following operators
5109 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~}@.
5111 The operations behave like C++ @code{valarrays}. Addition is defined as
5112 the addition of the corresponding elements of the operands. For
5113 example, in the code below, each of the 4 elements in @var{a} will be
5114 added to the corresponding 4 elements in @var{b} and the resulting
5115 vector will be stored in @var{c}.
5118 typedef int v4si __attribute__ ((vector_size (16)));
5125 Subtraction, multiplication, division, and the logical operations
5126 operate in a similar manner. Likewise, the result of using the unary
5127 minus or complement operators on a vector type is a vector whose
5128 elements are the negative or complemented values of the corresponding
5129 elements in the operand.
5131 You can declare variables and use them in function calls and returns, as
5132 well as in assignments and some casts. You can specify a vector type as
5133 a return type for a function. Vector types can also be used as function
5134 arguments. It is possible to cast from one vector type to another,
5135 provided they are of the same size (in fact, you can also cast vectors
5136 to and from other datatypes of the same size).
5138 You cannot operate between vectors of different lengths or different
5139 signedness without a cast.
5141 A port that supports hardware vector operations, usually provides a set
5142 of built-in functions that can be used to operate on vectors. For
5143 example, a function to add two vectors and multiply the result by a
5144 third could look like this:
5147 v4si f (v4si a, v4si b, v4si c)
5149 v4si tmp = __builtin_addv4si (a, b);
5150 return __builtin_mulv4si (tmp, c);
5157 @findex __builtin_offsetof
5159 GCC implements for both C and C++ a syntactic extension to implement
5160 the @code{offsetof} macro.
5164 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
5166 offsetof_member_designator:
5168 | offsetof_member_designator "." @code{identifier}
5169 | offsetof_member_designator "[" @code{expr} "]"
5172 This extension is sufficient such that
5175 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
5178 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
5179 may be dependent. In either case, @var{member} may consist of a single
5180 identifier, or a sequence of member accesses and array references.
5182 @node Atomic Builtins
5183 @section Built-in functions for atomic memory access
5185 The following builtins are intended to be compatible with those described
5186 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
5187 section 7.4. As such, they depart from the normal GCC practice of using
5188 the ``__builtin_'' prefix, and further that they are overloaded such that
5189 they work on multiple types.
5191 The definition given in the Intel documentation allows only for the use of
5192 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
5193 counterparts. GCC will allow any integral scalar or pointer type that is
5194 1, 2, 4 or 8 bytes in length.
5196 Not all operations are supported by all target processors. If a particular
5197 operation cannot be implemented on the target processor, a warning will be
5198 generated and a call an external function will be generated. The external
5199 function will carry the same name as the builtin, with an additional suffix
5200 @samp{_@var{n}} where @var{n} is the size of the data type.
5202 @c ??? Should we have a mechanism to suppress this warning? This is almost
5203 @c useful for implementing the operation under the control of an external
5206 In most cases, these builtins are considered a @dfn{full barrier}. That is,
5207 no memory operand will be moved across the operation, either forward or
5208 backward. Further, instructions will be issued as necessary to prevent the
5209 processor from speculating loads across the operation and from queuing stores
5210 after the operation.
5212 All of the routines are are described in the Intel documentation to take
5213 ``an optional list of variables protected by the memory barrier''. It's
5214 not clear what is meant by that; it could mean that @emph{only} the
5215 following variables are protected, or it could mean that these variables
5216 should in addition be protected. At present GCC ignores this list and
5217 protects all variables which are globally accessible. If in the future
5218 we make some use of this list, an empty list will continue to mean all
5219 globally accessible variables.
5222 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
5223 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
5224 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
5225 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
5226 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
5227 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
5228 @findex __sync_fetch_and_add
5229 @findex __sync_fetch_and_sub
5230 @findex __sync_fetch_and_or
5231 @findex __sync_fetch_and_and
5232 @findex __sync_fetch_and_xor
5233 @findex __sync_fetch_and_nand
5234 These builtins perform the operation suggested by the name, and
5235 returns the value that had previously been in memory. That is,
5238 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
5239 @{ tmp = *ptr; *ptr = ~tmp & value; return tmp; @} // nand
5242 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
5243 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
5244 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
5245 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
5246 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
5247 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
5248 @findex __sync_add_and_fetch
5249 @findex __sync_sub_and_fetch
5250 @findex __sync_or_and_fetch
5251 @findex __sync_and_and_fetch
5252 @findex __sync_xor_and_fetch
5253 @findex __sync_nand_and_fetch
5254 These builtins perform the operation suggested by the name, and
5255 return the new value. That is,
5258 @{ *ptr @var{op}= value; return *ptr; @}
5259 @{ *ptr = ~*ptr & value; return *ptr; @} // nand
5262 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
5263 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
5264 @findex __sync_bool_compare_and_swap
5265 @findex __sync_val_compare_and_swap
5266 These builtins perform an atomic compare and swap. That is, if the current
5267 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
5270 The ``bool'' version returns true if the comparison is successful and
5271 @var{newval} was written. The ``val'' version returns the contents
5272 of @code{*@var{ptr}} before the operation.
5274 @item __sync_synchronize (...)
5275 @findex __sync_synchronize
5276 This builtin issues a full memory barrier.
5278 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
5279 @findex __sync_lock_test_and_set
5280 This builtin, as described by Intel, is not a traditional test-and-set
5281 operation, but rather an atomic exchange operation. It writes @var{value}
5282 into @code{*@var{ptr}}, and returns the previous contents of
5285 Many targets have only minimal support for such locks, and do not support
5286 a full exchange operation. In this case, a target may support reduced
5287 functionality here by which the @emph{only} valid value to store is the
5288 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
5289 is implementation defined.
5291 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
5292 This means that references after the builtin cannot move to (or be
5293 speculated to) before the builtin, but previous memory stores may not
5294 be globally visible yet, and previous memory loads may not yet be
5297 @item void __sync_lock_release (@var{type} *ptr, ...)
5298 @findex __sync_lock_release
5299 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
5300 Normally this means writing the constant 0 to @code{*@var{ptr}}.
5302 This builtin is not a full barrier, but rather a @dfn{release barrier}.
5303 This means that all previous memory stores are globally visible, and all
5304 previous memory loads have been satisfied, but following memory reads
5305 are not prevented from being speculated to before the barrier.
5308 @node Object Size Checking
5309 @section Object Size Checking Builtins
5310 @findex __builtin_object_size
5311 @findex __builtin___memcpy_chk
5312 @findex __builtin___mempcpy_chk
5313 @findex __builtin___memmove_chk
5314 @findex __builtin___memset_chk
5315 @findex __builtin___strcpy_chk
5316 @findex __builtin___stpcpy_chk
5317 @findex __builtin___strncpy_chk
5318 @findex __builtin___strcat_chk
5319 @findex __builtin___strncat_chk
5320 @findex __builtin___sprintf_chk
5321 @findex __builtin___snprintf_chk
5322 @findex __builtin___vsprintf_chk
5323 @findex __builtin___vsnprintf_chk
5324 @findex __builtin___printf_chk
5325 @findex __builtin___vprintf_chk
5326 @findex __builtin___fprintf_chk
5327 @findex __builtin___vfprintf_chk
5329 GCC implements a limited buffer overflow protection mechanism
5330 that can prevent some buffer overflow attacks.
5332 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
5333 is a built-in construct that returns a constant number of bytes from
5334 @var{ptr} to the end of the object @var{ptr} pointer points to
5335 (if known at compile time). @code{__builtin_object_size} never evaluates
5336 its arguments for side-effects. If there are any side-effects in them, it
5337 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
5338 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
5339 point to and all of them are known at compile time, the returned number
5340 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
5341 0 and minimum if nonzero. If it is not possible to determine which objects
5342 @var{ptr} points to at compile time, @code{__builtin_object_size} should
5343 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
5344 for @var{type} 2 or 3.
5346 @var{type} is an integer constant from 0 to 3. If the least significant
5347 bit is clear, objects are whole variables, if it is set, a closest
5348 surrounding subobject is considered the object a pointer points to.
5349 The second bit determines if maximum or minimum of remaining bytes
5353 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
5354 char *p = &var.buf1[1], *q = &var.b;
5356 /* Here the object p points to is var. */
5357 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
5358 /* The subobject p points to is var.buf1. */
5359 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
5360 /* The object q points to is var. */
5361 assert (__builtin_object_size (q, 0)
5362 == (char *) (&var + 1) - (char *) &var.b);
5363 /* The subobject q points to is var.b. */
5364 assert (__builtin_object_size (q, 1) == sizeof (var.b));
5368 There are built-in functions added for many common string operation
5369 functions, e.g. for @code{memcpy} @code{__builtin___memcpy_chk}
5370 built-in is provided. This built-in has an additional last argument,
5371 which is the number of bytes remaining in object the @var{dest}
5372 argument points to or @code{(size_t) -1} if the size is not known.
5374 The built-in functions are optimized into the normal string functions
5375 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
5376 it is known at compile time that the destination object will not
5377 be overflown. If the compiler can determine at compile time the
5378 object will be always overflown, it issues a warning.
5380 The intended use can be e.g.
5384 #define bos0(dest) __builtin_object_size (dest, 0)
5385 #define memcpy(dest, src, n) \
5386 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
5390 /* It is unknown what object p points to, so this is optimized
5391 into plain memcpy - no checking is possible. */
5392 memcpy (p, "abcde", n);
5393 /* Destination is known and length too. It is known at compile
5394 time there will be no overflow. */
5395 memcpy (&buf[5], "abcde", 5);
5396 /* Destination is known, but the length is not known at compile time.
5397 This will result in __memcpy_chk call that can check for overflow
5399 memcpy (&buf[5], "abcde", n);
5400 /* Destination is known and it is known at compile time there will
5401 be overflow. There will be a warning and __memcpy_chk call that
5402 will abort the program at runtime. */
5403 memcpy (&buf[6], "abcde", 5);
5406 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
5407 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
5408 @code{strcat} and @code{strncat}.
5410 There are also checking built-in functions for formatted output functions.
5412 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
5413 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
5414 const char *fmt, ...);
5415 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
5417 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
5418 const char *fmt, va_list ap);
5421 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
5422 etc. functions and can contain implementation specific flags on what
5423 additional security measures the checking function might take, such as
5424 handling @code{%n} differently.
5426 The @var{os} argument is the object size @var{s} points to, like in the
5427 other built-in functions. There is a small difference in the behavior
5428 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
5429 optimized into the non-checking functions only if @var{flag} is 0, otherwise
5430 the checking function is called with @var{os} argument set to
5433 In addition to this, there are checking built-in functions
5434 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
5435 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
5436 These have just one additional argument, @var{flag}, right before
5437 format string @var{fmt}. If the compiler is able to optimize them to
5438 @code{fputc} etc. functions, it will, otherwise the checking function
5439 should be called and the @var{flag} argument passed to it.
5441 @node Other Builtins
5442 @section Other built-in functions provided by GCC
5443 @cindex built-in functions
5444 @findex __builtin_isfinite
5445 @findex __builtin_isnormal
5446 @findex __builtin_isgreater
5447 @findex __builtin_isgreaterequal
5448 @findex __builtin_isless
5449 @findex __builtin_islessequal
5450 @findex __builtin_islessgreater
5451 @findex __builtin_isunordered
5452 @findex __builtin_powi
5453 @findex __builtin_powif
5454 @findex __builtin_powil
5612 @findex fprintf_unlocked
5614 @findex fputs_unlocked
5731 @findex printf_unlocked
5763 @findex significandf
5764 @findex significandl
5835 GCC provides a large number of built-in functions other than the ones
5836 mentioned above. Some of these are for internal use in the processing
5837 of exceptions or variable-length argument lists and will not be
5838 documented here because they may change from time to time; we do not
5839 recommend general use of these functions.
5841 The remaining functions are provided for optimization purposes.
5843 @opindex fno-builtin
5844 GCC includes built-in versions of many of the functions in the standard
5845 C library. The versions prefixed with @code{__builtin_} will always be
5846 treated as having the same meaning as the C library function even if you
5847 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
5848 Many of these functions are only optimized in certain cases; if they are
5849 not optimized in a particular case, a call to the library function will
5854 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
5855 @option{-std=c99}), the functions
5856 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
5857 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
5858 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
5859 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
5860 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
5861 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
5862 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
5863 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
5864 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
5865 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
5866 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
5867 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
5868 @code{signbitd64}, @code{signbitd128}, @code{significandf},
5869 @code{significandl}, @code{significand}, @code{sincosf},
5870 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
5871 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
5872 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
5873 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
5875 may be handled as built-in functions.
5876 All these functions have corresponding versions
5877 prefixed with @code{__builtin_}, which may be used even in strict C89
5880 The ISO C99 functions
5881 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
5882 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
5883 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
5884 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
5885 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
5886 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
5887 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
5888 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
5889 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
5890 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
5891 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
5892 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
5893 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
5894 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
5895 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
5896 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
5897 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
5898 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
5899 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
5900 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
5901 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
5902 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
5903 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
5904 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
5905 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
5906 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
5907 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
5908 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
5909 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
5910 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
5911 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
5912 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
5913 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
5914 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
5915 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
5916 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
5917 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
5918 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
5919 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
5920 are handled as built-in functions
5921 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5923 There are also built-in versions of the ISO C99 functions
5924 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
5925 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
5926 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
5927 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
5928 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
5929 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
5930 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
5931 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
5932 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
5933 that are recognized in any mode since ISO C90 reserves these names for
5934 the purpose to which ISO C99 puts them. All these functions have
5935 corresponding versions prefixed with @code{__builtin_}.
5937 The ISO C94 functions
5938 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
5939 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
5940 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
5942 are handled as built-in functions
5943 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5945 The ISO C90 functions
5946 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
5947 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
5948 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
5949 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
5950 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
5951 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
5952 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
5953 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
5954 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
5955 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
5956 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
5957 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
5958 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
5959 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
5960 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
5961 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
5962 are all recognized as built-in functions unless
5963 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
5964 is specified for an individual function). All of these functions have
5965 corresponding versions prefixed with @code{__builtin_}.
5967 GCC provides built-in versions of the ISO C99 floating point comparison
5968 macros that avoid raising exceptions for unordered operands. They have
5969 the same names as the standard macros ( @code{isgreater},
5970 @code{isgreaterequal}, @code{isless}, @code{islessequal},
5971 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
5972 prefixed. We intend for a library implementor to be able to simply
5973 @code{#define} each standard macro to its built-in equivalent.
5974 In the same fashion, GCC provides @code{isfinite} and @code{isnormal}
5975 built-ins used with @code{__builtin_} prefixed.
5977 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
5979 You can use the built-in function @code{__builtin_types_compatible_p} to
5980 determine whether two types are the same.
5982 This built-in function returns 1 if the unqualified versions of the
5983 types @var{type1} and @var{type2} (which are types, not expressions) are
5984 compatible, 0 otherwise. The result of this built-in function can be
5985 used in integer constant expressions.
5987 This built-in function ignores top level qualifiers (e.g., @code{const},
5988 @code{volatile}). For example, @code{int} is equivalent to @code{const
5991 The type @code{int[]} and @code{int[5]} are compatible. On the other
5992 hand, @code{int} and @code{char *} are not compatible, even if the size
5993 of their types, on the particular architecture are the same. Also, the
5994 amount of pointer indirection is taken into account when determining
5995 similarity. Consequently, @code{short *} is not similar to
5996 @code{short **}. Furthermore, two types that are typedefed are
5997 considered compatible if their underlying types are compatible.
5999 An @code{enum} type is not considered to be compatible with another
6000 @code{enum} type even if both are compatible with the same integer
6001 type; this is what the C standard specifies.
6002 For example, @code{enum @{foo, bar@}} is not similar to
6003 @code{enum @{hot, dog@}}.
6005 You would typically use this function in code whose execution varies
6006 depending on the arguments' types. For example:
6011 typeof (x) tmp = (x); \
6012 if (__builtin_types_compatible_p (typeof (x), long double)) \
6013 tmp = foo_long_double (tmp); \
6014 else if (__builtin_types_compatible_p (typeof (x), double)) \
6015 tmp = foo_double (tmp); \
6016 else if (__builtin_types_compatible_p (typeof (x), float)) \
6017 tmp = foo_float (tmp); \
6024 @emph{Note:} This construct is only available for C@.
6028 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
6030 You can use the built-in function @code{__builtin_choose_expr} to
6031 evaluate code depending on the value of a constant expression. This
6032 built-in function returns @var{exp1} if @var{const_exp}, which is a
6033 constant expression that must be able to be determined at compile time,
6034 is nonzero. Otherwise it returns 0.
6036 This built-in function is analogous to the @samp{? :} operator in C,
6037 except that the expression returned has its type unaltered by promotion
6038 rules. Also, the built-in function does not evaluate the expression
6039 that was not chosen. For example, if @var{const_exp} evaluates to true,
6040 @var{exp2} is not evaluated even if it has side-effects.
6042 This built-in function can return an lvalue if the chosen argument is an
6045 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
6046 type. Similarly, if @var{exp2} is returned, its return type is the same
6053 __builtin_choose_expr ( \
6054 __builtin_types_compatible_p (typeof (x), double), \
6056 __builtin_choose_expr ( \
6057 __builtin_types_compatible_p (typeof (x), float), \
6059 /* @r{The void expression results in a compile-time error} \
6060 @r{when assigning the result to something.} */ \
6064 @emph{Note:} This construct is only available for C@. Furthermore, the
6065 unused expression (@var{exp1} or @var{exp2} depending on the value of
6066 @var{const_exp}) may still generate syntax errors. This may change in
6071 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
6072 You can use the built-in function @code{__builtin_constant_p} to
6073 determine if a value is known to be constant at compile-time and hence
6074 that GCC can perform constant-folding on expressions involving that
6075 value. The argument of the function is the value to test. The function
6076 returns the integer 1 if the argument is known to be a compile-time
6077 constant and 0 if it is not known to be a compile-time constant. A
6078 return of 0 does not indicate that the value is @emph{not} a constant,
6079 but merely that GCC cannot prove it is a constant with the specified
6080 value of the @option{-O} option.
6082 You would typically use this function in an embedded application where
6083 memory was a critical resource. If you have some complex calculation,
6084 you may want it to be folded if it involves constants, but need to call
6085 a function if it does not. For example:
6088 #define Scale_Value(X) \
6089 (__builtin_constant_p (X) \
6090 ? ((X) * SCALE + OFFSET) : Scale (X))
6093 You may use this built-in function in either a macro or an inline
6094 function. However, if you use it in an inlined function and pass an
6095 argument of the function as the argument to the built-in, GCC will
6096 never return 1 when you call the inline function with a string constant
6097 or compound literal (@pxref{Compound Literals}) and will not return 1
6098 when you pass a constant numeric value to the inline function unless you
6099 specify the @option{-O} option.
6101 You may also use @code{__builtin_constant_p} in initializers for static
6102 data. For instance, you can write
6105 static const int table[] = @{
6106 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
6112 This is an acceptable initializer even if @var{EXPRESSION} is not a
6113 constant expression. GCC must be more conservative about evaluating the
6114 built-in in this case, because it has no opportunity to perform
6117 Previous versions of GCC did not accept this built-in in data
6118 initializers. The earliest version where it is completely safe is
6122 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
6123 @opindex fprofile-arcs
6124 You may use @code{__builtin_expect} to provide the compiler with
6125 branch prediction information. In general, you should prefer to
6126 use actual profile feedback for this (@option{-fprofile-arcs}), as
6127 programmers are notoriously bad at predicting how their programs
6128 actually perform. However, there are applications in which this
6129 data is hard to collect.
6131 The return value is the value of @var{exp}, which should be an integral
6132 expression. The semantics of the built-in are that it is expected that
6133 @var{exp} == @var{c}. For example:
6136 if (__builtin_expect (x, 0))
6141 would indicate that we do not expect to call @code{foo}, since
6142 we expect @code{x} to be zero. Since you are limited to integral
6143 expressions for @var{exp}, you should use constructions such as
6146 if (__builtin_expect (ptr != NULL, 1))
6151 when testing pointer or floating-point values.
6154 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
6155 This function is used to flush the processor's instruction cache for
6156 the region of memory between @var{begin} inclusive and @var{end}
6157 exclusive. Some targets require that the instruction cache be
6158 flushed, after modifying memory containing code, in order to obtain
6159 deterministic behavior.
6161 If the target does not require instruction cache flushes,
6162 @code{__builtin___clear_cache} has no effect. Otherwise either
6163 instructions are emitted in-line to clear the instruction cache or a
6164 call to the @code{__clear_cache} function in libgcc is made.
6167 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
6168 This function is used to minimize cache-miss latency by moving data into
6169 a cache before it is accessed.
6170 You can insert calls to @code{__builtin_prefetch} into code for which
6171 you know addresses of data in memory that is likely to be accessed soon.
6172 If the target supports them, data prefetch instructions will be generated.
6173 If the prefetch is done early enough before the access then the data will
6174 be in the cache by the time it is accessed.
6176 The value of @var{addr} is the address of the memory to prefetch.
6177 There are two optional arguments, @var{rw} and @var{locality}.
6178 The value of @var{rw} is a compile-time constant one or zero; one
6179 means that the prefetch is preparing for a write to the memory address
6180 and zero, the default, means that the prefetch is preparing for a read.
6181 The value @var{locality} must be a compile-time constant integer between
6182 zero and three. A value of zero means that the data has no temporal
6183 locality, so it need not be left in the cache after the access. A value
6184 of three means that the data has a high degree of temporal locality and
6185 should be left in all levels of cache possible. Values of one and two
6186 mean, respectively, a low or moderate degree of temporal locality. The
6190 for (i = 0; i < n; i++)
6193 __builtin_prefetch (&a[i+j], 1, 1);
6194 __builtin_prefetch (&b[i+j], 0, 1);
6199 Data prefetch does not generate faults if @var{addr} is invalid, but
6200 the address expression itself must be valid. For example, a prefetch
6201 of @code{p->next} will not fault if @code{p->next} is not a valid
6202 address, but evaluation will fault if @code{p} is not a valid address.
6204 If the target does not support data prefetch, the address expression
6205 is evaluated if it includes side effects but no other code is generated
6206 and GCC does not issue a warning.
6209 @deftypefn {Built-in Function} double __builtin_huge_val (void)
6210 Returns a positive infinity, if supported by the floating-point format,
6211 else @code{DBL_MAX}. This function is suitable for implementing the
6212 ISO C macro @code{HUGE_VAL}.
6215 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
6216 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
6219 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
6220 Similar to @code{__builtin_huge_val}, except the return
6221 type is @code{long double}.
6224 @deftypefn {Built-in Function} double __builtin_inf (void)
6225 Similar to @code{__builtin_huge_val}, except a warning is generated
6226 if the target floating-point format does not support infinities.
6229 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
6230 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
6233 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
6234 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
6237 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
6238 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
6241 @deftypefn {Built-in Function} float __builtin_inff (void)
6242 Similar to @code{__builtin_inf}, except the return type is @code{float}.
6243 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
6246 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
6247 Similar to @code{__builtin_inf}, except the return
6248 type is @code{long double}.
6251 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
6252 This is an implementation of the ISO C99 function @code{nan}.
6254 Since ISO C99 defines this function in terms of @code{strtod}, which we
6255 do not implement, a description of the parsing is in order. The string
6256 is parsed as by @code{strtol}; that is, the base is recognized by
6257 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
6258 in the significand such that the least significant bit of the number
6259 is at the least significant bit of the significand. The number is
6260 truncated to fit the significand field provided. The significand is
6261 forced to be a quiet NaN@.
6263 This function, if given a string literal all of which would have been
6264 consumed by strtol, is evaluated early enough that it is considered a
6265 compile-time constant.
6268 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
6269 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
6272 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
6273 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
6276 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
6277 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
6280 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
6281 Similar to @code{__builtin_nan}, except the return type is @code{float}.
6284 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
6285 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
6288 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
6289 Similar to @code{__builtin_nan}, except the significand is forced
6290 to be a signaling NaN@. The @code{nans} function is proposed by
6291 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
6294 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
6295 Similar to @code{__builtin_nans}, except the return type is @code{float}.
6298 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
6299 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
6302 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
6303 Returns one plus the index of the least significant 1-bit of @var{x}, or
6304 if @var{x} is zero, returns zero.
6307 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
6308 Returns the number of leading 0-bits in @var{x}, starting at the most
6309 significant bit position. If @var{x} is 0, the result is undefined.
6312 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
6313 Returns the number of trailing 0-bits in @var{x}, starting at the least
6314 significant bit position. If @var{x} is 0, the result is undefined.
6317 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
6318 Returns the number of 1-bits in @var{x}.
6321 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
6322 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
6326 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
6327 Similar to @code{__builtin_ffs}, except the argument type is
6328 @code{unsigned long}.
6331 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
6332 Similar to @code{__builtin_clz}, except the argument type is
6333 @code{unsigned long}.
6336 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
6337 Similar to @code{__builtin_ctz}, except the argument type is
6338 @code{unsigned long}.
6341 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
6342 Similar to @code{__builtin_popcount}, except the argument type is
6343 @code{unsigned long}.
6346 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
6347 Similar to @code{__builtin_parity}, except the argument type is
6348 @code{unsigned long}.
6351 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
6352 Similar to @code{__builtin_ffs}, except the argument type is
6353 @code{unsigned long long}.
6356 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
6357 Similar to @code{__builtin_clz}, except the argument type is
6358 @code{unsigned long long}.
6361 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
6362 Similar to @code{__builtin_ctz}, except the argument type is
6363 @code{unsigned long long}.
6366 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
6367 Similar to @code{__builtin_popcount}, except the argument type is
6368 @code{unsigned long long}.
6371 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
6372 Similar to @code{__builtin_parity}, except the argument type is
6373 @code{unsigned long long}.
6376 @deftypefn {Built-in Function} double __builtin_powi (double, int)
6377 Returns the first argument raised to the power of the second. Unlike the
6378 @code{pow} function no guarantees about precision and rounding are made.
6381 @deftypefn {Built-in Function} float __builtin_powif (float, int)
6382 Similar to @code{__builtin_powi}, except the argument and return types
6386 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
6387 Similar to @code{__builtin_powi}, except the argument and return types
6388 are @code{long double}.
6391 @deftypefn {Built-in Function} int32_t __builtin_bswap32 (int32_t x)
6392 Returns @var{x} with the order of the bytes reversed; for example,
6393 @code{0xaabbccdd} becomes @code{0xddccbbaa}. Byte here always means
6397 @deftypefn {Built-in Function} int64_t __builtin_bswap64 (int64_t x)
6398 Similar to @code{__builtin_bswap32}, except the argument and return types
6402 @node Target Builtins
6403 @section Built-in Functions Specific to Particular Target Machines
6405 On some target machines, GCC supports many built-in functions specific
6406 to those machines. Generally these generate calls to specific machine
6407 instructions, but allow the compiler to schedule those calls.
6410 * Alpha Built-in Functions::
6411 * ARM iWMMXt Built-in Functions::
6412 * ARM NEON Intrinsics::
6413 * Blackfin Built-in Functions::
6414 * FR-V Built-in Functions::
6415 * X86 Built-in Functions::
6416 * MIPS DSP Built-in Functions::
6417 * MIPS Paired-Single Support::
6418 * PowerPC AltiVec Built-in Functions::
6419 * SPARC VIS Built-in Functions::
6420 * SPU Built-in Functions::
6423 @node Alpha Built-in Functions
6424 @subsection Alpha Built-in Functions
6426 These built-in functions are available for the Alpha family of
6427 processors, depending on the command-line switches used.
6429 The following built-in functions are always available. They
6430 all generate the machine instruction that is part of the name.
6433 long __builtin_alpha_implver (void)
6434 long __builtin_alpha_rpcc (void)
6435 long __builtin_alpha_amask (long)
6436 long __builtin_alpha_cmpbge (long, long)
6437 long __builtin_alpha_extbl (long, long)
6438 long __builtin_alpha_extwl (long, long)
6439 long __builtin_alpha_extll (long, long)
6440 long __builtin_alpha_extql (long, long)
6441 long __builtin_alpha_extwh (long, long)
6442 long __builtin_alpha_extlh (long, long)
6443 long __builtin_alpha_extqh (long, long)
6444 long __builtin_alpha_insbl (long, long)
6445 long __builtin_alpha_inswl (long, long)
6446 long __builtin_alpha_insll (long, long)
6447 long __builtin_alpha_insql (long, long)
6448 long __builtin_alpha_inswh (long, long)
6449 long __builtin_alpha_inslh (long, long)
6450 long __builtin_alpha_insqh (long, long)
6451 long __builtin_alpha_mskbl (long, long)
6452 long __builtin_alpha_mskwl (long, long)
6453 long __builtin_alpha_mskll (long, long)
6454 long __builtin_alpha_mskql (long, long)
6455 long __builtin_alpha_mskwh (long, long)
6456 long __builtin_alpha_msklh (long, long)
6457 long __builtin_alpha_mskqh (long, long)
6458 long __builtin_alpha_umulh (long, long)
6459 long __builtin_alpha_zap (long, long)
6460 long __builtin_alpha_zapnot (long, long)
6463 The following built-in functions are always with @option{-mmax}
6464 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
6465 later. They all generate the machine instruction that is part
6469 long __builtin_alpha_pklb (long)
6470 long __builtin_alpha_pkwb (long)
6471 long __builtin_alpha_unpkbl (long)
6472 long __builtin_alpha_unpkbw (long)
6473 long __builtin_alpha_minub8 (long, long)
6474 long __builtin_alpha_minsb8 (long, long)
6475 long __builtin_alpha_minuw4 (long, long)
6476 long __builtin_alpha_minsw4 (long, long)
6477 long __builtin_alpha_maxub8 (long, long)
6478 long __builtin_alpha_maxsb8 (long, long)
6479 long __builtin_alpha_maxuw4 (long, long)
6480 long __builtin_alpha_maxsw4 (long, long)
6481 long __builtin_alpha_perr (long, long)
6484 The following built-in functions are always with @option{-mcix}
6485 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
6486 later. They all generate the machine instruction that is part
6490 long __builtin_alpha_cttz (long)
6491 long __builtin_alpha_ctlz (long)
6492 long __builtin_alpha_ctpop (long)
6495 The following builtins are available on systems that use the OSF/1
6496 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
6497 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
6498 @code{rdval} and @code{wrval}.
6501 void *__builtin_thread_pointer (void)
6502 void __builtin_set_thread_pointer (void *)
6505 @node ARM iWMMXt Built-in Functions
6506 @subsection ARM iWMMXt Built-in Functions
6508 These built-in functions are available for the ARM family of
6509 processors when the @option{-mcpu=iwmmxt} switch is used:
6512 typedef int v2si __attribute__ ((vector_size (8)));
6513 typedef short v4hi __attribute__ ((vector_size (8)));
6514 typedef char v8qi __attribute__ ((vector_size (8)));
6516 int __builtin_arm_getwcx (int)
6517 void __builtin_arm_setwcx (int, int)
6518 int __builtin_arm_textrmsb (v8qi, int)
6519 int __builtin_arm_textrmsh (v4hi, int)
6520 int __builtin_arm_textrmsw (v2si, int)
6521 int __builtin_arm_textrmub (v8qi, int)
6522 int __builtin_arm_textrmuh (v4hi, int)
6523 int __builtin_arm_textrmuw (v2si, int)
6524 v8qi __builtin_arm_tinsrb (v8qi, int)
6525 v4hi __builtin_arm_tinsrh (v4hi, int)
6526 v2si __builtin_arm_tinsrw (v2si, int)
6527 long long __builtin_arm_tmia (long long, int, int)
6528 long long __builtin_arm_tmiabb (long long, int, int)
6529 long long __builtin_arm_tmiabt (long long, int, int)
6530 long long __builtin_arm_tmiaph (long long, int, int)
6531 long long __builtin_arm_tmiatb (long long, int, int)
6532 long long __builtin_arm_tmiatt (long long, int, int)
6533 int __builtin_arm_tmovmskb (v8qi)
6534 int __builtin_arm_tmovmskh (v4hi)
6535 int __builtin_arm_tmovmskw (v2si)
6536 long long __builtin_arm_waccb (v8qi)
6537 long long __builtin_arm_wacch (v4hi)
6538 long long __builtin_arm_waccw (v2si)
6539 v8qi __builtin_arm_waddb (v8qi, v8qi)
6540 v8qi __builtin_arm_waddbss (v8qi, v8qi)
6541 v8qi __builtin_arm_waddbus (v8qi, v8qi)
6542 v4hi __builtin_arm_waddh (v4hi, v4hi)
6543 v4hi __builtin_arm_waddhss (v4hi, v4hi)
6544 v4hi __builtin_arm_waddhus (v4hi, v4hi)
6545 v2si __builtin_arm_waddw (v2si, v2si)
6546 v2si __builtin_arm_waddwss (v2si, v2si)
6547 v2si __builtin_arm_waddwus (v2si, v2si)
6548 v8qi __builtin_arm_walign (v8qi, v8qi, int)
6549 long long __builtin_arm_wand(long long, long long)
6550 long long __builtin_arm_wandn (long long, long long)
6551 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
6552 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
6553 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
6554 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
6555 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
6556 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
6557 v2si __builtin_arm_wcmpeqw (v2si, v2si)
6558 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
6559 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
6560 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
6561 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
6562 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
6563 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
6564 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
6565 long long __builtin_arm_wmacsz (v4hi, v4hi)
6566 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
6567 long long __builtin_arm_wmacuz (v4hi, v4hi)
6568 v4hi __builtin_arm_wmadds (v4hi, v4hi)
6569 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
6570 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
6571 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
6572 v2si __builtin_arm_wmaxsw (v2si, v2si)
6573 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
6574 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
6575 v2si __builtin_arm_wmaxuw (v2si, v2si)
6576 v8qi __builtin_arm_wminsb (v8qi, v8qi)
6577 v4hi __builtin_arm_wminsh (v4hi, v4hi)
6578 v2si __builtin_arm_wminsw (v2si, v2si)
6579 v8qi __builtin_arm_wminub (v8qi, v8qi)
6580 v4hi __builtin_arm_wminuh (v4hi, v4hi)
6581 v2si __builtin_arm_wminuw (v2si, v2si)
6582 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
6583 v4hi __builtin_arm_wmulul (v4hi, v4hi)
6584 v4hi __builtin_arm_wmulum (v4hi, v4hi)
6585 long long __builtin_arm_wor (long long, long long)
6586 v2si __builtin_arm_wpackdss (long long, long long)
6587 v2si __builtin_arm_wpackdus (long long, long long)
6588 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
6589 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
6590 v4hi __builtin_arm_wpackwss (v2si, v2si)
6591 v4hi __builtin_arm_wpackwus (v2si, v2si)
6592 long long __builtin_arm_wrord (long long, long long)
6593 long long __builtin_arm_wrordi (long long, int)
6594 v4hi __builtin_arm_wrorh (v4hi, long long)
6595 v4hi __builtin_arm_wrorhi (v4hi, int)
6596 v2si __builtin_arm_wrorw (v2si, long long)
6597 v2si __builtin_arm_wrorwi (v2si, int)
6598 v2si __builtin_arm_wsadb (v8qi, v8qi)
6599 v2si __builtin_arm_wsadbz (v8qi, v8qi)
6600 v2si __builtin_arm_wsadh (v4hi, v4hi)
6601 v2si __builtin_arm_wsadhz (v4hi, v4hi)
6602 v4hi __builtin_arm_wshufh (v4hi, int)
6603 long long __builtin_arm_wslld (long long, long long)
6604 long long __builtin_arm_wslldi (long long, int)
6605 v4hi __builtin_arm_wsllh (v4hi, long long)
6606 v4hi __builtin_arm_wsllhi (v4hi, int)
6607 v2si __builtin_arm_wsllw (v2si, long long)
6608 v2si __builtin_arm_wsllwi (v2si, int)
6609 long long __builtin_arm_wsrad (long long, long long)
6610 long long __builtin_arm_wsradi (long long, int)
6611 v4hi __builtin_arm_wsrah (v4hi, long long)
6612 v4hi __builtin_arm_wsrahi (v4hi, int)
6613 v2si __builtin_arm_wsraw (v2si, long long)
6614 v2si __builtin_arm_wsrawi (v2si, int)
6615 long long __builtin_arm_wsrld (long long, long long)
6616 long long __builtin_arm_wsrldi (long long, int)
6617 v4hi __builtin_arm_wsrlh (v4hi, long long)
6618 v4hi __builtin_arm_wsrlhi (v4hi, int)
6619 v2si __builtin_arm_wsrlw (v2si, long long)
6620 v2si __builtin_arm_wsrlwi (v2si, int)
6621 v8qi __builtin_arm_wsubb (v8qi, v8qi)
6622 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
6623 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
6624 v4hi __builtin_arm_wsubh (v4hi, v4hi)
6625 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
6626 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
6627 v2si __builtin_arm_wsubw (v2si, v2si)
6628 v2si __builtin_arm_wsubwss (v2si, v2si)
6629 v2si __builtin_arm_wsubwus (v2si, v2si)
6630 v4hi __builtin_arm_wunpckehsb (v8qi)
6631 v2si __builtin_arm_wunpckehsh (v4hi)
6632 long long __builtin_arm_wunpckehsw (v2si)
6633 v4hi __builtin_arm_wunpckehub (v8qi)
6634 v2si __builtin_arm_wunpckehuh (v4hi)
6635 long long __builtin_arm_wunpckehuw (v2si)
6636 v4hi __builtin_arm_wunpckelsb (v8qi)
6637 v2si __builtin_arm_wunpckelsh (v4hi)
6638 long long __builtin_arm_wunpckelsw (v2si)
6639 v4hi __builtin_arm_wunpckelub (v8qi)
6640 v2si __builtin_arm_wunpckeluh (v4hi)
6641 long long __builtin_arm_wunpckeluw (v2si)
6642 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
6643 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
6644 v2si __builtin_arm_wunpckihw (v2si, v2si)
6645 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
6646 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
6647 v2si __builtin_arm_wunpckilw (v2si, v2si)
6648 long long __builtin_arm_wxor (long long, long long)
6649 long long __builtin_arm_wzero ()
6652 @node ARM NEON Intrinsics
6653 @subsection ARM NEON Intrinsics
6655 These built-in intrinsics for the ARM Advanced SIMD extension are available
6656 when the @option{-mfpu=neon} switch is used:
6658 @include arm-neon-intrinsics.texi
6660 @node Blackfin Built-in Functions
6661 @subsection Blackfin Built-in Functions
6663 Currently, there are two Blackfin-specific built-in functions. These are
6664 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
6665 using inline assembly; by using these built-in functions the compiler can
6666 automatically add workarounds for hardware errata involving these
6667 instructions. These functions are named as follows:
6670 void __builtin_bfin_csync (void)
6671 void __builtin_bfin_ssync (void)
6674 @node FR-V Built-in Functions
6675 @subsection FR-V Built-in Functions
6677 GCC provides many FR-V-specific built-in functions. In general,
6678 these functions are intended to be compatible with those described
6679 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
6680 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
6681 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
6682 pointer rather than by value.
6684 Most of the functions are named after specific FR-V instructions.
6685 Such functions are said to be ``directly mapped'' and are summarized
6686 here in tabular form.
6690 * Directly-mapped Integer Functions::
6691 * Directly-mapped Media Functions::
6692 * Raw read/write Functions::
6693 * Other Built-in Functions::
6696 @node Argument Types
6697 @subsubsection Argument Types
6699 The arguments to the built-in functions can be divided into three groups:
6700 register numbers, compile-time constants and run-time values. In order
6701 to make this classification clear at a glance, the arguments and return
6702 values are given the following pseudo types:
6704 @multitable @columnfractions .20 .30 .15 .35
6705 @item Pseudo type @tab Real C type @tab Constant? @tab Description
6706 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
6707 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
6708 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
6709 @item @code{uw2} @tab @code{unsigned long long} @tab No
6710 @tab an unsigned doubleword
6711 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
6712 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
6713 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
6714 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
6717 These pseudo types are not defined by GCC, they are simply a notational
6718 convenience used in this manual.
6720 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
6721 and @code{sw2} are evaluated at run time. They correspond to
6722 register operands in the underlying FR-V instructions.
6724 @code{const} arguments represent immediate operands in the underlying
6725 FR-V instructions. They must be compile-time constants.
6727 @code{acc} arguments are evaluated at compile time and specify the number
6728 of an accumulator register. For example, an @code{acc} argument of 2
6729 will select the ACC2 register.
6731 @code{iacc} arguments are similar to @code{acc} arguments but specify the
6732 number of an IACC register. See @pxref{Other Built-in Functions}
6735 @node Directly-mapped Integer Functions
6736 @subsubsection Directly-mapped Integer Functions
6738 The functions listed below map directly to FR-V I-type instructions.
6740 @multitable @columnfractions .45 .32 .23
6741 @item Function prototype @tab Example usage @tab Assembly output
6742 @item @code{sw1 __ADDSS (sw1, sw1)}
6743 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
6744 @tab @code{ADDSS @var{a},@var{b},@var{c}}
6745 @item @code{sw1 __SCAN (sw1, sw1)}
6746 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
6747 @tab @code{SCAN @var{a},@var{b},@var{c}}
6748 @item @code{sw1 __SCUTSS (sw1)}
6749 @tab @code{@var{b} = __SCUTSS (@var{a})}
6750 @tab @code{SCUTSS @var{a},@var{b}}
6751 @item @code{sw1 __SLASS (sw1, sw1)}
6752 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
6753 @tab @code{SLASS @var{a},@var{b},@var{c}}
6754 @item @code{void __SMASS (sw1, sw1)}
6755 @tab @code{__SMASS (@var{a}, @var{b})}
6756 @tab @code{SMASS @var{a},@var{b}}
6757 @item @code{void __SMSSS (sw1, sw1)}
6758 @tab @code{__SMSSS (@var{a}, @var{b})}
6759 @tab @code{SMSSS @var{a},@var{b}}
6760 @item @code{void __SMU (sw1, sw1)}
6761 @tab @code{__SMU (@var{a}, @var{b})}
6762 @tab @code{SMU @var{a},@var{b}}
6763 @item @code{sw2 __SMUL (sw1, sw1)}
6764 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
6765 @tab @code{SMUL @var{a},@var{b},@var{c}}
6766 @item @code{sw1 __SUBSS (sw1, sw1)}
6767 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
6768 @tab @code{SUBSS @var{a},@var{b},@var{c}}
6769 @item @code{uw2 __UMUL (uw1, uw1)}
6770 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
6771 @tab @code{UMUL @var{a},@var{b},@var{c}}
6774 @node Directly-mapped Media Functions
6775 @subsubsection Directly-mapped Media Functions
6777 The functions listed below map directly to FR-V M-type instructions.
6779 @multitable @columnfractions .45 .32 .23
6780 @item Function prototype @tab Example usage @tab Assembly output
6781 @item @code{uw1 __MABSHS (sw1)}
6782 @tab @code{@var{b} = __MABSHS (@var{a})}
6783 @tab @code{MABSHS @var{a},@var{b}}
6784 @item @code{void __MADDACCS (acc, acc)}
6785 @tab @code{__MADDACCS (@var{b}, @var{a})}
6786 @tab @code{MADDACCS @var{a},@var{b}}
6787 @item @code{sw1 __MADDHSS (sw1, sw1)}
6788 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
6789 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
6790 @item @code{uw1 __MADDHUS (uw1, uw1)}
6791 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
6792 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
6793 @item @code{uw1 __MAND (uw1, uw1)}
6794 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
6795 @tab @code{MAND @var{a},@var{b},@var{c}}
6796 @item @code{void __MASACCS (acc, acc)}
6797 @tab @code{__MASACCS (@var{b}, @var{a})}
6798 @tab @code{MASACCS @var{a},@var{b}}
6799 @item @code{uw1 __MAVEH (uw1, uw1)}
6800 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
6801 @tab @code{MAVEH @var{a},@var{b},@var{c}}
6802 @item @code{uw2 __MBTOH (uw1)}
6803 @tab @code{@var{b} = __MBTOH (@var{a})}
6804 @tab @code{MBTOH @var{a},@var{b}}
6805 @item @code{void __MBTOHE (uw1 *, uw1)}
6806 @tab @code{__MBTOHE (&@var{b}, @var{a})}
6807 @tab @code{MBTOHE @var{a},@var{b}}
6808 @item @code{void __MCLRACC (acc)}
6809 @tab @code{__MCLRACC (@var{a})}
6810 @tab @code{MCLRACC @var{a}}
6811 @item @code{void __MCLRACCA (void)}
6812 @tab @code{__MCLRACCA ()}
6813 @tab @code{MCLRACCA}
6814 @item @code{uw1 __Mcop1 (uw1, uw1)}
6815 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
6816 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
6817 @item @code{uw1 __Mcop2 (uw1, uw1)}
6818 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
6819 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
6820 @item @code{uw1 __MCPLHI (uw2, const)}
6821 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
6822 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
6823 @item @code{uw1 __MCPLI (uw2, const)}
6824 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
6825 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
6826 @item @code{void __MCPXIS (acc, sw1, sw1)}
6827 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
6828 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
6829 @item @code{void __MCPXIU (acc, uw1, uw1)}
6830 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
6831 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
6832 @item @code{void __MCPXRS (acc, sw1, sw1)}
6833 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
6834 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
6835 @item @code{void __MCPXRU (acc, uw1, uw1)}
6836 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
6837 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
6838 @item @code{uw1 __MCUT (acc, uw1)}
6839 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
6840 @tab @code{MCUT @var{a},@var{b},@var{c}}
6841 @item @code{uw1 __MCUTSS (acc, sw1)}
6842 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
6843 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
6844 @item @code{void __MDADDACCS (acc, acc)}
6845 @tab @code{__MDADDACCS (@var{b}, @var{a})}
6846 @tab @code{MDADDACCS @var{a},@var{b}}
6847 @item @code{void __MDASACCS (acc, acc)}
6848 @tab @code{__MDASACCS (@var{b}, @var{a})}
6849 @tab @code{MDASACCS @var{a},@var{b}}
6850 @item @code{uw2 __MDCUTSSI (acc, const)}
6851 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
6852 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
6853 @item @code{uw2 __MDPACKH (uw2, uw2)}
6854 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
6855 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
6856 @item @code{uw2 __MDROTLI (uw2, const)}
6857 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
6858 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
6859 @item @code{void __MDSUBACCS (acc, acc)}
6860 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
6861 @tab @code{MDSUBACCS @var{a},@var{b}}
6862 @item @code{void __MDUNPACKH (uw1 *, uw2)}
6863 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
6864 @tab @code{MDUNPACKH @var{a},@var{b}}
6865 @item @code{uw2 __MEXPDHD (uw1, const)}
6866 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
6867 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
6868 @item @code{uw1 __MEXPDHW (uw1, const)}
6869 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
6870 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
6871 @item @code{uw1 __MHDSETH (uw1, const)}
6872 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
6873 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
6874 @item @code{sw1 __MHDSETS (const)}
6875 @tab @code{@var{b} = __MHDSETS (@var{a})}
6876 @tab @code{MHDSETS #@var{a},@var{b}}
6877 @item @code{uw1 __MHSETHIH (uw1, const)}
6878 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
6879 @tab @code{MHSETHIH #@var{a},@var{b}}
6880 @item @code{sw1 __MHSETHIS (sw1, const)}
6881 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
6882 @tab @code{MHSETHIS #@var{a},@var{b}}
6883 @item @code{uw1 __MHSETLOH (uw1, const)}
6884 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
6885 @tab @code{MHSETLOH #@var{a},@var{b}}
6886 @item @code{sw1 __MHSETLOS (sw1, const)}
6887 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
6888 @tab @code{MHSETLOS #@var{a},@var{b}}
6889 @item @code{uw1 __MHTOB (uw2)}
6890 @tab @code{@var{b} = __MHTOB (@var{a})}
6891 @tab @code{MHTOB @var{a},@var{b}}
6892 @item @code{void __MMACHS (acc, sw1, sw1)}
6893 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
6894 @tab @code{MMACHS @var{a},@var{b},@var{c}}
6895 @item @code{void __MMACHU (acc, uw1, uw1)}
6896 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
6897 @tab @code{MMACHU @var{a},@var{b},@var{c}}
6898 @item @code{void __MMRDHS (acc, sw1, sw1)}
6899 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
6900 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
6901 @item @code{void __MMRDHU (acc, uw1, uw1)}
6902 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
6903 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
6904 @item @code{void __MMULHS (acc, sw1, sw1)}
6905 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
6906 @tab @code{MMULHS @var{a},@var{b},@var{c}}
6907 @item @code{void __MMULHU (acc, uw1, uw1)}
6908 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
6909 @tab @code{MMULHU @var{a},@var{b},@var{c}}
6910 @item @code{void __MMULXHS (acc, sw1, sw1)}
6911 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
6912 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
6913 @item @code{void __MMULXHU (acc, uw1, uw1)}
6914 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
6915 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
6916 @item @code{uw1 __MNOT (uw1)}
6917 @tab @code{@var{b} = __MNOT (@var{a})}
6918 @tab @code{MNOT @var{a},@var{b}}
6919 @item @code{uw1 __MOR (uw1, uw1)}
6920 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
6921 @tab @code{MOR @var{a},@var{b},@var{c}}
6922 @item @code{uw1 __MPACKH (uh, uh)}
6923 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
6924 @tab @code{MPACKH @var{a},@var{b},@var{c}}
6925 @item @code{sw2 __MQADDHSS (sw2, sw2)}
6926 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
6927 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
6928 @item @code{uw2 __MQADDHUS (uw2, uw2)}
6929 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
6930 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
6931 @item @code{void __MQCPXIS (acc, sw2, sw2)}
6932 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
6933 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
6934 @item @code{void __MQCPXIU (acc, uw2, uw2)}
6935 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
6936 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
6937 @item @code{void __MQCPXRS (acc, sw2, sw2)}
6938 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
6939 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
6940 @item @code{void __MQCPXRU (acc, uw2, uw2)}
6941 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
6942 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
6943 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
6944 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
6945 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
6946 @item @code{sw2 __MQLMTHS (sw2, sw2)}
6947 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
6948 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
6949 @item @code{void __MQMACHS (acc, sw2, sw2)}
6950 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
6951 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
6952 @item @code{void __MQMACHU (acc, uw2, uw2)}
6953 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
6954 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
6955 @item @code{void __MQMACXHS (acc, sw2, sw2)}
6956 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
6957 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
6958 @item @code{void __MQMULHS (acc, sw2, sw2)}
6959 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
6960 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
6961 @item @code{void __MQMULHU (acc, uw2, uw2)}
6962 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
6963 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
6964 @item @code{void __MQMULXHS (acc, sw2, sw2)}
6965 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
6966 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
6967 @item @code{void __MQMULXHU (acc, uw2, uw2)}
6968 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
6969 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
6970 @item @code{sw2 __MQSATHS (sw2, sw2)}
6971 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
6972 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
6973 @item @code{uw2 __MQSLLHI (uw2, int)}
6974 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
6975 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
6976 @item @code{sw2 __MQSRAHI (sw2, int)}
6977 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
6978 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
6979 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
6980 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
6981 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
6982 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
6983 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
6984 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
6985 @item @code{void __MQXMACHS (acc, sw2, sw2)}
6986 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
6987 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
6988 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
6989 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
6990 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
6991 @item @code{uw1 __MRDACC (acc)}
6992 @tab @code{@var{b} = __MRDACC (@var{a})}
6993 @tab @code{MRDACC @var{a},@var{b}}
6994 @item @code{uw1 __MRDACCG (acc)}
6995 @tab @code{@var{b} = __MRDACCG (@var{a})}
6996 @tab @code{MRDACCG @var{a},@var{b}}
6997 @item @code{uw1 __MROTLI (uw1, const)}
6998 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
6999 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
7000 @item @code{uw1 __MROTRI (uw1, const)}
7001 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
7002 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
7003 @item @code{sw1 __MSATHS (sw1, sw1)}
7004 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
7005 @tab @code{MSATHS @var{a},@var{b},@var{c}}
7006 @item @code{uw1 __MSATHU (uw1, uw1)}
7007 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
7008 @tab @code{MSATHU @var{a},@var{b},@var{c}}
7009 @item @code{uw1 __MSLLHI (uw1, const)}
7010 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
7011 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
7012 @item @code{sw1 __MSRAHI (sw1, const)}
7013 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
7014 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
7015 @item @code{uw1 __MSRLHI (uw1, const)}
7016 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
7017 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
7018 @item @code{void __MSUBACCS (acc, acc)}
7019 @tab @code{__MSUBACCS (@var{b}, @var{a})}
7020 @tab @code{MSUBACCS @var{a},@var{b}}
7021 @item @code{sw1 __MSUBHSS (sw1, sw1)}
7022 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
7023 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
7024 @item @code{uw1 __MSUBHUS (uw1, uw1)}
7025 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
7026 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
7027 @item @code{void __MTRAP (void)}
7028 @tab @code{__MTRAP ()}
7030 @item @code{uw2 __MUNPACKH (uw1)}
7031 @tab @code{@var{b} = __MUNPACKH (@var{a})}
7032 @tab @code{MUNPACKH @var{a},@var{b}}
7033 @item @code{uw1 __MWCUT (uw2, uw1)}
7034 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
7035 @tab @code{MWCUT @var{a},@var{b},@var{c}}
7036 @item @code{void __MWTACC (acc, uw1)}
7037 @tab @code{__MWTACC (@var{b}, @var{a})}
7038 @tab @code{MWTACC @var{a},@var{b}}
7039 @item @code{void __MWTACCG (acc, uw1)}
7040 @tab @code{__MWTACCG (@var{b}, @var{a})}
7041 @tab @code{MWTACCG @var{a},@var{b}}
7042 @item @code{uw1 __MXOR (uw1, uw1)}
7043 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
7044 @tab @code{MXOR @var{a},@var{b},@var{c}}
7047 @node Raw read/write Functions
7048 @subsubsection Raw read/write Functions
7050 This sections describes built-in functions related to read and write
7051 instructions to access memory. These functions generate
7052 @code{membar} instructions to flush the I/O load and stores where
7053 appropriate, as described in Fujitsu's manual described above.
7057 @item unsigned char __builtin_read8 (void *@var{data})
7058 @item unsigned short __builtin_read16 (void *@var{data})
7059 @item unsigned long __builtin_read32 (void *@var{data})
7060 @item unsigned long long __builtin_read64 (void *@var{data})
7062 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
7063 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
7064 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
7065 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
7068 @node Other Built-in Functions
7069 @subsubsection Other Built-in Functions
7071 This section describes built-in functions that are not named after
7072 a specific FR-V instruction.
7075 @item sw2 __IACCreadll (iacc @var{reg})
7076 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
7077 for future expansion and must be 0.
7079 @item sw1 __IACCreadl (iacc @var{reg})
7080 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
7081 Other values of @var{reg} are rejected as invalid.
7083 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
7084 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
7085 is reserved for future expansion and must be 0.
7087 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
7088 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
7089 is 1. Other values of @var{reg} are rejected as invalid.
7091 @item void __data_prefetch0 (const void *@var{x})
7092 Use the @code{dcpl} instruction to load the contents of address @var{x}
7093 into the data cache.
7095 @item void __data_prefetch (const void *@var{x})
7096 Use the @code{nldub} instruction to load the contents of address @var{x}
7097 into the data cache. The instruction will be issued in slot I1@.
7100 @node X86 Built-in Functions
7101 @subsection X86 Built-in Functions
7103 These built-in functions are available for the i386 and x86-64 family
7104 of computers, depending on the command-line switches used.
7106 Note that, if you specify command-line switches such as @option{-msse},
7107 the compiler could use the extended instruction sets even if the built-ins
7108 are not used explicitly in the program. For this reason, applications
7109 which perform runtime CPU detection must compile separate files for each
7110 supported architecture, using the appropriate flags. In particular,
7111 the file containing the CPU detection code should be compiled without
7114 The following machine modes are available for use with MMX built-in functions
7115 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
7116 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
7117 vector of eight 8-bit integers. Some of the built-in functions operate on
7118 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
7120 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
7121 of two 32-bit floating point values.
7123 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
7124 floating point values. Some instructions use a vector of four 32-bit
7125 integers, these use @code{V4SI}. Finally, some instructions operate on an
7126 entire vector register, interpreting it as a 128-bit integer, these use mode
7129 In the 64-bit mode, x86-64 family of processors uses additional built-in
7130 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
7131 floating point and @code{TC} 128-bit complex floating point values.
7133 The following floating point built-in functions are made available in the
7134 64-bit mode. All of them implement the function that is part of the name.
7137 __float128 __builtin_fabsq (__float128)
7138 __float128 __builtin_copysignq (__float128, __float128)
7141 The following floating point built-in functions are made available in the
7145 @item __float128 __builtin_infq (void)
7146 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
7149 The following built-in functions are made available by @option{-mmmx}.
7150 All of them generate the machine instruction that is part of the name.
7153 v8qi __builtin_ia32_paddb (v8qi, v8qi)
7154 v4hi __builtin_ia32_paddw (v4hi, v4hi)
7155 v2si __builtin_ia32_paddd (v2si, v2si)
7156 v8qi __builtin_ia32_psubb (v8qi, v8qi)
7157 v4hi __builtin_ia32_psubw (v4hi, v4hi)
7158 v2si __builtin_ia32_psubd (v2si, v2si)
7159 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
7160 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
7161 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
7162 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
7163 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
7164 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
7165 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
7166 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
7167 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
7168 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
7169 di __builtin_ia32_pand (di, di)
7170 di __builtin_ia32_pandn (di,di)
7171 di __builtin_ia32_por (di, di)
7172 di __builtin_ia32_pxor (di, di)
7173 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
7174 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
7175 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
7176 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
7177 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
7178 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
7179 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
7180 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
7181 v2si __builtin_ia32_punpckhdq (v2si, v2si)
7182 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
7183 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
7184 v2si __builtin_ia32_punpckldq (v2si, v2si)
7185 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
7186 v4hi __builtin_ia32_packssdw (v2si, v2si)
7187 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
7190 The following built-in functions are made available either with
7191 @option{-msse}, or with a combination of @option{-m3dnow} and
7192 @option{-march=athlon}. All of them generate the machine
7193 instruction that is part of the name.
7196 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
7197 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
7198 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
7199 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
7200 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
7201 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
7202 v8qi __builtin_ia32_pminub (v8qi, v8qi)
7203 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
7204 int __builtin_ia32_pextrw (v4hi, int)
7205 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
7206 int __builtin_ia32_pmovmskb (v8qi)
7207 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
7208 void __builtin_ia32_movntq (di *, di)
7209 void __builtin_ia32_sfence (void)
7212 The following built-in functions are available when @option{-msse} is used.
7213 All of them generate the machine instruction that is part of the name.
7216 int __builtin_ia32_comieq (v4sf, v4sf)
7217 int __builtin_ia32_comineq (v4sf, v4sf)
7218 int __builtin_ia32_comilt (v4sf, v4sf)
7219 int __builtin_ia32_comile (v4sf, v4sf)
7220 int __builtin_ia32_comigt (v4sf, v4sf)
7221 int __builtin_ia32_comige (v4sf, v4sf)
7222 int __builtin_ia32_ucomieq (v4sf, v4sf)
7223 int __builtin_ia32_ucomineq (v4sf, v4sf)
7224 int __builtin_ia32_ucomilt (v4sf, v4sf)
7225 int __builtin_ia32_ucomile (v4sf, v4sf)
7226 int __builtin_ia32_ucomigt (v4sf, v4sf)
7227 int __builtin_ia32_ucomige (v4sf, v4sf)
7228 v4sf __builtin_ia32_addps (v4sf, v4sf)
7229 v4sf __builtin_ia32_subps (v4sf, v4sf)
7230 v4sf __builtin_ia32_mulps (v4sf, v4sf)
7231 v4sf __builtin_ia32_divps (v4sf, v4sf)
7232 v4sf __builtin_ia32_addss (v4sf, v4sf)
7233 v4sf __builtin_ia32_subss (v4sf, v4sf)
7234 v4sf __builtin_ia32_mulss (v4sf, v4sf)
7235 v4sf __builtin_ia32_divss (v4sf, v4sf)
7236 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
7237 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
7238 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
7239 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
7240 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
7241 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
7242 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
7243 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
7244 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
7245 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
7246 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
7247 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
7248 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
7249 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
7250 v4si __builtin_ia32_cmpless (v4sf, v4sf)
7251 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
7252 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
7253 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
7254 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
7255 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
7256 v4sf __builtin_ia32_maxps (v4sf, v4sf)
7257 v4sf __builtin_ia32_maxss (v4sf, v4sf)
7258 v4sf __builtin_ia32_minps (v4sf, v4sf)
7259 v4sf __builtin_ia32_minss (v4sf, v4sf)
7260 v4sf __builtin_ia32_andps (v4sf, v4sf)
7261 v4sf __builtin_ia32_andnps (v4sf, v4sf)
7262 v4sf __builtin_ia32_orps (v4sf, v4sf)
7263 v4sf __builtin_ia32_xorps (v4sf, v4sf)
7264 v4sf __builtin_ia32_movss (v4sf, v4sf)
7265 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
7266 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
7267 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
7268 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
7269 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
7270 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
7271 v2si __builtin_ia32_cvtps2pi (v4sf)
7272 int __builtin_ia32_cvtss2si (v4sf)
7273 v2si __builtin_ia32_cvttps2pi (v4sf)
7274 int __builtin_ia32_cvttss2si (v4sf)
7275 v4sf __builtin_ia32_rcpps (v4sf)
7276 v4sf __builtin_ia32_rsqrtps (v4sf)
7277 v4sf __builtin_ia32_sqrtps (v4sf)
7278 v4sf __builtin_ia32_rcpss (v4sf)
7279 v4sf __builtin_ia32_rsqrtss (v4sf)
7280 v4sf __builtin_ia32_sqrtss (v4sf)
7281 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
7282 void __builtin_ia32_movntps (float *, v4sf)
7283 int __builtin_ia32_movmskps (v4sf)
7286 The following built-in functions are available when @option{-msse} is used.
7289 @item v4sf __builtin_ia32_loadaps (float *)
7290 Generates the @code{movaps} machine instruction as a load from memory.
7291 @item void __builtin_ia32_storeaps (float *, v4sf)
7292 Generates the @code{movaps} machine instruction as a store to memory.
7293 @item v4sf __builtin_ia32_loadups (float *)
7294 Generates the @code{movups} machine instruction as a load from memory.
7295 @item void __builtin_ia32_storeups (float *, v4sf)
7296 Generates the @code{movups} machine instruction as a store to memory.
7297 @item v4sf __builtin_ia32_loadsss (float *)
7298 Generates the @code{movss} machine instruction as a load from memory.
7299 @item void __builtin_ia32_storess (float *, v4sf)
7300 Generates the @code{movss} machine instruction as a store to memory.
7301 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
7302 Generates the @code{movhps} machine instruction as a load from memory.
7303 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
7304 Generates the @code{movlps} machine instruction as a load from memory
7305 @item void __builtin_ia32_storehps (v4sf, v2si *)
7306 Generates the @code{movhps} machine instruction as a store to memory.
7307 @item void __builtin_ia32_storelps (v4sf, v2si *)
7308 Generates the @code{movlps} machine instruction as a store to memory.
7311 The following built-in functions are available when @option{-msse2} is used.
7312 All of them generate the machine instruction that is part of the name.
7315 int __builtin_ia32_comisdeq (v2df, v2df)
7316 int __builtin_ia32_comisdlt (v2df, v2df)
7317 int __builtin_ia32_comisdle (v2df, v2df)
7318 int __builtin_ia32_comisdgt (v2df, v2df)
7319 int __builtin_ia32_comisdge (v2df, v2df)
7320 int __builtin_ia32_comisdneq (v2df, v2df)
7321 int __builtin_ia32_ucomisdeq (v2df, v2df)
7322 int __builtin_ia32_ucomisdlt (v2df, v2df)
7323 int __builtin_ia32_ucomisdle (v2df, v2df)
7324 int __builtin_ia32_ucomisdgt (v2df, v2df)
7325 int __builtin_ia32_ucomisdge (v2df, v2df)
7326 int __builtin_ia32_ucomisdneq (v2df, v2df)
7327 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
7328 v2df __builtin_ia32_cmpltpd (v2df, v2df)
7329 v2df __builtin_ia32_cmplepd (v2df, v2df)
7330 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
7331 v2df __builtin_ia32_cmpgepd (v2df, v2df)
7332 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
7333 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
7334 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
7335 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
7336 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
7337 v2df __builtin_ia32_cmpngepd (v2df, v2df)
7338 v2df __builtin_ia32_cmpordpd (v2df, v2df)
7339 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
7340 v2df __builtin_ia32_cmpltsd (v2df, v2df)
7341 v2df __builtin_ia32_cmplesd (v2df, v2df)
7342 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
7343 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
7344 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
7345 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
7346 v2df __builtin_ia32_cmpordsd (v2df, v2df)
7347 v2di __builtin_ia32_paddq (v2di, v2di)
7348 v2di __builtin_ia32_psubq (v2di, v2di)
7349 v2df __builtin_ia32_addpd (v2df, v2df)
7350 v2df __builtin_ia32_subpd (v2df, v2df)
7351 v2df __builtin_ia32_mulpd (v2df, v2df)
7352 v2df __builtin_ia32_divpd (v2df, v2df)
7353 v2df __builtin_ia32_addsd (v2df, v2df)
7354 v2df __builtin_ia32_subsd (v2df, v2df)
7355 v2df __builtin_ia32_mulsd (v2df, v2df)
7356 v2df __builtin_ia32_divsd (v2df, v2df)
7357 v2df __builtin_ia32_minpd (v2df, v2df)
7358 v2df __builtin_ia32_maxpd (v2df, v2df)
7359 v2df __builtin_ia32_minsd (v2df, v2df)
7360 v2df __builtin_ia32_maxsd (v2df, v2df)
7361 v2df __builtin_ia32_andpd (v2df, v2df)
7362 v2df __builtin_ia32_andnpd (v2df, v2df)
7363 v2df __builtin_ia32_orpd (v2df, v2df)
7364 v2df __builtin_ia32_xorpd (v2df, v2df)
7365 v2df __builtin_ia32_movsd (v2df, v2df)
7366 v2df __builtin_ia32_unpckhpd (v2df, v2df)
7367 v2df __builtin_ia32_unpcklpd (v2df, v2df)
7368 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
7369 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
7370 v4si __builtin_ia32_paddd128 (v4si, v4si)
7371 v2di __builtin_ia32_paddq128 (v2di, v2di)
7372 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
7373 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
7374 v4si __builtin_ia32_psubd128 (v4si, v4si)
7375 v2di __builtin_ia32_psubq128 (v2di, v2di)
7376 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
7377 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
7378 v2di __builtin_ia32_pand128 (v2di, v2di)
7379 v2di __builtin_ia32_pandn128 (v2di, v2di)
7380 v2di __builtin_ia32_por128 (v2di, v2di)
7381 v2di __builtin_ia32_pxor128 (v2di, v2di)
7382 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
7383 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
7384 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
7385 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
7386 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
7387 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
7388 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
7389 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
7390 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
7391 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
7392 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
7393 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
7394 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
7395 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
7396 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
7397 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
7398 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
7399 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
7400 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
7401 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
7402 v16qi __builtin_ia32_packsswb128 (v16qi, v16qi)
7403 v8hi __builtin_ia32_packssdw128 (v8hi, v8hi)
7404 v16qi __builtin_ia32_packuswb128 (v16qi, v16qi)
7405 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
7406 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
7407 v2df __builtin_ia32_loadupd (double *)
7408 void __builtin_ia32_storeupd (double *, v2df)
7409 v2df __builtin_ia32_loadhpd (v2df, double *)
7410 v2df __builtin_ia32_loadlpd (v2df, double *)
7411 int __builtin_ia32_movmskpd (v2df)
7412 int __builtin_ia32_pmovmskb128 (v16qi)
7413 void __builtin_ia32_movnti (int *, int)
7414 void __builtin_ia32_movntpd (double *, v2df)
7415 void __builtin_ia32_movntdq (v2df *, v2df)
7416 v4si __builtin_ia32_pshufd (v4si, int)
7417 v8hi __builtin_ia32_pshuflw (v8hi, int)
7418 v8hi __builtin_ia32_pshufhw (v8hi, int)
7419 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
7420 v2df __builtin_ia32_sqrtpd (v2df)
7421 v2df __builtin_ia32_sqrtsd (v2df)
7422 v2df __builtin_ia32_shufpd (v2df, v2df, int)
7423 v2df __builtin_ia32_cvtdq2pd (v4si)
7424 v4sf __builtin_ia32_cvtdq2ps (v4si)
7425 v4si __builtin_ia32_cvtpd2dq (v2df)
7426 v2si __builtin_ia32_cvtpd2pi (v2df)
7427 v4sf __builtin_ia32_cvtpd2ps (v2df)
7428 v4si __builtin_ia32_cvttpd2dq (v2df)
7429 v2si __builtin_ia32_cvttpd2pi (v2df)
7430 v2df __builtin_ia32_cvtpi2pd (v2si)
7431 int __builtin_ia32_cvtsd2si (v2df)
7432 int __builtin_ia32_cvttsd2si (v2df)
7433 long long __builtin_ia32_cvtsd2si64 (v2df)
7434 long long __builtin_ia32_cvttsd2si64 (v2df)
7435 v4si __builtin_ia32_cvtps2dq (v4sf)
7436 v2df __builtin_ia32_cvtps2pd (v4sf)
7437 v4si __builtin_ia32_cvttps2dq (v4sf)
7438 v2df __builtin_ia32_cvtsi2sd (v2df, int)
7439 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
7440 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
7441 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
7442 void __builtin_ia32_clflush (const void *)
7443 void __builtin_ia32_lfence (void)
7444 void __builtin_ia32_mfence (void)
7445 v16qi __builtin_ia32_loaddqu (const char *)
7446 void __builtin_ia32_storedqu (char *, v16qi)
7447 unsigned long long __builtin_ia32_pmuludq (v2si, v2si)
7448 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
7449 v8hi __builtin_ia32_psllw128 (v8hi, v2di)
7450 v4si __builtin_ia32_pslld128 (v4si, v2di)
7451 v2di __builtin_ia32_psllq128 (v4si, v2di)
7452 v8hi __builtin_ia32_psrlw128 (v8hi, v2di)
7453 v4si __builtin_ia32_psrld128 (v4si, v2di)
7454 v2di __builtin_ia32_psrlq128 (v2di, v2di)
7455 v8hi __builtin_ia32_psraw128 (v8hi, v2di)
7456 v4si __builtin_ia32_psrad128 (v4si, v2di)
7457 v2di __builtin_ia32_pslldqi128 (v2di, int)
7458 v8hi __builtin_ia32_psllwi128 (v8hi, int)
7459 v4si __builtin_ia32_pslldi128 (v4si, int)
7460 v2di __builtin_ia32_psllqi128 (v2di, int)
7461 v2di __builtin_ia32_psrldqi128 (v2di, int)
7462 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
7463 v4si __builtin_ia32_psrldi128 (v4si, int)
7464 v2di __builtin_ia32_psrlqi128 (v2di, int)
7465 v8hi __builtin_ia32_psrawi128 (v8hi, int)
7466 v4si __builtin_ia32_psradi128 (v4si, int)
7467 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
7470 The following built-in functions are available when @option{-msse3} is used.
7471 All of them generate the machine instruction that is part of the name.
7474 v2df __builtin_ia32_addsubpd (v2df, v2df)
7475 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
7476 v2df __builtin_ia32_haddpd (v2df, v2df)
7477 v4sf __builtin_ia32_haddps (v4sf, v4sf)
7478 v2df __builtin_ia32_hsubpd (v2df, v2df)
7479 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
7480 v16qi __builtin_ia32_lddqu (char const *)
7481 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
7482 v2df __builtin_ia32_movddup (v2df)
7483 v4sf __builtin_ia32_movshdup (v4sf)
7484 v4sf __builtin_ia32_movsldup (v4sf)
7485 void __builtin_ia32_mwait (unsigned int, unsigned int)
7488 The following built-in functions are available when @option{-msse3} is used.
7491 @item v2df __builtin_ia32_loadddup (double const *)
7492 Generates the @code{movddup} machine instruction as a load from memory.
7495 The following built-in functions are available when @option{-mssse3} is used.
7496 All of them generate the machine instruction that is part of the name
7500 v2si __builtin_ia32_phaddd (v2si, v2si)
7501 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
7502 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
7503 v2si __builtin_ia32_phsubd (v2si, v2si)
7504 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
7505 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
7506 v8qi __builtin_ia32_pmaddubsw (v8qi, v8qi)
7507 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
7508 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
7509 v8qi __builtin_ia32_psignb (v8qi, v8qi)
7510 v2si __builtin_ia32_psignd (v2si, v2si)
7511 v4hi __builtin_ia32_psignw (v4hi, v4hi)
7512 long long __builtin_ia32_palignr (long long, long long, int)
7513 v8qi __builtin_ia32_pabsb (v8qi)
7514 v2si __builtin_ia32_pabsd (v2si)
7515 v4hi __builtin_ia32_pabsw (v4hi)
7518 The following built-in functions are available when @option{-mssse3} is used.
7519 All of them generate the machine instruction that is part of the name
7523 v4si __builtin_ia32_phaddd128 (v4si, v4si)
7524 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
7525 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
7526 v4si __builtin_ia32_phsubd128 (v4si, v4si)
7527 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
7528 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
7529 v16qi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
7530 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
7531 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
7532 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
7533 v4si __builtin_ia32_psignd128 (v4si, v4si)
7534 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
7535 v2di __builtin_ia32_palignr (v2di, v2di, int)
7536 v16qi __builtin_ia32_pabsb128 (v16qi)
7537 v4si __builtin_ia32_pabsd128 (v4si)
7538 v8hi __builtin_ia32_pabsw128 (v8hi)
7541 The following built-in functions are available when @option{-msse4.1} is
7542 used. All of them generate the machine instruction that is part of the
7546 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
7547 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
7548 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
7549 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
7550 v2df __builtin_ia32_dppd (v2df, v2df, const int)
7551 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
7552 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
7553 v2di __builtin_ia32_movntdqa (v2di *);
7554 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
7555 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
7556 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
7557 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
7558 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
7559 v8hi __builtin_ia32_phminposuw128 (v8hi)
7560 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
7561 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
7562 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
7563 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
7564 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
7565 v4si __builtin_ia32_pminsd128 (v4si, v4si)
7566 v4si __builtin_ia32_pminud128 (v4si, v4si)
7567 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
7568 v4si __builtin_ia32_pmovsxbd128 (v16qi)
7569 v2di __builtin_ia32_pmovsxbq128 (v16qi)
7570 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
7571 v2di __builtin_ia32_pmovsxdq128 (v4si)
7572 v4si __builtin_ia32_pmovsxwd128 (v8hi)
7573 v2di __builtin_ia32_pmovsxwq128 (v8hi)
7574 v4si __builtin_ia32_pmovzxbd128 (v16qi)
7575 v2di __builtin_ia32_pmovzxbq128 (v16qi)
7576 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
7577 v2di __builtin_ia32_pmovzxdq128 (v4si)
7578 v4si __builtin_ia32_pmovzxwd128 (v8hi)
7579 v2di __builtin_ia32_pmovzxwq128 (v8hi)
7580 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
7581 v4si __builtin_ia32_pmulld128 (v4si, v4si)
7582 int __builtin_ia32_ptestc128 (v2di, v2di)
7583 int __builtin_ia32_ptestnzc128 (v2di, v2di)
7584 int __builtin_ia32_ptestz128 (v2di, v2di)
7585 v2df __builtin_ia32_roundpd (v2df, const int)
7586 v4sf __builtin_ia32_roundps (v4sf, const int)
7587 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
7588 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
7591 The following built-in functions are available when @option{-msse4.1} is
7595 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
7596 Generates the @code{insertps} machine instruction.
7597 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
7598 Generates the @code{pextrb} machine instruction.
7599 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
7600 Generates the @code{pinsrb} machine instruction.
7601 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
7602 Generates the @code{pinsrd} machine instruction.
7603 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
7604 Generates the @code{pinsrq} machine instruction in 64bit mode.
7607 The following built-in functions are changed to generate new SSE4.1
7608 instructions when @option{-msse4.1} is used.
7611 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
7612 Generates the @code{extractps} machine instruction.
7613 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
7614 Generates the @code{pextrd} machine instruction.
7615 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
7616 Generates the @code{pextrq} machine instruction in 64bit mode.
7619 The following built-in functions are available when @option{-msse4.2} is
7620 used. All of them generate the machine instruction that is part of the
7624 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
7625 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
7626 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
7627 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
7628 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
7629 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
7630 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
7631 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
7632 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
7633 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
7634 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
7635 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
7636 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
7637 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
7638 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
7641 The following built-in functions are available when @option{-msse4.2} is
7645 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
7646 Generates the @code{crc32b} machine instruction.
7647 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
7648 Generates the @code{crc32w} machine instruction.
7649 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
7650 Generates the @code{crc32l} machine instruction.
7651 @item unsigned long long __builtin_ia32_crc32di (unsigned int, unsigned long long)
7654 The following built-in functions are changed to generate new SSE4.2
7655 instructions when @option{-msse4.2} is used.
7658 @item int __builtin_popcount (unsigned int)
7659 Generates the @code{popcntl} machine instruction.
7660 @item int __builtin_popcountl (unsigned long)
7661 Generates the @code{popcntl} or @code{popcntq} machine instruction,
7662 depending on the size of @code{unsigned long}.
7663 @item int __builtin_popcountll (unsigned long long)
7664 Generates the @code{popcntq} machine instruction.
7667 The following built-in functions are available when @option{-msse4a} is used.
7668 All of them generate the machine instruction that is part of the name.
7671 void __builtin_ia32_movntsd (double *, v2df)
7672 void __builtin_ia32_movntss (float *, v4sf)
7673 v2di __builtin_ia32_extrq (v2di, v16qi)
7674 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
7675 v2di __builtin_ia32_insertq (v2di, v2di)
7676 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
7679 The following built-in functions are available when @option{-m3dnow} is used.
7680 All of them generate the machine instruction that is part of the name.
7683 void __builtin_ia32_femms (void)
7684 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
7685 v2si __builtin_ia32_pf2id (v2sf)
7686 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
7687 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
7688 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
7689 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
7690 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
7691 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
7692 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
7693 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
7694 v2sf __builtin_ia32_pfrcp (v2sf)
7695 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
7696 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
7697 v2sf __builtin_ia32_pfrsqrt (v2sf)
7698 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
7699 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
7700 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
7701 v2sf __builtin_ia32_pi2fd (v2si)
7702 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
7705 The following built-in functions are available when both @option{-m3dnow}
7706 and @option{-march=athlon} are used. All of them generate the machine
7707 instruction that is part of the name.
7710 v2si __builtin_ia32_pf2iw (v2sf)
7711 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
7712 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
7713 v2sf __builtin_ia32_pi2fw (v2si)
7714 v2sf __builtin_ia32_pswapdsf (v2sf)
7715 v2si __builtin_ia32_pswapdsi (v2si)
7718 @node MIPS DSP Built-in Functions
7719 @subsection MIPS DSP Built-in Functions
7721 The MIPS DSP Application-Specific Extension (ASE) includes new
7722 instructions that are designed to improve the performance of DSP and
7723 media applications. It provides instructions that operate on packed
7724 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
7726 GCC supports MIPS DSP operations using both the generic
7727 vector extensions (@pxref{Vector Extensions}) and a collection of
7728 MIPS-specific built-in functions. Both kinds of support are
7729 enabled by the @option{-mdsp} command-line option.
7731 Revision 2 of the ASE was introduced in the second half of 2006.
7732 This revision adds extra instructions to the original ASE, but is
7733 otherwise backwards-compatible with it. You can select revision 2
7734 using the command-line option @option{-mdspr2}; this option implies
7737 At present, GCC only provides support for operations on 32-bit
7738 vectors. The vector type associated with 8-bit integer data is
7739 usually called @code{v4i8}, the vector type associated with Q7
7740 is usually called @code{v4q7}, the vector type associated with 16-bit
7741 integer data is usually called @code{v2i16}, and the vector type
7742 associated with Q15 is usually called @code{v2q15}. They can be
7743 defined in C as follows:
7746 typedef signed char v4i8 __attribute__ ((vector_size(4)));
7747 typedef signed char v4q7 __attribute__ ((vector_size(4)));
7748 typedef short v2i16 __attribute__ ((vector_size(4)));
7749 typedef short v2q15 __attribute__ ((vector_size(4)));
7752 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
7753 initialized in the same way as aggregates. For example:
7756 v4i8 a = @{1, 2, 3, 4@};
7758 b = (v4i8) @{5, 6, 7, 8@};
7760 v2q15 c = @{0x0fcb, 0x3a75@};
7762 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
7765 @emph{Note:} The CPU's endianness determines the order in which values
7766 are packed. On little-endian targets, the first value is the least
7767 significant and the last value is the most significant. The opposite
7768 order applies to big-endian targets. For example, the code above will
7769 set the lowest byte of @code{a} to @code{1} on little-endian targets
7770 and @code{4} on big-endian targets.
7772 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
7773 representation. As shown in this example, the integer representation
7774 of a Q7 value can be obtained by multiplying the fractional value by
7775 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
7776 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
7779 The table below lists the @code{v4i8} and @code{v2q15} operations for which
7780 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
7781 and @code{c} and @code{d} are @code{v2q15} values.
7783 @multitable @columnfractions .50 .50
7784 @item C code @tab MIPS instruction
7785 @item @code{a + b} @tab @code{addu.qb}
7786 @item @code{c + d} @tab @code{addq.ph}
7787 @item @code{a - b} @tab @code{subu.qb}
7788 @item @code{c - d} @tab @code{subq.ph}
7791 The table below lists the @code{v2i16} operation for which
7792 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
7793 @code{v2i16} values.
7795 @multitable @columnfractions .50 .50
7796 @item C code @tab MIPS instruction
7797 @item @code{e * f} @tab @code{mul.ph}
7800 It is easier to describe the DSP built-in functions if we first define
7801 the following types:
7806 typedef unsigned int ui32;
7807 typedef long long a64;
7810 @code{q31} and @code{i32} are actually the same as @code{int}, but we
7811 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
7812 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
7813 @code{long long}, but we use @code{a64} to indicate values that will
7814 be placed in one of the four DSP accumulators (@code{$ac0},
7815 @code{$ac1}, @code{$ac2} or @code{$ac3}).
7817 Also, some built-in functions prefer or require immediate numbers as
7818 parameters, because the corresponding DSP instructions accept both immediate
7819 numbers and register operands, or accept immediate numbers only. The
7820 immediate parameters are listed as follows.
7829 imm_n32_31: -32 to 31.
7830 imm_n512_511: -512 to 511.
7833 The following built-in functions map directly to a particular MIPS DSP
7834 instruction. Please refer to the architecture specification
7835 for details on what each instruction does.
7838 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
7839 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
7840 q31 __builtin_mips_addq_s_w (q31, q31)
7841 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
7842 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
7843 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
7844 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
7845 q31 __builtin_mips_subq_s_w (q31, q31)
7846 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
7847 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
7848 i32 __builtin_mips_addsc (i32, i32)
7849 i32 __builtin_mips_addwc (i32, i32)
7850 i32 __builtin_mips_modsub (i32, i32)
7851 i32 __builtin_mips_raddu_w_qb (v4i8)
7852 v2q15 __builtin_mips_absq_s_ph (v2q15)
7853 q31 __builtin_mips_absq_s_w (q31)
7854 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
7855 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
7856 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
7857 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
7858 q31 __builtin_mips_preceq_w_phl (v2q15)
7859 q31 __builtin_mips_preceq_w_phr (v2q15)
7860 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
7861 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
7862 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
7863 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
7864 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
7865 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
7866 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
7867 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
7868 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
7869 v4i8 __builtin_mips_shll_qb (v4i8, i32)
7870 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
7871 v2q15 __builtin_mips_shll_ph (v2q15, i32)
7872 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
7873 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
7874 q31 __builtin_mips_shll_s_w (q31, imm0_31)
7875 q31 __builtin_mips_shll_s_w (q31, i32)
7876 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
7877 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
7878 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
7879 v2q15 __builtin_mips_shra_ph (v2q15, i32)
7880 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
7881 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
7882 q31 __builtin_mips_shra_r_w (q31, imm0_31)
7883 q31 __builtin_mips_shra_r_w (q31, i32)
7884 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
7885 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
7886 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
7887 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
7888 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
7889 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
7890 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
7891 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
7892 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
7893 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
7894 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
7895 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
7896 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
7897 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
7898 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
7899 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
7900 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
7901 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
7902 i32 __builtin_mips_bitrev (i32)
7903 i32 __builtin_mips_insv (i32, i32)
7904 v4i8 __builtin_mips_repl_qb (imm0_255)
7905 v4i8 __builtin_mips_repl_qb (i32)
7906 v2q15 __builtin_mips_repl_ph (imm_n512_511)
7907 v2q15 __builtin_mips_repl_ph (i32)
7908 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
7909 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
7910 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
7911 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
7912 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
7913 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
7914 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
7915 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
7916 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
7917 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
7918 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
7919 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
7920 i32 __builtin_mips_extr_w (a64, imm0_31)
7921 i32 __builtin_mips_extr_w (a64, i32)
7922 i32 __builtin_mips_extr_r_w (a64, imm0_31)
7923 i32 __builtin_mips_extr_s_h (a64, i32)
7924 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
7925 i32 __builtin_mips_extr_rs_w (a64, i32)
7926 i32 __builtin_mips_extr_s_h (a64, imm0_31)
7927 i32 __builtin_mips_extr_r_w (a64, i32)
7928 i32 __builtin_mips_extp (a64, imm0_31)
7929 i32 __builtin_mips_extp (a64, i32)
7930 i32 __builtin_mips_extpdp (a64, imm0_31)
7931 i32 __builtin_mips_extpdp (a64, i32)
7932 a64 __builtin_mips_shilo (a64, imm_n32_31)
7933 a64 __builtin_mips_shilo (a64, i32)
7934 a64 __builtin_mips_mthlip (a64, i32)
7935 void __builtin_mips_wrdsp (i32, imm0_63)
7936 i32 __builtin_mips_rddsp (imm0_63)
7937 i32 __builtin_mips_lbux (void *, i32)
7938 i32 __builtin_mips_lhx (void *, i32)
7939 i32 __builtin_mips_lwx (void *, i32)
7940 i32 __builtin_mips_bposge32 (void)
7943 The following built-in functions map directly to a particular MIPS DSP REV 2
7944 instruction. Please refer to the architecture specification
7945 for details on what each instruction does.
7948 v4q7 __builtin_mips_absq_s_qb (v4q7);
7949 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
7950 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
7951 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
7952 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
7953 i32 __builtin_mips_append (i32, i32, imm0_31);
7954 i32 __builtin_mips_balign (i32, i32, imm0_3);
7955 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
7956 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
7957 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
7958 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
7959 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
7960 a64 __builtin_mips_madd (a64, i32, i32);
7961 a64 __builtin_mips_maddu (a64, ui32, ui32);
7962 a64 __builtin_mips_msub (a64, i32, i32);
7963 a64 __builtin_mips_msubu (a64, ui32, ui32);
7964 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
7965 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
7966 q31 __builtin_mips_mulq_rs_w (q31, q31);
7967 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
7968 q31 __builtin_mips_mulq_s_w (q31, q31);
7969 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
7970 a64 __builtin_mips_mult (i32, i32);
7971 a64 __builtin_mips_multu (ui32, ui32);
7972 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
7973 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
7974 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
7975 i32 __builtin_mips_prepend (i32, i32, imm0_31);
7976 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
7977 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
7978 v4i8 __builtin_mips_shra_qb (v4i8, i32);
7979 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
7980 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
7981 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
7982 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
7983 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
7984 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
7985 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
7986 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
7987 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
7988 q31 __builtin_mips_addqh_w (q31, q31);
7989 q31 __builtin_mips_addqh_r_w (q31, q31);
7990 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
7991 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
7992 q31 __builtin_mips_subqh_w (q31, q31);
7993 q31 __builtin_mips_subqh_r_w (q31, q31);
7994 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
7995 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
7996 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
7997 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
7998 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
7999 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
8003 @node MIPS Paired-Single Support
8004 @subsection MIPS Paired-Single Support
8006 The MIPS64 architecture includes a number of instructions that
8007 operate on pairs of single-precision floating-point values.
8008 Each pair is packed into a 64-bit floating-point register,
8009 with one element being designated the ``upper half'' and
8010 the other being designated the ``lower half''.
8012 GCC supports paired-single operations using both the generic
8013 vector extensions (@pxref{Vector Extensions}) and a collection of
8014 MIPS-specific built-in functions. Both kinds of support are
8015 enabled by the @option{-mpaired-single} command-line option.
8017 The vector type associated with paired-single values is usually
8018 called @code{v2sf}. It can be defined in C as follows:
8021 typedef float v2sf __attribute__ ((vector_size (8)));
8024 @code{v2sf} values are initialized in the same way as aggregates.
8028 v2sf a = @{1.5, 9.1@};
8031 b = (v2sf) @{e, f@};
8034 @emph{Note:} The CPU's endianness determines which value is stored in
8035 the upper half of a register and which value is stored in the lower half.
8036 On little-endian targets, the first value is the lower one and the second
8037 value is the upper one. The opposite order applies to big-endian targets.
8038 For example, the code above will set the lower half of @code{a} to
8039 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
8042 * Paired-Single Arithmetic::
8043 * Paired-Single Built-in Functions::
8044 * MIPS-3D Built-in Functions::
8047 @node Paired-Single Arithmetic
8048 @subsubsection Paired-Single Arithmetic
8050 The table below lists the @code{v2sf} operations for which hardware
8051 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
8052 values and @code{x} is an integral value.
8054 @multitable @columnfractions .50 .50
8055 @item C code @tab MIPS instruction
8056 @item @code{a + b} @tab @code{add.ps}
8057 @item @code{a - b} @tab @code{sub.ps}
8058 @item @code{-a} @tab @code{neg.ps}
8059 @item @code{a * b} @tab @code{mul.ps}
8060 @item @code{a * b + c} @tab @code{madd.ps}
8061 @item @code{a * b - c} @tab @code{msub.ps}
8062 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
8063 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
8064 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
8067 Note that the multiply-accumulate instructions can be disabled
8068 using the command-line option @code{-mno-fused-madd}.
8070 @node Paired-Single Built-in Functions
8071 @subsubsection Paired-Single Built-in Functions
8073 The following paired-single functions map directly to a particular
8074 MIPS instruction. Please refer to the architecture specification
8075 for details on what each instruction does.
8078 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
8079 Pair lower lower (@code{pll.ps}).
8081 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
8082 Pair upper lower (@code{pul.ps}).
8084 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
8085 Pair lower upper (@code{plu.ps}).
8087 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
8088 Pair upper upper (@code{puu.ps}).
8090 @item v2sf __builtin_mips_cvt_ps_s (float, float)
8091 Convert pair to paired single (@code{cvt.ps.s}).
8093 @item float __builtin_mips_cvt_s_pl (v2sf)
8094 Convert pair lower to single (@code{cvt.s.pl}).
8096 @item float __builtin_mips_cvt_s_pu (v2sf)
8097 Convert pair upper to single (@code{cvt.s.pu}).
8099 @item v2sf __builtin_mips_abs_ps (v2sf)
8100 Absolute value (@code{abs.ps}).
8102 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
8103 Align variable (@code{alnv.ps}).
8105 @emph{Note:} The value of the third parameter must be 0 or 4
8106 modulo 8, otherwise the result will be unpredictable. Please read the
8107 instruction description for details.
8110 The following multi-instruction functions are also available.
8111 In each case, @var{cond} can be any of the 16 floating-point conditions:
8112 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
8113 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
8114 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
8117 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8118 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8119 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
8120 @code{movt.ps}/@code{movf.ps}).
8122 The @code{movt} functions return the value @var{x} computed by:
8125 c.@var{cond}.ps @var{cc},@var{a},@var{b}
8126 mov.ps @var{x},@var{c}
8127 movt.ps @var{x},@var{d},@var{cc}
8130 The @code{movf} functions are similar but use @code{movf.ps} instead
8133 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8134 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8135 Comparison of two paired-single values (@code{c.@var{cond}.ps},
8136 @code{bc1t}/@code{bc1f}).
8138 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
8139 and return either the upper or lower half of the result. For example:
8143 if (__builtin_mips_upper_c_eq_ps (a, b))
8144 upper_halves_are_equal ();
8146 upper_halves_are_unequal ();
8148 if (__builtin_mips_lower_c_eq_ps (a, b))
8149 lower_halves_are_equal ();
8151 lower_halves_are_unequal ();
8155 @node MIPS-3D Built-in Functions
8156 @subsubsection MIPS-3D Built-in Functions
8158 The MIPS-3D Application-Specific Extension (ASE) includes additional
8159 paired-single instructions that are designed to improve the performance
8160 of 3D graphics operations. Support for these instructions is controlled
8161 by the @option{-mips3d} command-line option.
8163 The functions listed below map directly to a particular MIPS-3D
8164 instruction. Please refer to the architecture specification for
8165 more details on what each instruction does.
8168 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
8169 Reduction add (@code{addr.ps}).
8171 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
8172 Reduction multiply (@code{mulr.ps}).
8174 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
8175 Convert paired single to paired word (@code{cvt.pw.ps}).
8177 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
8178 Convert paired word to paired single (@code{cvt.ps.pw}).
8180 @item float __builtin_mips_recip1_s (float)
8181 @itemx double __builtin_mips_recip1_d (double)
8182 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
8183 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
8185 @item float __builtin_mips_recip2_s (float, float)
8186 @itemx double __builtin_mips_recip2_d (double, double)
8187 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
8188 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
8190 @item float __builtin_mips_rsqrt1_s (float)
8191 @itemx double __builtin_mips_rsqrt1_d (double)
8192 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
8193 Reduced precision reciprocal square root (sequence step 1)
8194 (@code{rsqrt1.@var{fmt}}).
8196 @item float __builtin_mips_rsqrt2_s (float, float)
8197 @itemx double __builtin_mips_rsqrt2_d (double, double)
8198 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
8199 Reduced precision reciprocal square root (sequence step 2)
8200 (@code{rsqrt2.@var{fmt}}).
8203 The following multi-instruction functions are also available.
8204 In each case, @var{cond} can be any of the 16 floating-point conditions:
8205 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
8206 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
8207 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
8210 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
8211 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
8212 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
8213 @code{bc1t}/@code{bc1f}).
8215 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
8216 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
8221 if (__builtin_mips_cabs_eq_s (a, b))
8227 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8228 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8229 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
8230 @code{bc1t}/@code{bc1f}).
8232 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
8233 and return either the upper or lower half of the result. For example:
8237 if (__builtin_mips_upper_cabs_eq_ps (a, b))
8238 upper_halves_are_equal ();
8240 upper_halves_are_unequal ();
8242 if (__builtin_mips_lower_cabs_eq_ps (a, b))
8243 lower_halves_are_equal ();
8245 lower_halves_are_unequal ();
8248 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8249 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8250 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
8251 @code{movt.ps}/@code{movf.ps}).
8253 The @code{movt} functions return the value @var{x} computed by:
8256 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
8257 mov.ps @var{x},@var{c}
8258 movt.ps @var{x},@var{d},@var{cc}
8261 The @code{movf} functions are similar but use @code{movf.ps} instead
8264 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8265 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8266 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8267 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8268 Comparison of two paired-single values
8269 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
8270 @code{bc1any2t}/@code{bc1any2f}).
8272 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
8273 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
8274 result is true and the @code{all} forms return true if both results are true.
8279 if (__builtin_mips_any_c_eq_ps (a, b))
8284 if (__builtin_mips_all_c_eq_ps (a, b))
8290 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8291 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8292 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8293 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8294 Comparison of four paired-single values
8295 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
8296 @code{bc1any4t}/@code{bc1any4f}).
8298 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
8299 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
8300 The @code{any} forms return true if any of the four results are true
8301 and the @code{all} forms return true if all four results are true.
8306 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
8311 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
8318 @node PowerPC AltiVec Built-in Functions
8319 @subsection PowerPC AltiVec Built-in Functions
8321 GCC provides an interface for the PowerPC family of processors to access
8322 the AltiVec operations described in Motorola's AltiVec Programming
8323 Interface Manual. The interface is made available by including
8324 @code{<altivec.h>} and using @option{-maltivec} and
8325 @option{-mabi=altivec}. The interface supports the following vector
8329 vector unsigned char
8333 vector unsigned short
8344 GCC's implementation of the high-level language interface available from
8345 C and C++ code differs from Motorola's documentation in several ways.
8350 A vector constant is a list of constant expressions within curly braces.
8353 A vector initializer requires no cast if the vector constant is of the
8354 same type as the variable it is initializing.
8357 If @code{signed} or @code{unsigned} is omitted, the signedness of the
8358 vector type is the default signedness of the base type. The default
8359 varies depending on the operating system, so a portable program should
8360 always specify the signedness.
8363 Compiling with @option{-maltivec} adds keywords @code{__vector},
8364 @code{__pixel}, and @code{__bool}. Macros @option{vector},
8365 @code{pixel}, and @code{bool} are defined in @code{<altivec.h>} and can
8369 GCC allows using a @code{typedef} name as the type specifier for a
8373 For C, overloaded functions are implemented with macros so the following
8377 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
8380 Since @code{vec_add} is a macro, the vector constant in the example
8381 is treated as four separate arguments. Wrap the entire argument in
8382 parentheses for this to work.
8385 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
8386 Internally, GCC uses built-in functions to achieve the functionality in
8387 the aforementioned header file, but they are not supported and are
8388 subject to change without notice.
8390 The following interfaces are supported for the generic and specific
8391 AltiVec operations and the AltiVec predicates. In cases where there
8392 is a direct mapping between generic and specific operations, only the
8393 generic names are shown here, although the specific operations can also
8396 Arguments that are documented as @code{const int} require literal
8397 integral values within the range required for that operation.
8400 vector signed char vec_abs (vector signed char);
8401 vector signed short vec_abs (vector signed short);
8402 vector signed int vec_abs (vector signed int);
8403 vector float vec_abs (vector float);
8405 vector signed char vec_abss (vector signed char);
8406 vector signed short vec_abss (vector signed short);
8407 vector signed int vec_abss (vector signed int);
8409 vector signed char vec_add (vector bool char, vector signed char);
8410 vector signed char vec_add (vector signed char, vector bool char);
8411 vector signed char vec_add (vector signed char, vector signed char);
8412 vector unsigned char vec_add (vector bool char, vector unsigned char);
8413 vector unsigned char vec_add (vector unsigned char, vector bool char);
8414 vector unsigned char vec_add (vector unsigned char,
8415 vector unsigned char);
8416 vector signed short vec_add (vector bool short, vector signed short);
8417 vector signed short vec_add (vector signed short, vector bool short);
8418 vector signed short vec_add (vector signed short, vector signed short);
8419 vector unsigned short vec_add (vector bool short,
8420 vector unsigned short);
8421 vector unsigned short vec_add (vector unsigned short,
8423 vector unsigned short vec_add (vector unsigned short,
8424 vector unsigned short);
8425 vector signed int vec_add (vector bool int, vector signed int);
8426 vector signed int vec_add (vector signed int, vector bool int);
8427 vector signed int vec_add (vector signed int, vector signed int);
8428 vector unsigned int vec_add (vector bool int, vector unsigned int);
8429 vector unsigned int vec_add (vector unsigned int, vector bool int);
8430 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
8431 vector float vec_add (vector float, vector float);
8433 vector float vec_vaddfp (vector float, vector float);
8435 vector signed int vec_vadduwm (vector bool int, vector signed int);
8436 vector signed int vec_vadduwm (vector signed int, vector bool int);
8437 vector signed int vec_vadduwm (vector signed int, vector signed int);
8438 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
8439 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
8440 vector unsigned int vec_vadduwm (vector unsigned int,
8441 vector unsigned int);
8443 vector signed short vec_vadduhm (vector bool short,
8444 vector signed short);
8445 vector signed short vec_vadduhm (vector signed short,
8447 vector signed short vec_vadduhm (vector signed short,
8448 vector signed short);
8449 vector unsigned short vec_vadduhm (vector bool short,
8450 vector unsigned short);
8451 vector unsigned short vec_vadduhm (vector unsigned short,
8453 vector unsigned short vec_vadduhm (vector unsigned short,
8454 vector unsigned short);
8456 vector signed char vec_vaddubm (vector bool char, vector signed char);
8457 vector signed char vec_vaddubm (vector signed char, vector bool char);
8458 vector signed char vec_vaddubm (vector signed char, vector signed char);
8459 vector unsigned char vec_vaddubm (vector bool char,
8460 vector unsigned char);
8461 vector unsigned char vec_vaddubm (vector unsigned char,
8463 vector unsigned char vec_vaddubm (vector unsigned char,
8464 vector unsigned char);
8466 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
8468 vector unsigned char vec_adds (vector bool char, vector unsigned char);
8469 vector unsigned char vec_adds (vector unsigned char, vector bool char);
8470 vector unsigned char vec_adds (vector unsigned char,
8471 vector unsigned char);
8472 vector signed char vec_adds (vector bool char, vector signed char);
8473 vector signed char vec_adds (vector signed char, vector bool char);
8474 vector signed char vec_adds (vector signed char, vector signed char);
8475 vector unsigned short vec_adds (vector bool short,
8476 vector unsigned short);
8477 vector unsigned short vec_adds (vector unsigned short,
8479 vector unsigned short vec_adds (vector unsigned short,
8480 vector unsigned short);
8481 vector signed short vec_adds (vector bool short, vector signed short);
8482 vector signed short vec_adds (vector signed short, vector bool short);
8483 vector signed short vec_adds (vector signed short, vector signed short);
8484 vector unsigned int vec_adds (vector bool int, vector unsigned int);
8485 vector unsigned int vec_adds (vector unsigned int, vector bool int);
8486 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
8487 vector signed int vec_adds (vector bool int, vector signed int);
8488 vector signed int vec_adds (vector signed int, vector bool int);
8489 vector signed int vec_adds (vector signed int, vector signed int);
8491 vector signed int vec_vaddsws (vector bool int, vector signed int);
8492 vector signed int vec_vaddsws (vector signed int, vector bool int);
8493 vector signed int vec_vaddsws (vector signed int, vector signed int);
8495 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
8496 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
8497 vector unsigned int vec_vadduws (vector unsigned int,
8498 vector unsigned int);
8500 vector signed short vec_vaddshs (vector bool short,
8501 vector signed short);
8502 vector signed short vec_vaddshs (vector signed short,
8504 vector signed short vec_vaddshs (vector signed short,
8505 vector signed short);
8507 vector unsigned short vec_vadduhs (vector bool short,
8508 vector unsigned short);
8509 vector unsigned short vec_vadduhs (vector unsigned short,
8511 vector unsigned short vec_vadduhs (vector unsigned short,
8512 vector unsigned short);
8514 vector signed char vec_vaddsbs (vector bool char, vector signed char);
8515 vector signed char vec_vaddsbs (vector signed char, vector bool char);
8516 vector signed char vec_vaddsbs (vector signed char, vector signed char);
8518 vector unsigned char vec_vaddubs (vector bool char,
8519 vector unsigned char);
8520 vector unsigned char vec_vaddubs (vector unsigned char,
8522 vector unsigned char vec_vaddubs (vector unsigned char,
8523 vector unsigned char);
8525 vector float vec_and (vector float, vector float);
8526 vector float vec_and (vector float, vector bool int);
8527 vector float vec_and (vector bool int, vector float);
8528 vector bool int vec_and (vector bool int, vector bool int);
8529 vector signed int vec_and (vector bool int, vector signed int);
8530 vector signed int vec_and (vector signed int, vector bool int);
8531 vector signed int vec_and (vector signed int, vector signed int);
8532 vector unsigned int vec_and (vector bool int, vector unsigned int);
8533 vector unsigned int vec_and (vector unsigned int, vector bool int);
8534 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
8535 vector bool short vec_and (vector bool short, vector bool short);
8536 vector signed short vec_and (vector bool short, vector signed short);
8537 vector signed short vec_and (vector signed short, vector bool short);
8538 vector signed short vec_and (vector signed short, vector signed short);
8539 vector unsigned short vec_and (vector bool short,
8540 vector unsigned short);
8541 vector unsigned short vec_and (vector unsigned short,
8543 vector unsigned short vec_and (vector unsigned short,
8544 vector unsigned short);
8545 vector signed char vec_and (vector bool char, vector signed char);
8546 vector bool char vec_and (vector bool char, vector bool char);
8547 vector signed char vec_and (vector signed char, vector bool char);
8548 vector signed char vec_and (vector signed char, vector signed char);
8549 vector unsigned char vec_and (vector bool char, vector unsigned char);
8550 vector unsigned char vec_and (vector unsigned char, vector bool char);
8551 vector unsigned char vec_and (vector unsigned char,
8552 vector unsigned char);
8554 vector float vec_andc (vector float, vector float);
8555 vector float vec_andc (vector float, vector bool int);
8556 vector float vec_andc (vector bool int, vector float);
8557 vector bool int vec_andc (vector bool int, vector bool int);
8558 vector signed int vec_andc (vector bool int, vector signed int);
8559 vector signed int vec_andc (vector signed int, vector bool int);
8560 vector signed int vec_andc (vector signed int, vector signed int);
8561 vector unsigned int vec_andc (vector bool int, vector unsigned int);
8562 vector unsigned int vec_andc (vector unsigned int, vector bool int);
8563 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
8564 vector bool short vec_andc (vector bool short, vector bool short);
8565 vector signed short vec_andc (vector bool short, vector signed short);
8566 vector signed short vec_andc (vector signed short, vector bool short);
8567 vector signed short vec_andc (vector signed short, vector signed short);
8568 vector unsigned short vec_andc (vector bool short,
8569 vector unsigned short);
8570 vector unsigned short vec_andc (vector unsigned short,
8572 vector unsigned short vec_andc (vector unsigned short,
8573 vector unsigned short);
8574 vector signed char vec_andc (vector bool char, vector signed char);
8575 vector bool char vec_andc (vector bool char, vector bool char);
8576 vector signed char vec_andc (vector signed char, vector bool char);
8577 vector signed char vec_andc (vector signed char, vector signed char);
8578 vector unsigned char vec_andc (vector bool char, vector unsigned char);
8579 vector unsigned char vec_andc (vector unsigned char, vector bool char);
8580 vector unsigned char vec_andc (vector unsigned char,
8581 vector unsigned char);
8583 vector unsigned char vec_avg (vector unsigned char,
8584 vector unsigned char);
8585 vector signed char vec_avg (vector signed char, vector signed char);
8586 vector unsigned short vec_avg (vector unsigned short,
8587 vector unsigned short);
8588 vector signed short vec_avg (vector signed short, vector signed short);
8589 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
8590 vector signed int vec_avg (vector signed int, vector signed int);
8592 vector signed int vec_vavgsw (vector signed int, vector signed int);
8594 vector unsigned int vec_vavguw (vector unsigned int,
8595 vector unsigned int);
8597 vector signed short vec_vavgsh (vector signed short,
8598 vector signed short);
8600 vector unsigned short vec_vavguh (vector unsigned short,
8601 vector unsigned short);
8603 vector signed char vec_vavgsb (vector signed char, vector signed char);
8605 vector unsigned char vec_vavgub (vector unsigned char,
8606 vector unsigned char);
8608 vector float vec_ceil (vector float);
8610 vector signed int vec_cmpb (vector float, vector float);
8612 vector bool char vec_cmpeq (vector signed char, vector signed char);
8613 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
8614 vector bool short vec_cmpeq (vector signed short, vector signed short);
8615 vector bool short vec_cmpeq (vector unsigned short,
8616 vector unsigned short);
8617 vector bool int vec_cmpeq (vector signed int, vector signed int);
8618 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
8619 vector bool int vec_cmpeq (vector float, vector float);
8621 vector bool int vec_vcmpeqfp (vector float, vector float);
8623 vector bool int vec_vcmpequw (vector signed int, vector signed int);
8624 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
8626 vector bool short vec_vcmpequh (vector signed short,
8627 vector signed short);
8628 vector bool short vec_vcmpequh (vector unsigned short,
8629 vector unsigned short);
8631 vector bool char vec_vcmpequb (vector signed char, vector signed char);
8632 vector bool char vec_vcmpequb (vector unsigned char,
8633 vector unsigned char);
8635 vector bool int vec_cmpge (vector float, vector float);
8637 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
8638 vector bool char vec_cmpgt (vector signed char, vector signed char);
8639 vector bool short vec_cmpgt (vector unsigned short,
8640 vector unsigned short);
8641 vector bool short vec_cmpgt (vector signed short, vector signed short);
8642 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
8643 vector bool int vec_cmpgt (vector signed int, vector signed int);
8644 vector bool int vec_cmpgt (vector float, vector float);
8646 vector bool int vec_vcmpgtfp (vector float, vector float);
8648 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
8650 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
8652 vector bool short vec_vcmpgtsh (vector signed short,
8653 vector signed short);
8655 vector bool short vec_vcmpgtuh (vector unsigned short,
8656 vector unsigned short);
8658 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
8660 vector bool char vec_vcmpgtub (vector unsigned char,
8661 vector unsigned char);
8663 vector bool int vec_cmple (vector float, vector float);
8665 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
8666 vector bool char vec_cmplt (vector signed char, vector signed char);
8667 vector bool short vec_cmplt (vector unsigned short,
8668 vector unsigned short);
8669 vector bool short vec_cmplt (vector signed short, vector signed short);
8670 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
8671 vector bool int vec_cmplt (vector signed int, vector signed int);
8672 vector bool int vec_cmplt (vector float, vector float);
8674 vector float vec_ctf (vector unsigned int, const int);
8675 vector float vec_ctf (vector signed int, const int);
8677 vector float vec_vcfsx (vector signed int, const int);
8679 vector float vec_vcfux (vector unsigned int, const int);
8681 vector signed int vec_cts (vector float, const int);
8683 vector unsigned int vec_ctu (vector float, const int);
8685 void vec_dss (const int);
8687 void vec_dssall (void);
8689 void vec_dst (const vector unsigned char *, int, const int);
8690 void vec_dst (const vector signed char *, int, const int);
8691 void vec_dst (const vector bool char *, int, const int);
8692 void vec_dst (const vector unsigned short *, int, const int);
8693 void vec_dst (const vector signed short *, int, const int);
8694 void vec_dst (const vector bool short *, int, const int);
8695 void vec_dst (const vector pixel *, int, const int);
8696 void vec_dst (const vector unsigned int *, int, const int);
8697 void vec_dst (const vector signed int *, int, const int);
8698 void vec_dst (const vector bool int *, int, const int);
8699 void vec_dst (const vector float *, int, const int);
8700 void vec_dst (const unsigned char *, int, const int);
8701 void vec_dst (const signed char *, int, const int);
8702 void vec_dst (const unsigned short *, int, const int);
8703 void vec_dst (const short *, int, const int);
8704 void vec_dst (const unsigned int *, int, const int);
8705 void vec_dst (const int *, int, const int);
8706 void vec_dst (const unsigned long *, int, const int);
8707 void vec_dst (const long *, int, const int);
8708 void vec_dst (const float *, int, const int);
8710 void vec_dstst (const vector unsigned char *, int, const int);
8711 void vec_dstst (const vector signed char *, int, const int);
8712 void vec_dstst (const vector bool char *, int, const int);
8713 void vec_dstst (const vector unsigned short *, int, const int);
8714 void vec_dstst (const vector signed short *, int, const int);
8715 void vec_dstst (const vector bool short *, int, const int);
8716 void vec_dstst (const vector pixel *, int, const int);
8717 void vec_dstst (const vector unsigned int *, int, const int);
8718 void vec_dstst (const vector signed int *, int, const int);
8719 void vec_dstst (const vector bool int *, int, const int);
8720 void vec_dstst (const vector float *, int, const int);
8721 void vec_dstst (const unsigned char *, int, const int);
8722 void vec_dstst (const signed char *, int, const int);
8723 void vec_dstst (const unsigned short *, int, const int);
8724 void vec_dstst (const short *, int, const int);
8725 void vec_dstst (const unsigned int *, int, const int);
8726 void vec_dstst (const int *, int, const int);
8727 void vec_dstst (const unsigned long *, int, const int);
8728 void vec_dstst (const long *, int, const int);
8729 void vec_dstst (const float *, int, const int);
8731 void vec_dststt (const vector unsigned char *, int, const int);
8732 void vec_dststt (const vector signed char *, int, const int);
8733 void vec_dststt (const vector bool char *, int, const int);
8734 void vec_dststt (const vector unsigned short *, int, const int);
8735 void vec_dststt (const vector signed short *, int, const int);
8736 void vec_dststt (const vector bool short *, int, const int);
8737 void vec_dststt (const vector pixel *, int, const int);
8738 void vec_dststt (const vector unsigned int *, int, const int);
8739 void vec_dststt (const vector signed int *, int, const int);
8740 void vec_dststt (const vector bool int *, int, const int);
8741 void vec_dststt (const vector float *, int, const int);
8742 void vec_dststt (const unsigned char *, int, const int);
8743 void vec_dststt (const signed char *, int, const int);
8744 void vec_dststt (const unsigned short *, int, const int);
8745 void vec_dststt (const short *, int, const int);
8746 void vec_dststt (const unsigned int *, int, const int);
8747 void vec_dststt (const int *, int, const int);
8748 void vec_dststt (const unsigned long *, int, const int);
8749 void vec_dststt (const long *, int, const int);
8750 void vec_dststt (const float *, int, const int);
8752 void vec_dstt (const vector unsigned char *, int, const int);
8753 void vec_dstt (const vector signed char *, int, const int);
8754 void vec_dstt (const vector bool char *, int, const int);
8755 void vec_dstt (const vector unsigned short *, int, const int);
8756 void vec_dstt (const vector signed short *, int, const int);
8757 void vec_dstt (const vector bool short *, int, const int);
8758 void vec_dstt (const vector pixel *, int, const int);
8759 void vec_dstt (const vector unsigned int *, int, const int);
8760 void vec_dstt (const vector signed int *, int, const int);
8761 void vec_dstt (const vector bool int *, int, const int);
8762 void vec_dstt (const vector float *, int, const int);
8763 void vec_dstt (const unsigned char *, int, const int);
8764 void vec_dstt (const signed char *, int, const int);
8765 void vec_dstt (const unsigned short *, int, const int);
8766 void vec_dstt (const short *, int, const int);
8767 void vec_dstt (const unsigned int *, int, const int);
8768 void vec_dstt (const int *, int, const int);
8769 void vec_dstt (const unsigned long *, int, const int);
8770 void vec_dstt (const long *, int, const int);
8771 void vec_dstt (const float *, int, const int);
8773 vector float vec_expte (vector float);
8775 vector float vec_floor (vector float);
8777 vector float vec_ld (int, const vector float *);
8778 vector float vec_ld (int, const float *);
8779 vector bool int vec_ld (int, const vector bool int *);
8780 vector signed int vec_ld (int, const vector signed int *);
8781 vector signed int vec_ld (int, const int *);
8782 vector signed int vec_ld (int, const long *);
8783 vector unsigned int vec_ld (int, const vector unsigned int *);
8784 vector unsigned int vec_ld (int, const unsigned int *);
8785 vector unsigned int vec_ld (int, const unsigned long *);
8786 vector bool short vec_ld (int, const vector bool short *);
8787 vector pixel vec_ld (int, const vector pixel *);
8788 vector signed short vec_ld (int, const vector signed short *);
8789 vector signed short vec_ld (int, const short *);
8790 vector unsigned short vec_ld (int, const vector unsigned short *);
8791 vector unsigned short vec_ld (int, const unsigned short *);
8792 vector bool char vec_ld (int, const vector bool char *);
8793 vector signed char vec_ld (int, const vector signed char *);
8794 vector signed char vec_ld (int, const signed char *);
8795 vector unsigned char vec_ld (int, const vector unsigned char *);
8796 vector unsigned char vec_ld (int, const unsigned char *);
8798 vector signed char vec_lde (int, const signed char *);
8799 vector unsigned char vec_lde (int, const unsigned char *);
8800 vector signed short vec_lde (int, const short *);
8801 vector unsigned short vec_lde (int, const unsigned short *);
8802 vector float vec_lde (int, const float *);
8803 vector signed int vec_lde (int, const int *);
8804 vector unsigned int vec_lde (int, const unsigned int *);
8805 vector signed int vec_lde (int, const long *);
8806 vector unsigned int vec_lde (int, const unsigned long *);
8808 vector float vec_lvewx (int, float *);
8809 vector signed int vec_lvewx (int, int *);
8810 vector unsigned int vec_lvewx (int, unsigned int *);
8811 vector signed int vec_lvewx (int, long *);
8812 vector unsigned int vec_lvewx (int, unsigned long *);
8814 vector signed short vec_lvehx (int, short *);
8815 vector unsigned short vec_lvehx (int, unsigned short *);
8817 vector signed char vec_lvebx (int, char *);
8818 vector unsigned char vec_lvebx (int, unsigned char *);
8820 vector float vec_ldl (int, const vector float *);
8821 vector float vec_ldl (int, const float *);
8822 vector bool int vec_ldl (int, const vector bool int *);
8823 vector signed int vec_ldl (int, const vector signed int *);
8824 vector signed int vec_ldl (int, const int *);
8825 vector signed int vec_ldl (int, const long *);
8826 vector unsigned int vec_ldl (int, const vector unsigned int *);
8827 vector unsigned int vec_ldl (int, const unsigned int *);
8828 vector unsigned int vec_ldl (int, const unsigned long *);
8829 vector bool short vec_ldl (int, const vector bool short *);
8830 vector pixel vec_ldl (int, const vector pixel *);
8831 vector signed short vec_ldl (int, const vector signed short *);
8832 vector signed short vec_ldl (int, const short *);
8833 vector unsigned short vec_ldl (int, const vector unsigned short *);
8834 vector unsigned short vec_ldl (int, const unsigned short *);
8835 vector bool char vec_ldl (int, const vector bool char *);
8836 vector signed char vec_ldl (int, const vector signed char *);
8837 vector signed char vec_ldl (int, const signed char *);
8838 vector unsigned char vec_ldl (int, const vector unsigned char *);
8839 vector unsigned char vec_ldl (int, const unsigned char *);
8841 vector float vec_loge (vector float);
8843 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
8844 vector unsigned char vec_lvsl (int, const volatile signed char *);
8845 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
8846 vector unsigned char vec_lvsl (int, const volatile short *);
8847 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
8848 vector unsigned char vec_lvsl (int, const volatile int *);
8849 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
8850 vector unsigned char vec_lvsl (int, const volatile long *);
8851 vector unsigned char vec_lvsl (int, const volatile float *);
8853 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
8854 vector unsigned char vec_lvsr (int, const volatile signed char *);
8855 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
8856 vector unsigned char vec_lvsr (int, const volatile short *);
8857 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
8858 vector unsigned char vec_lvsr (int, const volatile int *);
8859 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
8860 vector unsigned char vec_lvsr (int, const volatile long *);
8861 vector unsigned char vec_lvsr (int, const volatile float *);
8863 vector float vec_madd (vector float, vector float, vector float);
8865 vector signed short vec_madds (vector signed short,
8866 vector signed short,
8867 vector signed short);
8869 vector unsigned char vec_max (vector bool char, vector unsigned char);
8870 vector unsigned char vec_max (vector unsigned char, vector bool char);
8871 vector unsigned char vec_max (vector unsigned char,
8872 vector unsigned char);
8873 vector signed char vec_max (vector bool char, vector signed char);
8874 vector signed char vec_max (vector signed char, vector bool char);
8875 vector signed char vec_max (vector signed char, vector signed char);
8876 vector unsigned short vec_max (vector bool short,
8877 vector unsigned short);
8878 vector unsigned short vec_max (vector unsigned short,
8880 vector unsigned short vec_max (vector unsigned short,
8881 vector unsigned short);
8882 vector signed short vec_max (vector bool short, vector signed short);
8883 vector signed short vec_max (vector signed short, vector bool short);
8884 vector signed short vec_max (vector signed short, vector signed short);
8885 vector unsigned int vec_max (vector bool int, vector unsigned int);
8886 vector unsigned int vec_max (vector unsigned int, vector bool int);
8887 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
8888 vector signed int vec_max (vector bool int, vector signed int);
8889 vector signed int vec_max (vector signed int, vector bool int);
8890 vector signed int vec_max (vector signed int, vector signed int);
8891 vector float vec_max (vector float, vector float);
8893 vector float vec_vmaxfp (vector float, vector float);
8895 vector signed int vec_vmaxsw (vector bool int, vector signed int);
8896 vector signed int vec_vmaxsw (vector signed int, vector bool int);
8897 vector signed int vec_vmaxsw (vector signed int, vector signed int);
8899 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
8900 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
8901 vector unsigned int vec_vmaxuw (vector unsigned int,
8902 vector unsigned int);
8904 vector signed short vec_vmaxsh (vector bool short, vector signed short);
8905 vector signed short vec_vmaxsh (vector signed short, vector bool short);
8906 vector signed short vec_vmaxsh (vector signed short,
8907 vector signed short);
8909 vector unsigned short vec_vmaxuh (vector bool short,
8910 vector unsigned short);
8911 vector unsigned short vec_vmaxuh (vector unsigned short,
8913 vector unsigned short vec_vmaxuh (vector unsigned short,
8914 vector unsigned short);
8916 vector signed char vec_vmaxsb (vector bool char, vector signed char);
8917 vector signed char vec_vmaxsb (vector signed char, vector bool char);
8918 vector signed char vec_vmaxsb (vector signed char, vector signed char);
8920 vector unsigned char vec_vmaxub (vector bool char,
8921 vector unsigned char);
8922 vector unsigned char vec_vmaxub (vector unsigned char,
8924 vector unsigned char vec_vmaxub (vector unsigned char,
8925 vector unsigned char);
8927 vector bool char vec_mergeh (vector bool char, vector bool char);
8928 vector signed char vec_mergeh (vector signed char, vector signed char);
8929 vector unsigned char vec_mergeh (vector unsigned char,
8930 vector unsigned char);
8931 vector bool short vec_mergeh (vector bool short, vector bool short);
8932 vector pixel vec_mergeh (vector pixel, vector pixel);
8933 vector signed short vec_mergeh (vector signed short,
8934 vector signed short);
8935 vector unsigned short vec_mergeh (vector unsigned short,
8936 vector unsigned short);
8937 vector float vec_mergeh (vector float, vector float);
8938 vector bool int vec_mergeh (vector bool int, vector bool int);
8939 vector signed int vec_mergeh (vector signed int, vector signed int);
8940 vector unsigned int vec_mergeh (vector unsigned int,
8941 vector unsigned int);
8943 vector float vec_vmrghw (vector float, vector float);
8944 vector bool int vec_vmrghw (vector bool int, vector bool int);
8945 vector signed int vec_vmrghw (vector signed int, vector signed int);
8946 vector unsigned int vec_vmrghw (vector unsigned int,
8947 vector unsigned int);
8949 vector bool short vec_vmrghh (vector bool short, vector bool short);
8950 vector signed short vec_vmrghh (vector signed short,
8951 vector signed short);
8952 vector unsigned short vec_vmrghh (vector unsigned short,
8953 vector unsigned short);
8954 vector pixel vec_vmrghh (vector pixel, vector pixel);
8956 vector bool char vec_vmrghb (vector bool char, vector bool char);
8957 vector signed char vec_vmrghb (vector signed char, vector signed char);
8958 vector unsigned char vec_vmrghb (vector unsigned char,
8959 vector unsigned char);
8961 vector bool char vec_mergel (vector bool char, vector bool char);
8962 vector signed char vec_mergel (vector signed char, vector signed char);
8963 vector unsigned char vec_mergel (vector unsigned char,
8964 vector unsigned char);
8965 vector bool short vec_mergel (vector bool short, vector bool short);
8966 vector pixel vec_mergel (vector pixel, vector pixel);
8967 vector signed short vec_mergel (vector signed short,
8968 vector signed short);
8969 vector unsigned short vec_mergel (vector unsigned short,
8970 vector unsigned short);
8971 vector float vec_mergel (vector float, vector float);
8972 vector bool int vec_mergel (vector bool int, vector bool int);
8973 vector signed int vec_mergel (vector signed int, vector signed int);
8974 vector unsigned int vec_mergel (vector unsigned int,
8975 vector unsigned int);
8977 vector float vec_vmrglw (vector float, vector float);
8978 vector signed int vec_vmrglw (vector signed int, vector signed int);
8979 vector unsigned int vec_vmrglw (vector unsigned int,
8980 vector unsigned int);
8981 vector bool int vec_vmrglw (vector bool int, vector bool int);
8983 vector bool short vec_vmrglh (vector bool short, vector bool short);
8984 vector signed short vec_vmrglh (vector signed short,
8985 vector signed short);
8986 vector unsigned short vec_vmrglh (vector unsigned short,
8987 vector unsigned short);
8988 vector pixel vec_vmrglh (vector pixel, vector pixel);
8990 vector bool char vec_vmrglb (vector bool char, vector bool char);
8991 vector signed char vec_vmrglb (vector signed char, vector signed char);
8992 vector unsigned char vec_vmrglb (vector unsigned char,
8993 vector unsigned char);
8995 vector unsigned short vec_mfvscr (void);
8997 vector unsigned char vec_min (vector bool char, vector unsigned char);
8998 vector unsigned char vec_min (vector unsigned char, vector bool char);
8999 vector unsigned char vec_min (vector unsigned char,
9000 vector unsigned char);
9001 vector signed char vec_min (vector bool char, vector signed char);
9002 vector signed char vec_min (vector signed char, vector bool char);
9003 vector signed char vec_min (vector signed char, vector signed char);
9004 vector unsigned short vec_min (vector bool short,
9005 vector unsigned short);
9006 vector unsigned short vec_min (vector unsigned short,
9008 vector unsigned short vec_min (vector unsigned short,
9009 vector unsigned short);
9010 vector signed short vec_min (vector bool short, vector signed short);
9011 vector signed short vec_min (vector signed short, vector bool short);
9012 vector signed short vec_min (vector signed short, vector signed short);
9013 vector unsigned int vec_min (vector bool int, vector unsigned int);
9014 vector unsigned int vec_min (vector unsigned int, vector bool int);
9015 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
9016 vector signed int vec_min (vector bool int, vector signed int);
9017 vector signed int vec_min (vector signed int, vector bool int);
9018 vector signed int vec_min (vector signed int, vector signed int);
9019 vector float vec_min (vector float, vector float);
9021 vector float vec_vminfp (vector float, vector float);
9023 vector signed int vec_vminsw (vector bool int, vector signed int);
9024 vector signed int vec_vminsw (vector signed int, vector bool int);
9025 vector signed int vec_vminsw (vector signed int, vector signed int);
9027 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
9028 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
9029 vector unsigned int vec_vminuw (vector unsigned int,
9030 vector unsigned int);
9032 vector signed short vec_vminsh (vector bool short, vector signed short);
9033 vector signed short vec_vminsh (vector signed short, vector bool short);
9034 vector signed short vec_vminsh (vector signed short,
9035 vector signed short);
9037 vector unsigned short vec_vminuh (vector bool short,
9038 vector unsigned short);
9039 vector unsigned short vec_vminuh (vector unsigned short,
9041 vector unsigned short vec_vminuh (vector unsigned short,
9042 vector unsigned short);
9044 vector signed char vec_vminsb (vector bool char, vector signed char);
9045 vector signed char vec_vminsb (vector signed char, vector bool char);
9046 vector signed char vec_vminsb (vector signed char, vector signed char);
9048 vector unsigned char vec_vminub (vector bool char,
9049 vector unsigned char);
9050 vector unsigned char vec_vminub (vector unsigned char,
9052 vector unsigned char vec_vminub (vector unsigned char,
9053 vector unsigned char);
9055 vector signed short vec_mladd (vector signed short,
9056 vector signed short,
9057 vector signed short);
9058 vector signed short vec_mladd (vector signed short,
9059 vector unsigned short,
9060 vector unsigned short);
9061 vector signed short vec_mladd (vector unsigned short,
9062 vector signed short,
9063 vector signed short);
9064 vector unsigned short vec_mladd (vector unsigned short,
9065 vector unsigned short,
9066 vector unsigned short);
9068 vector signed short vec_mradds (vector signed short,
9069 vector signed short,
9070 vector signed short);
9072 vector unsigned int vec_msum (vector unsigned char,
9073 vector unsigned char,
9074 vector unsigned int);
9075 vector signed int vec_msum (vector signed char,
9076 vector unsigned char,
9078 vector unsigned int vec_msum (vector unsigned short,
9079 vector unsigned short,
9080 vector unsigned int);
9081 vector signed int vec_msum (vector signed short,
9082 vector signed short,
9085 vector signed int vec_vmsumshm (vector signed short,
9086 vector signed short,
9089 vector unsigned int vec_vmsumuhm (vector unsigned short,
9090 vector unsigned short,
9091 vector unsigned int);
9093 vector signed int vec_vmsummbm (vector signed char,
9094 vector unsigned char,
9097 vector unsigned int vec_vmsumubm (vector unsigned char,
9098 vector unsigned char,
9099 vector unsigned int);
9101 vector unsigned int vec_msums (vector unsigned short,
9102 vector unsigned short,
9103 vector unsigned int);
9104 vector signed int vec_msums (vector signed short,
9105 vector signed short,
9108 vector signed int vec_vmsumshs (vector signed short,
9109 vector signed short,
9112 vector unsigned int vec_vmsumuhs (vector unsigned short,
9113 vector unsigned short,
9114 vector unsigned int);
9116 void vec_mtvscr (vector signed int);
9117 void vec_mtvscr (vector unsigned int);
9118 void vec_mtvscr (vector bool int);
9119 void vec_mtvscr (vector signed short);
9120 void vec_mtvscr (vector unsigned short);
9121 void vec_mtvscr (vector bool short);
9122 void vec_mtvscr (vector pixel);
9123 void vec_mtvscr (vector signed char);
9124 void vec_mtvscr (vector unsigned char);
9125 void vec_mtvscr (vector bool char);
9127 vector unsigned short vec_mule (vector unsigned char,
9128 vector unsigned char);
9129 vector signed short vec_mule (vector signed char,
9130 vector signed char);
9131 vector unsigned int vec_mule (vector unsigned short,
9132 vector unsigned short);
9133 vector signed int vec_mule (vector signed short, vector signed short);
9135 vector signed int vec_vmulesh (vector signed short,
9136 vector signed short);
9138 vector unsigned int vec_vmuleuh (vector unsigned short,
9139 vector unsigned short);
9141 vector signed short vec_vmulesb (vector signed char,
9142 vector signed char);
9144 vector unsigned short vec_vmuleub (vector unsigned char,
9145 vector unsigned char);
9147 vector unsigned short vec_mulo (vector unsigned char,
9148 vector unsigned char);
9149 vector signed short vec_mulo (vector signed char, vector signed char);
9150 vector unsigned int vec_mulo (vector unsigned short,
9151 vector unsigned short);
9152 vector signed int vec_mulo (vector signed short, vector signed short);
9154 vector signed int vec_vmulosh (vector signed short,
9155 vector signed short);
9157 vector unsigned int vec_vmulouh (vector unsigned short,
9158 vector unsigned short);
9160 vector signed short vec_vmulosb (vector signed char,
9161 vector signed char);
9163 vector unsigned short vec_vmuloub (vector unsigned char,
9164 vector unsigned char);
9166 vector float vec_nmsub (vector float, vector float, vector float);
9168 vector float vec_nor (vector float, vector float);
9169 vector signed int vec_nor (vector signed int, vector signed int);
9170 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
9171 vector bool int vec_nor (vector bool int, vector bool int);
9172 vector signed short vec_nor (vector signed short, vector signed short);
9173 vector unsigned short vec_nor (vector unsigned short,
9174 vector unsigned short);
9175 vector bool short vec_nor (vector bool short, vector bool short);
9176 vector signed char vec_nor (vector signed char, vector signed char);
9177 vector unsigned char vec_nor (vector unsigned char,
9178 vector unsigned char);
9179 vector bool char vec_nor (vector bool char, vector bool char);
9181 vector float vec_or (vector float, vector float);
9182 vector float vec_or (vector float, vector bool int);
9183 vector float vec_or (vector bool int, vector float);
9184 vector bool int vec_or (vector bool int, vector bool int);
9185 vector signed int vec_or (vector bool int, vector signed int);
9186 vector signed int vec_or (vector signed int, vector bool int);
9187 vector signed int vec_or (vector signed int, vector signed int);
9188 vector unsigned int vec_or (vector bool int, vector unsigned int);
9189 vector unsigned int vec_or (vector unsigned int, vector bool int);
9190 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
9191 vector bool short vec_or (vector bool short, vector bool short);
9192 vector signed short vec_or (vector bool short, vector signed short);
9193 vector signed short vec_or (vector signed short, vector bool short);
9194 vector signed short vec_or (vector signed short, vector signed short);
9195 vector unsigned short vec_or (vector bool short, vector unsigned short);
9196 vector unsigned short vec_or (vector unsigned short, vector bool short);
9197 vector unsigned short vec_or (vector unsigned short,
9198 vector unsigned short);
9199 vector signed char vec_or (vector bool char, vector signed char);
9200 vector bool char vec_or (vector bool char, vector bool char);
9201 vector signed char vec_or (vector signed char, vector bool char);
9202 vector signed char vec_or (vector signed char, vector signed char);
9203 vector unsigned char vec_or (vector bool char, vector unsigned char);
9204 vector unsigned char vec_or (vector unsigned char, vector bool char);
9205 vector unsigned char vec_or (vector unsigned char,
9206 vector unsigned char);
9208 vector signed char vec_pack (vector signed short, vector signed short);
9209 vector unsigned char vec_pack (vector unsigned short,
9210 vector unsigned short);
9211 vector bool char vec_pack (vector bool short, vector bool short);
9212 vector signed short vec_pack (vector signed int, vector signed int);
9213 vector unsigned short vec_pack (vector unsigned int,
9214 vector unsigned int);
9215 vector bool short vec_pack (vector bool int, vector bool int);
9217 vector bool short vec_vpkuwum (vector bool int, vector bool int);
9218 vector signed short vec_vpkuwum (vector signed int, vector signed int);
9219 vector unsigned short vec_vpkuwum (vector unsigned int,
9220 vector unsigned int);
9222 vector bool char vec_vpkuhum (vector bool short, vector bool short);
9223 vector signed char vec_vpkuhum (vector signed short,
9224 vector signed short);
9225 vector unsigned char vec_vpkuhum (vector unsigned short,
9226 vector unsigned short);
9228 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
9230 vector unsigned char vec_packs (vector unsigned short,
9231 vector unsigned short);
9232 vector signed char vec_packs (vector signed short, vector signed short);
9233 vector unsigned short vec_packs (vector unsigned int,
9234 vector unsigned int);
9235 vector signed short vec_packs (vector signed int, vector signed int);
9237 vector signed short vec_vpkswss (vector signed int, vector signed int);
9239 vector unsigned short vec_vpkuwus (vector unsigned int,
9240 vector unsigned int);
9242 vector signed char vec_vpkshss (vector signed short,
9243 vector signed short);
9245 vector unsigned char vec_vpkuhus (vector unsigned short,
9246 vector unsigned short);
9248 vector unsigned char vec_packsu (vector unsigned short,
9249 vector unsigned short);
9250 vector unsigned char vec_packsu (vector signed short,
9251 vector signed short);
9252 vector unsigned short vec_packsu (vector unsigned int,
9253 vector unsigned int);
9254 vector unsigned short vec_packsu (vector signed int, vector signed int);
9256 vector unsigned short vec_vpkswus (vector signed int,
9259 vector unsigned char vec_vpkshus (vector signed short,
9260 vector signed short);
9262 vector float vec_perm (vector float,
9264 vector unsigned char);
9265 vector signed int vec_perm (vector signed int,
9267 vector unsigned char);
9268 vector unsigned int vec_perm (vector unsigned int,
9269 vector unsigned int,
9270 vector unsigned char);
9271 vector bool int vec_perm (vector bool int,
9273 vector unsigned char);
9274 vector signed short vec_perm (vector signed short,
9275 vector signed short,
9276 vector unsigned char);
9277 vector unsigned short vec_perm (vector unsigned short,
9278 vector unsigned short,
9279 vector unsigned char);
9280 vector bool short vec_perm (vector bool short,
9282 vector unsigned char);
9283 vector pixel vec_perm (vector pixel,
9285 vector unsigned char);
9286 vector signed char vec_perm (vector signed char,
9288 vector unsigned char);
9289 vector unsigned char vec_perm (vector unsigned char,
9290 vector unsigned char,
9291 vector unsigned char);
9292 vector bool char vec_perm (vector bool char,
9294 vector unsigned char);
9296 vector float vec_re (vector float);
9298 vector signed char vec_rl (vector signed char,
9299 vector unsigned char);
9300 vector unsigned char vec_rl (vector unsigned char,
9301 vector unsigned char);
9302 vector signed short vec_rl (vector signed short, vector unsigned short);
9303 vector unsigned short vec_rl (vector unsigned short,
9304 vector unsigned short);
9305 vector signed int vec_rl (vector signed int, vector unsigned int);
9306 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
9308 vector signed int vec_vrlw (vector signed int, vector unsigned int);
9309 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
9311 vector signed short vec_vrlh (vector signed short,
9312 vector unsigned short);
9313 vector unsigned short vec_vrlh (vector unsigned short,
9314 vector unsigned short);
9316 vector signed char vec_vrlb (vector signed char, vector unsigned char);
9317 vector unsigned char vec_vrlb (vector unsigned char,
9318 vector unsigned char);
9320 vector float vec_round (vector float);
9322 vector float vec_rsqrte (vector float);
9324 vector float vec_sel (vector float, vector float, vector bool int);
9325 vector float vec_sel (vector float, vector float, vector unsigned int);
9326 vector signed int vec_sel (vector signed int,
9329 vector signed int vec_sel (vector signed int,
9331 vector unsigned int);
9332 vector unsigned int vec_sel (vector unsigned int,
9333 vector unsigned int,
9335 vector unsigned int vec_sel (vector unsigned int,
9336 vector unsigned int,
9337 vector unsigned int);
9338 vector bool int vec_sel (vector bool int,
9341 vector bool int vec_sel (vector bool int,
9343 vector unsigned int);
9344 vector signed short vec_sel (vector signed short,
9345 vector signed short,
9347 vector signed short vec_sel (vector signed short,
9348 vector signed short,
9349 vector unsigned short);
9350 vector unsigned short vec_sel (vector unsigned short,
9351 vector unsigned short,
9353 vector unsigned short vec_sel (vector unsigned short,
9354 vector unsigned short,
9355 vector unsigned short);
9356 vector bool short vec_sel (vector bool short,
9359 vector bool short vec_sel (vector bool short,
9361 vector unsigned short);
9362 vector signed char vec_sel (vector signed char,
9365 vector signed char vec_sel (vector signed char,
9367 vector unsigned char);
9368 vector unsigned char vec_sel (vector unsigned char,
9369 vector unsigned char,
9371 vector unsigned char vec_sel (vector unsigned char,
9372 vector unsigned char,
9373 vector unsigned char);
9374 vector bool char vec_sel (vector bool char,
9377 vector bool char vec_sel (vector bool char,
9379 vector unsigned char);
9381 vector signed char vec_sl (vector signed char,
9382 vector unsigned char);
9383 vector unsigned char vec_sl (vector unsigned char,
9384 vector unsigned char);
9385 vector signed short vec_sl (vector signed short, vector unsigned short);
9386 vector unsigned short vec_sl (vector unsigned short,
9387 vector unsigned short);
9388 vector signed int vec_sl (vector signed int, vector unsigned int);
9389 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
9391 vector signed int vec_vslw (vector signed int, vector unsigned int);
9392 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
9394 vector signed short vec_vslh (vector signed short,
9395 vector unsigned short);
9396 vector unsigned short vec_vslh (vector unsigned short,
9397 vector unsigned short);
9399 vector signed char vec_vslb (vector signed char, vector unsigned char);
9400 vector unsigned char vec_vslb (vector unsigned char,
9401 vector unsigned char);
9403 vector float vec_sld (vector float, vector float, const int);
9404 vector signed int vec_sld (vector signed int,
9407 vector unsigned int vec_sld (vector unsigned int,
9408 vector unsigned int,
9410 vector bool int vec_sld (vector bool int,
9413 vector signed short vec_sld (vector signed short,
9414 vector signed short,
9416 vector unsigned short vec_sld (vector unsigned short,
9417 vector unsigned short,
9419 vector bool short vec_sld (vector bool short,
9422 vector pixel vec_sld (vector pixel,
9425 vector signed char vec_sld (vector signed char,
9428 vector unsigned char vec_sld (vector unsigned char,
9429 vector unsigned char,
9431 vector bool char vec_sld (vector bool char,
9435 vector signed int vec_sll (vector signed int,
9436 vector unsigned int);
9437 vector signed int vec_sll (vector signed int,
9438 vector unsigned short);
9439 vector signed int vec_sll (vector signed int,
9440 vector unsigned char);
9441 vector unsigned int vec_sll (vector unsigned int,
9442 vector unsigned int);
9443 vector unsigned int vec_sll (vector unsigned int,
9444 vector unsigned short);
9445 vector unsigned int vec_sll (vector unsigned int,
9446 vector unsigned char);
9447 vector bool int vec_sll (vector bool int,
9448 vector unsigned int);
9449 vector bool int vec_sll (vector bool int,
9450 vector unsigned short);
9451 vector bool int vec_sll (vector bool int,
9452 vector unsigned char);
9453 vector signed short vec_sll (vector signed short,
9454 vector unsigned int);
9455 vector signed short vec_sll (vector signed short,
9456 vector unsigned short);
9457 vector signed short vec_sll (vector signed short,
9458 vector unsigned char);
9459 vector unsigned short vec_sll (vector unsigned short,
9460 vector unsigned int);
9461 vector unsigned short vec_sll (vector unsigned short,
9462 vector unsigned short);
9463 vector unsigned short vec_sll (vector unsigned short,
9464 vector unsigned char);
9465 vector bool short vec_sll (vector bool short, vector unsigned int);
9466 vector bool short vec_sll (vector bool short, vector unsigned short);
9467 vector bool short vec_sll (vector bool short, vector unsigned char);
9468 vector pixel vec_sll (vector pixel, vector unsigned int);
9469 vector pixel vec_sll (vector pixel, vector unsigned short);
9470 vector pixel vec_sll (vector pixel, vector unsigned char);
9471 vector signed char vec_sll (vector signed char, vector unsigned int);
9472 vector signed char vec_sll (vector signed char, vector unsigned short);
9473 vector signed char vec_sll (vector signed char, vector unsigned char);
9474 vector unsigned char vec_sll (vector unsigned char,
9475 vector unsigned int);
9476 vector unsigned char vec_sll (vector unsigned char,
9477 vector unsigned short);
9478 vector unsigned char vec_sll (vector unsigned char,
9479 vector unsigned char);
9480 vector bool char vec_sll (vector bool char, vector unsigned int);
9481 vector bool char vec_sll (vector bool char, vector unsigned short);
9482 vector bool char vec_sll (vector bool char, vector unsigned char);
9484 vector float vec_slo (vector float, vector signed char);
9485 vector float vec_slo (vector float, vector unsigned char);
9486 vector signed int vec_slo (vector signed int, vector signed char);
9487 vector signed int vec_slo (vector signed int, vector unsigned char);
9488 vector unsigned int vec_slo (vector unsigned int, vector signed char);
9489 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
9490 vector signed short vec_slo (vector signed short, vector signed char);
9491 vector signed short vec_slo (vector signed short, vector unsigned char);
9492 vector unsigned short vec_slo (vector unsigned short,
9493 vector signed char);
9494 vector unsigned short vec_slo (vector unsigned short,
9495 vector unsigned char);
9496 vector pixel vec_slo (vector pixel, vector signed char);
9497 vector pixel vec_slo (vector pixel, vector unsigned char);
9498 vector signed char vec_slo (vector signed char, vector signed char);
9499 vector signed char vec_slo (vector signed char, vector unsigned char);
9500 vector unsigned char vec_slo (vector unsigned char, vector signed char);
9501 vector unsigned char vec_slo (vector unsigned char,
9502 vector unsigned char);
9504 vector signed char vec_splat (vector signed char, const int);
9505 vector unsigned char vec_splat (vector unsigned char, const int);
9506 vector bool char vec_splat (vector bool char, const int);
9507 vector signed short vec_splat (vector signed short, const int);
9508 vector unsigned short vec_splat (vector unsigned short, const int);
9509 vector bool short vec_splat (vector bool short, const int);
9510 vector pixel vec_splat (vector pixel, const int);
9511 vector float vec_splat (vector float, const int);
9512 vector signed int vec_splat (vector signed int, const int);
9513 vector unsigned int vec_splat (vector unsigned int, const int);
9514 vector bool int vec_splat (vector bool int, const int);
9516 vector float vec_vspltw (vector float, const int);
9517 vector signed int vec_vspltw (vector signed int, const int);
9518 vector unsigned int vec_vspltw (vector unsigned int, const int);
9519 vector bool int vec_vspltw (vector bool int, const int);
9521 vector bool short vec_vsplth (vector bool short, const int);
9522 vector signed short vec_vsplth (vector signed short, const int);
9523 vector unsigned short vec_vsplth (vector unsigned short, const int);
9524 vector pixel vec_vsplth (vector pixel, const int);
9526 vector signed char vec_vspltb (vector signed char, const int);
9527 vector unsigned char vec_vspltb (vector unsigned char, const int);
9528 vector bool char vec_vspltb (vector bool char, const int);
9530 vector signed char vec_splat_s8 (const int);
9532 vector signed short vec_splat_s16 (const int);
9534 vector signed int vec_splat_s32 (const int);
9536 vector unsigned char vec_splat_u8 (const int);
9538 vector unsigned short vec_splat_u16 (const int);
9540 vector unsigned int vec_splat_u32 (const int);
9542 vector signed char vec_sr (vector signed char, vector unsigned char);
9543 vector unsigned char vec_sr (vector unsigned char,
9544 vector unsigned char);
9545 vector signed short vec_sr (vector signed short,
9546 vector unsigned short);
9547 vector unsigned short vec_sr (vector unsigned short,
9548 vector unsigned short);
9549 vector signed int vec_sr (vector signed int, vector unsigned int);
9550 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
9552 vector signed int vec_vsrw (vector signed int, vector unsigned int);
9553 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
9555 vector signed short vec_vsrh (vector signed short,
9556 vector unsigned short);
9557 vector unsigned short vec_vsrh (vector unsigned short,
9558 vector unsigned short);
9560 vector signed char vec_vsrb (vector signed char, vector unsigned char);
9561 vector unsigned char vec_vsrb (vector unsigned char,
9562 vector unsigned char);
9564 vector signed char vec_sra (vector signed char, vector unsigned char);
9565 vector unsigned char vec_sra (vector unsigned char,
9566 vector unsigned char);
9567 vector signed short vec_sra (vector signed short,
9568 vector unsigned short);
9569 vector unsigned short vec_sra (vector unsigned short,
9570 vector unsigned short);
9571 vector signed int vec_sra (vector signed int, vector unsigned int);
9572 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
9574 vector signed int vec_vsraw (vector signed int, vector unsigned int);
9575 vector unsigned int vec_vsraw (vector unsigned int,
9576 vector unsigned int);
9578 vector signed short vec_vsrah (vector signed short,
9579 vector unsigned short);
9580 vector unsigned short vec_vsrah (vector unsigned short,
9581 vector unsigned short);
9583 vector signed char vec_vsrab (vector signed char, vector unsigned char);
9584 vector unsigned char vec_vsrab (vector unsigned char,
9585 vector unsigned char);
9587 vector signed int vec_srl (vector signed int, vector unsigned int);
9588 vector signed int vec_srl (vector signed int, vector unsigned short);
9589 vector signed int vec_srl (vector signed int, vector unsigned char);
9590 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
9591 vector unsigned int vec_srl (vector unsigned int,
9592 vector unsigned short);
9593 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
9594 vector bool int vec_srl (vector bool int, vector unsigned int);
9595 vector bool int vec_srl (vector bool int, vector unsigned short);
9596 vector bool int vec_srl (vector bool int, vector unsigned char);
9597 vector signed short vec_srl (vector signed short, vector unsigned int);
9598 vector signed short vec_srl (vector signed short,
9599 vector unsigned short);
9600 vector signed short vec_srl (vector signed short, vector unsigned char);
9601 vector unsigned short vec_srl (vector unsigned short,
9602 vector unsigned int);
9603 vector unsigned short vec_srl (vector unsigned short,
9604 vector unsigned short);
9605 vector unsigned short vec_srl (vector unsigned short,
9606 vector unsigned char);
9607 vector bool short vec_srl (vector bool short, vector unsigned int);
9608 vector bool short vec_srl (vector bool short, vector unsigned short);
9609 vector bool short vec_srl (vector bool short, vector unsigned char);
9610 vector pixel vec_srl (vector pixel, vector unsigned int);
9611 vector pixel vec_srl (vector pixel, vector unsigned short);
9612 vector pixel vec_srl (vector pixel, vector unsigned char);
9613 vector signed char vec_srl (vector signed char, vector unsigned int);
9614 vector signed char vec_srl (vector signed char, vector unsigned short);
9615 vector signed char vec_srl (vector signed char, vector unsigned char);
9616 vector unsigned char vec_srl (vector unsigned char,
9617 vector unsigned int);
9618 vector unsigned char vec_srl (vector unsigned char,
9619 vector unsigned short);
9620 vector unsigned char vec_srl (vector unsigned char,
9621 vector unsigned char);
9622 vector bool char vec_srl (vector bool char, vector unsigned int);
9623 vector bool char vec_srl (vector bool char, vector unsigned short);
9624 vector bool char vec_srl (vector bool char, vector unsigned char);
9626 vector float vec_sro (vector float, vector signed char);
9627 vector float vec_sro (vector float, vector unsigned char);
9628 vector signed int vec_sro (vector signed int, vector signed char);
9629 vector signed int vec_sro (vector signed int, vector unsigned char);
9630 vector unsigned int vec_sro (vector unsigned int, vector signed char);
9631 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
9632 vector signed short vec_sro (vector signed short, vector signed char);
9633 vector signed short vec_sro (vector signed short, vector unsigned char);
9634 vector unsigned short vec_sro (vector unsigned short,
9635 vector signed char);
9636 vector unsigned short vec_sro (vector unsigned short,
9637 vector unsigned char);
9638 vector pixel vec_sro (vector pixel, vector signed char);
9639 vector pixel vec_sro (vector pixel, vector unsigned char);
9640 vector signed char vec_sro (vector signed char, vector signed char);
9641 vector signed char vec_sro (vector signed char, vector unsigned char);
9642 vector unsigned char vec_sro (vector unsigned char, vector signed char);
9643 vector unsigned char vec_sro (vector unsigned char,
9644 vector unsigned char);
9646 void vec_st (vector float, int, vector float *);
9647 void vec_st (vector float, int, float *);
9648 void vec_st (vector signed int, int, vector signed int *);
9649 void vec_st (vector signed int, int, int *);
9650 void vec_st (vector unsigned int, int, vector unsigned int *);
9651 void vec_st (vector unsigned int, int, unsigned int *);
9652 void vec_st (vector bool int, int, vector bool int *);
9653 void vec_st (vector bool int, int, unsigned int *);
9654 void vec_st (vector bool int, int, int *);
9655 void vec_st (vector signed short, int, vector signed short *);
9656 void vec_st (vector signed short, int, short *);
9657 void vec_st (vector unsigned short, int, vector unsigned short *);
9658 void vec_st (vector unsigned short, int, unsigned short *);
9659 void vec_st (vector bool short, int, vector bool short *);
9660 void vec_st (vector bool short, int, unsigned short *);
9661 void vec_st (vector pixel, int, vector pixel *);
9662 void vec_st (vector pixel, int, unsigned short *);
9663 void vec_st (vector pixel, int, short *);
9664 void vec_st (vector bool short, int, short *);
9665 void vec_st (vector signed char, int, vector signed char *);
9666 void vec_st (vector signed char, int, signed char *);
9667 void vec_st (vector unsigned char, int, vector unsigned char *);
9668 void vec_st (vector unsigned char, int, unsigned char *);
9669 void vec_st (vector bool char, int, vector bool char *);
9670 void vec_st (vector bool char, int, unsigned char *);
9671 void vec_st (vector bool char, int, signed char *);
9673 void vec_ste (vector signed char, int, signed char *);
9674 void vec_ste (vector unsigned char, int, unsigned char *);
9675 void vec_ste (vector bool char, int, signed char *);
9676 void vec_ste (vector bool char, int, unsigned char *);
9677 void vec_ste (vector signed short, int, short *);
9678 void vec_ste (vector unsigned short, int, unsigned short *);
9679 void vec_ste (vector bool short, int, short *);
9680 void vec_ste (vector bool short, int, unsigned short *);
9681 void vec_ste (vector pixel, int, short *);
9682 void vec_ste (vector pixel, int, unsigned short *);
9683 void vec_ste (vector float, int, float *);
9684 void vec_ste (vector signed int, int, int *);
9685 void vec_ste (vector unsigned int, int, unsigned int *);
9686 void vec_ste (vector bool int, int, int *);
9687 void vec_ste (vector bool int, int, unsigned int *);
9689 void vec_stvewx (vector float, int, float *);
9690 void vec_stvewx (vector signed int, int, int *);
9691 void vec_stvewx (vector unsigned int, int, unsigned int *);
9692 void vec_stvewx (vector bool int, int, int *);
9693 void vec_stvewx (vector bool int, int, unsigned int *);
9695 void vec_stvehx (vector signed short, int, short *);
9696 void vec_stvehx (vector unsigned short, int, unsigned short *);
9697 void vec_stvehx (vector bool short, int, short *);
9698 void vec_stvehx (vector bool short, int, unsigned short *);
9699 void vec_stvehx (vector pixel, int, short *);
9700 void vec_stvehx (vector pixel, int, unsigned short *);
9702 void vec_stvebx (vector signed char, int, signed char *);
9703 void vec_stvebx (vector unsigned char, int, unsigned char *);
9704 void vec_stvebx (vector bool char, int, signed char *);
9705 void vec_stvebx (vector bool char, int, unsigned char *);
9707 void vec_stl (vector float, int, vector float *);
9708 void vec_stl (vector float, int, float *);
9709 void vec_stl (vector signed int, int, vector signed int *);
9710 void vec_stl (vector signed int, int, int *);
9711 void vec_stl (vector unsigned int, int, vector unsigned int *);
9712 void vec_stl (vector unsigned int, int, unsigned int *);
9713 void vec_stl (vector bool int, int, vector bool int *);
9714 void vec_stl (vector bool int, int, unsigned int *);
9715 void vec_stl (vector bool int, int, int *);
9716 void vec_stl (vector signed short, int, vector signed short *);
9717 void vec_stl (vector signed short, int, short *);
9718 void vec_stl (vector unsigned short, int, vector unsigned short *);
9719 void vec_stl (vector unsigned short, int, unsigned short *);
9720 void vec_stl (vector bool short, int, vector bool short *);
9721 void vec_stl (vector bool short, int, unsigned short *);
9722 void vec_stl (vector bool short, int, short *);
9723 void vec_stl (vector pixel, int, vector pixel *);
9724 void vec_stl (vector pixel, int, unsigned short *);
9725 void vec_stl (vector pixel, int, short *);
9726 void vec_stl (vector signed char, int, vector signed char *);
9727 void vec_stl (vector signed char, int, signed char *);
9728 void vec_stl (vector unsigned char, int, vector unsigned char *);
9729 void vec_stl (vector unsigned char, int, unsigned char *);
9730 void vec_stl (vector bool char, int, vector bool char *);
9731 void vec_stl (vector bool char, int, unsigned char *);
9732 void vec_stl (vector bool char, int, signed char *);
9734 vector signed char vec_sub (vector bool char, vector signed char);
9735 vector signed char vec_sub (vector signed char, vector bool char);
9736 vector signed char vec_sub (vector signed char, vector signed char);
9737 vector unsigned char vec_sub (vector bool char, vector unsigned char);
9738 vector unsigned char vec_sub (vector unsigned char, vector bool char);
9739 vector unsigned char vec_sub (vector unsigned char,
9740 vector unsigned char);
9741 vector signed short vec_sub (vector bool short, vector signed short);
9742 vector signed short vec_sub (vector signed short, vector bool short);
9743 vector signed short vec_sub (vector signed short, vector signed short);
9744 vector unsigned short vec_sub (vector bool short,
9745 vector unsigned short);
9746 vector unsigned short vec_sub (vector unsigned short,
9748 vector unsigned short vec_sub (vector unsigned short,
9749 vector unsigned short);
9750 vector signed int vec_sub (vector bool int, vector signed int);
9751 vector signed int vec_sub (vector signed int, vector bool int);
9752 vector signed int vec_sub (vector signed int, vector signed int);
9753 vector unsigned int vec_sub (vector bool int, vector unsigned int);
9754 vector unsigned int vec_sub (vector unsigned int, vector bool int);
9755 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
9756 vector float vec_sub (vector float, vector float);
9758 vector float vec_vsubfp (vector float, vector float);
9760 vector signed int vec_vsubuwm (vector bool int, vector signed int);
9761 vector signed int vec_vsubuwm (vector signed int, vector bool int);
9762 vector signed int vec_vsubuwm (vector signed int, vector signed int);
9763 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
9764 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
9765 vector unsigned int vec_vsubuwm (vector unsigned int,
9766 vector unsigned int);
9768 vector signed short vec_vsubuhm (vector bool short,
9769 vector signed short);
9770 vector signed short vec_vsubuhm (vector signed short,
9772 vector signed short vec_vsubuhm (vector signed short,
9773 vector signed short);
9774 vector unsigned short vec_vsubuhm (vector bool short,
9775 vector unsigned short);
9776 vector unsigned short vec_vsubuhm (vector unsigned short,
9778 vector unsigned short vec_vsubuhm (vector unsigned short,
9779 vector unsigned short);
9781 vector signed char vec_vsububm (vector bool char, vector signed char);
9782 vector signed char vec_vsububm (vector signed char, vector bool char);
9783 vector signed char vec_vsububm (vector signed char, vector signed char);
9784 vector unsigned char vec_vsububm (vector bool char,
9785 vector unsigned char);
9786 vector unsigned char vec_vsububm (vector unsigned char,
9788 vector unsigned char vec_vsububm (vector unsigned char,
9789 vector unsigned char);
9791 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
9793 vector unsigned char vec_subs (vector bool char, vector unsigned char);
9794 vector unsigned char vec_subs (vector unsigned char, vector bool char);
9795 vector unsigned char vec_subs (vector unsigned char,
9796 vector unsigned char);
9797 vector signed char vec_subs (vector bool char, vector signed char);
9798 vector signed char vec_subs (vector signed char, vector bool char);
9799 vector signed char vec_subs (vector signed char, vector signed char);
9800 vector unsigned short vec_subs (vector bool short,
9801 vector unsigned short);
9802 vector unsigned short vec_subs (vector unsigned short,
9804 vector unsigned short vec_subs (vector unsigned short,
9805 vector unsigned short);
9806 vector signed short vec_subs (vector bool short, vector signed short);
9807 vector signed short vec_subs (vector signed short, vector bool short);
9808 vector signed short vec_subs (vector signed short, vector signed short);
9809 vector unsigned int vec_subs (vector bool int, vector unsigned int);
9810 vector unsigned int vec_subs (vector unsigned int, vector bool int);
9811 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
9812 vector signed int vec_subs (vector bool int, vector signed int);
9813 vector signed int vec_subs (vector signed int, vector bool int);
9814 vector signed int vec_subs (vector signed int, vector signed int);
9816 vector signed int vec_vsubsws (vector bool int, vector signed int);
9817 vector signed int vec_vsubsws (vector signed int, vector bool int);
9818 vector signed int vec_vsubsws (vector signed int, vector signed int);
9820 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
9821 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
9822 vector unsigned int vec_vsubuws (vector unsigned int,
9823 vector unsigned int);
9825 vector signed short vec_vsubshs (vector bool short,
9826 vector signed short);
9827 vector signed short vec_vsubshs (vector signed short,
9829 vector signed short vec_vsubshs (vector signed short,
9830 vector signed short);
9832 vector unsigned short vec_vsubuhs (vector bool short,
9833 vector unsigned short);
9834 vector unsigned short vec_vsubuhs (vector unsigned short,
9836 vector unsigned short vec_vsubuhs (vector unsigned short,
9837 vector unsigned short);
9839 vector signed char vec_vsubsbs (vector bool char, vector signed char);
9840 vector signed char vec_vsubsbs (vector signed char, vector bool char);
9841 vector signed char vec_vsubsbs (vector signed char, vector signed char);
9843 vector unsigned char vec_vsububs (vector bool char,
9844 vector unsigned char);
9845 vector unsigned char vec_vsububs (vector unsigned char,
9847 vector unsigned char vec_vsububs (vector unsigned char,
9848 vector unsigned char);
9850 vector unsigned int vec_sum4s (vector unsigned char,
9851 vector unsigned int);
9852 vector signed int vec_sum4s (vector signed char, vector signed int);
9853 vector signed int vec_sum4s (vector signed short, vector signed int);
9855 vector signed int vec_vsum4shs (vector signed short, vector signed int);
9857 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
9859 vector unsigned int vec_vsum4ubs (vector unsigned char,
9860 vector unsigned int);
9862 vector signed int vec_sum2s (vector signed int, vector signed int);
9864 vector signed int vec_sums (vector signed int, vector signed int);
9866 vector float vec_trunc (vector float);
9868 vector signed short vec_unpackh (vector signed char);
9869 vector bool short vec_unpackh (vector bool char);
9870 vector signed int vec_unpackh (vector signed short);
9871 vector bool int vec_unpackh (vector bool short);
9872 vector unsigned int vec_unpackh (vector pixel);
9874 vector bool int vec_vupkhsh (vector bool short);
9875 vector signed int vec_vupkhsh (vector signed short);
9877 vector unsigned int vec_vupkhpx (vector pixel);
9879 vector bool short vec_vupkhsb (vector bool char);
9880 vector signed short vec_vupkhsb (vector signed char);
9882 vector signed short vec_unpackl (vector signed char);
9883 vector bool short vec_unpackl (vector bool char);
9884 vector unsigned int vec_unpackl (vector pixel);
9885 vector signed int vec_unpackl (vector signed short);
9886 vector bool int vec_unpackl (vector bool short);
9888 vector unsigned int vec_vupklpx (vector pixel);
9890 vector bool int vec_vupklsh (vector bool short);
9891 vector signed int vec_vupklsh (vector signed short);
9893 vector bool short vec_vupklsb (vector bool char);
9894 vector signed short vec_vupklsb (vector signed char);
9896 vector float vec_xor (vector float, vector float);
9897 vector float vec_xor (vector float, vector bool int);
9898 vector float vec_xor (vector bool int, vector float);
9899 vector bool int vec_xor (vector bool int, vector bool int);
9900 vector signed int vec_xor (vector bool int, vector signed int);
9901 vector signed int vec_xor (vector signed int, vector bool int);
9902 vector signed int vec_xor (vector signed int, vector signed int);
9903 vector unsigned int vec_xor (vector bool int, vector unsigned int);
9904 vector unsigned int vec_xor (vector unsigned int, vector bool int);
9905 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
9906 vector bool short vec_xor (vector bool short, vector bool short);
9907 vector signed short vec_xor (vector bool short, vector signed short);
9908 vector signed short vec_xor (vector signed short, vector bool short);
9909 vector signed short vec_xor (vector signed short, vector signed short);
9910 vector unsigned short vec_xor (vector bool short,
9911 vector unsigned short);
9912 vector unsigned short vec_xor (vector unsigned short,
9914 vector unsigned short vec_xor (vector unsigned short,
9915 vector unsigned short);
9916 vector signed char vec_xor (vector bool char, vector signed char);
9917 vector bool char vec_xor (vector bool char, vector bool char);
9918 vector signed char vec_xor (vector signed char, vector bool char);
9919 vector signed char vec_xor (vector signed char, vector signed char);
9920 vector unsigned char vec_xor (vector bool char, vector unsigned char);
9921 vector unsigned char vec_xor (vector unsigned char, vector bool char);
9922 vector unsigned char vec_xor (vector unsigned char,
9923 vector unsigned char);
9925 int vec_all_eq (vector signed char, vector bool char);
9926 int vec_all_eq (vector signed char, vector signed char);
9927 int vec_all_eq (vector unsigned char, vector bool char);
9928 int vec_all_eq (vector unsigned char, vector unsigned char);
9929 int vec_all_eq (vector bool char, vector bool char);
9930 int vec_all_eq (vector bool char, vector unsigned char);
9931 int vec_all_eq (vector bool char, vector signed char);
9932 int vec_all_eq (vector signed short, vector bool short);
9933 int vec_all_eq (vector signed short, vector signed short);
9934 int vec_all_eq (vector unsigned short, vector bool short);
9935 int vec_all_eq (vector unsigned short, vector unsigned short);
9936 int vec_all_eq (vector bool short, vector bool short);
9937 int vec_all_eq (vector bool short, vector unsigned short);
9938 int vec_all_eq (vector bool short, vector signed short);
9939 int vec_all_eq (vector pixel, vector pixel);
9940 int vec_all_eq (vector signed int, vector bool int);
9941 int vec_all_eq (vector signed int, vector signed int);
9942 int vec_all_eq (vector unsigned int, vector bool int);
9943 int vec_all_eq (vector unsigned int, vector unsigned int);
9944 int vec_all_eq (vector bool int, vector bool int);
9945 int vec_all_eq (vector bool int, vector unsigned int);
9946 int vec_all_eq (vector bool int, vector signed int);
9947 int vec_all_eq (vector float, vector float);
9949 int vec_all_ge (vector bool char, vector unsigned char);
9950 int vec_all_ge (vector unsigned char, vector bool char);
9951 int vec_all_ge (vector unsigned char, vector unsigned char);
9952 int vec_all_ge (vector bool char, vector signed char);
9953 int vec_all_ge (vector signed char, vector bool char);
9954 int vec_all_ge (vector signed char, vector signed char);
9955 int vec_all_ge (vector bool short, vector unsigned short);
9956 int vec_all_ge (vector unsigned short, vector bool short);
9957 int vec_all_ge (vector unsigned short, vector unsigned short);
9958 int vec_all_ge (vector signed short, vector signed short);
9959 int vec_all_ge (vector bool short, vector signed short);
9960 int vec_all_ge (vector signed short, vector bool short);
9961 int vec_all_ge (vector bool int, vector unsigned int);
9962 int vec_all_ge (vector unsigned int, vector bool int);
9963 int vec_all_ge (vector unsigned int, vector unsigned int);
9964 int vec_all_ge (vector bool int, vector signed int);
9965 int vec_all_ge (vector signed int, vector bool int);
9966 int vec_all_ge (vector signed int, vector signed int);
9967 int vec_all_ge (vector float, vector float);
9969 int vec_all_gt (vector bool char, vector unsigned char);
9970 int vec_all_gt (vector unsigned char, vector bool char);
9971 int vec_all_gt (vector unsigned char, vector unsigned char);
9972 int vec_all_gt (vector bool char, vector signed char);
9973 int vec_all_gt (vector signed char, vector bool char);
9974 int vec_all_gt (vector signed char, vector signed char);
9975 int vec_all_gt (vector bool short, vector unsigned short);
9976 int vec_all_gt (vector unsigned short, vector bool short);
9977 int vec_all_gt (vector unsigned short, vector unsigned short);
9978 int vec_all_gt (vector bool short, vector signed short);
9979 int vec_all_gt (vector signed short, vector bool short);
9980 int vec_all_gt (vector signed short, vector signed short);
9981 int vec_all_gt (vector bool int, vector unsigned int);
9982 int vec_all_gt (vector unsigned int, vector bool int);
9983 int vec_all_gt (vector unsigned int, vector unsigned int);
9984 int vec_all_gt (vector bool int, vector signed int);
9985 int vec_all_gt (vector signed int, vector bool int);
9986 int vec_all_gt (vector signed int, vector signed int);
9987 int vec_all_gt (vector float, vector float);
9989 int vec_all_in (vector float, vector float);
9991 int vec_all_le (vector bool char, vector unsigned char);
9992 int vec_all_le (vector unsigned char, vector bool char);
9993 int vec_all_le (vector unsigned char, vector unsigned char);
9994 int vec_all_le (vector bool char, vector signed char);
9995 int vec_all_le (vector signed char, vector bool char);
9996 int vec_all_le (vector signed char, vector signed char);
9997 int vec_all_le (vector bool short, vector unsigned short);
9998 int vec_all_le (vector unsigned short, vector bool short);
9999 int vec_all_le (vector unsigned short, vector unsigned short);
10000 int vec_all_le (vector bool short, vector signed short);
10001 int vec_all_le (vector signed short, vector bool short);
10002 int vec_all_le (vector signed short, vector signed short);
10003 int vec_all_le (vector bool int, vector unsigned int);
10004 int vec_all_le (vector unsigned int, vector bool int);
10005 int vec_all_le (vector unsigned int, vector unsigned int);
10006 int vec_all_le (vector bool int, vector signed int);
10007 int vec_all_le (vector signed int, vector bool int);
10008 int vec_all_le (vector signed int, vector signed int);
10009 int vec_all_le (vector float, vector float);
10011 int vec_all_lt (vector bool char, vector unsigned char);
10012 int vec_all_lt (vector unsigned char, vector bool char);
10013 int vec_all_lt (vector unsigned char, vector unsigned char);
10014 int vec_all_lt (vector bool char, vector signed char);
10015 int vec_all_lt (vector signed char, vector bool char);
10016 int vec_all_lt (vector signed char, vector signed char);
10017 int vec_all_lt (vector bool short, vector unsigned short);
10018 int vec_all_lt (vector unsigned short, vector bool short);
10019 int vec_all_lt (vector unsigned short, vector unsigned short);
10020 int vec_all_lt (vector bool short, vector signed short);
10021 int vec_all_lt (vector signed short, vector bool short);
10022 int vec_all_lt (vector signed short, vector signed short);
10023 int vec_all_lt (vector bool int, vector unsigned int);
10024 int vec_all_lt (vector unsigned int, vector bool int);
10025 int vec_all_lt (vector unsigned int, vector unsigned int);
10026 int vec_all_lt (vector bool int, vector signed int);
10027 int vec_all_lt (vector signed int, vector bool int);
10028 int vec_all_lt (vector signed int, vector signed int);
10029 int vec_all_lt (vector float, vector float);
10031 int vec_all_nan (vector float);
10033 int vec_all_ne (vector signed char, vector bool char);
10034 int vec_all_ne (vector signed char, vector signed char);
10035 int vec_all_ne (vector unsigned char, vector bool char);
10036 int vec_all_ne (vector unsigned char, vector unsigned char);
10037 int vec_all_ne (vector bool char, vector bool char);
10038 int vec_all_ne (vector bool char, vector unsigned char);
10039 int vec_all_ne (vector bool char, vector signed char);
10040 int vec_all_ne (vector signed short, vector bool short);
10041 int vec_all_ne (vector signed short, vector signed short);
10042 int vec_all_ne (vector unsigned short, vector bool short);
10043 int vec_all_ne (vector unsigned short, vector unsigned short);
10044 int vec_all_ne (vector bool short, vector bool short);
10045 int vec_all_ne (vector bool short, vector unsigned short);
10046 int vec_all_ne (vector bool short, vector signed short);
10047 int vec_all_ne (vector pixel, vector pixel);
10048 int vec_all_ne (vector signed int, vector bool int);
10049 int vec_all_ne (vector signed int, vector signed int);
10050 int vec_all_ne (vector unsigned int, vector bool int);
10051 int vec_all_ne (vector unsigned int, vector unsigned int);
10052 int vec_all_ne (vector bool int, vector bool int);
10053 int vec_all_ne (vector bool int, vector unsigned int);
10054 int vec_all_ne (vector bool int, vector signed int);
10055 int vec_all_ne (vector float, vector float);
10057 int vec_all_nge (vector float, vector float);
10059 int vec_all_ngt (vector float, vector float);
10061 int vec_all_nle (vector float, vector float);
10063 int vec_all_nlt (vector float, vector float);
10065 int vec_all_numeric (vector float);
10067 int vec_any_eq (vector signed char, vector bool char);
10068 int vec_any_eq (vector signed char, vector signed char);
10069 int vec_any_eq (vector unsigned char, vector bool char);
10070 int vec_any_eq (vector unsigned char, vector unsigned char);
10071 int vec_any_eq (vector bool char, vector bool char);
10072 int vec_any_eq (vector bool char, vector unsigned char);
10073 int vec_any_eq (vector bool char, vector signed char);
10074 int vec_any_eq (vector signed short, vector bool short);
10075 int vec_any_eq (vector signed short, vector signed short);
10076 int vec_any_eq (vector unsigned short, vector bool short);
10077 int vec_any_eq (vector unsigned short, vector unsigned short);
10078 int vec_any_eq (vector bool short, vector bool short);
10079 int vec_any_eq (vector bool short, vector unsigned short);
10080 int vec_any_eq (vector bool short, vector signed short);
10081 int vec_any_eq (vector pixel, vector pixel);
10082 int vec_any_eq (vector signed int, vector bool int);
10083 int vec_any_eq (vector signed int, vector signed int);
10084 int vec_any_eq (vector unsigned int, vector bool int);
10085 int vec_any_eq (vector unsigned int, vector unsigned int);
10086 int vec_any_eq (vector bool int, vector bool int);
10087 int vec_any_eq (vector bool int, vector unsigned int);
10088 int vec_any_eq (vector bool int, vector signed int);
10089 int vec_any_eq (vector float, vector float);
10091 int vec_any_ge (vector signed char, vector bool char);
10092 int vec_any_ge (vector unsigned char, vector bool char);
10093 int vec_any_ge (vector unsigned char, vector unsigned char);
10094 int vec_any_ge (vector signed char, vector signed char);
10095 int vec_any_ge (vector bool char, vector unsigned char);
10096 int vec_any_ge (vector bool char, vector signed char);
10097 int vec_any_ge (vector unsigned short, vector bool short);
10098 int vec_any_ge (vector unsigned short, vector unsigned short);
10099 int vec_any_ge (vector signed short, vector signed short);
10100 int vec_any_ge (vector signed short, vector bool short);
10101 int vec_any_ge (vector bool short, vector unsigned short);
10102 int vec_any_ge (vector bool short, vector signed short);
10103 int vec_any_ge (vector signed int, vector bool int);
10104 int vec_any_ge (vector unsigned int, vector bool int);
10105 int vec_any_ge (vector unsigned int, vector unsigned int);
10106 int vec_any_ge (vector signed int, vector signed int);
10107 int vec_any_ge (vector bool int, vector unsigned int);
10108 int vec_any_ge (vector bool int, vector signed int);
10109 int vec_any_ge (vector float, vector float);
10111 int vec_any_gt (vector bool char, vector unsigned char);
10112 int vec_any_gt (vector unsigned char, vector bool char);
10113 int vec_any_gt (vector unsigned char, vector unsigned char);
10114 int vec_any_gt (vector bool char, vector signed char);
10115 int vec_any_gt (vector signed char, vector bool char);
10116 int vec_any_gt (vector signed char, vector signed char);
10117 int vec_any_gt (vector bool short, vector unsigned short);
10118 int vec_any_gt (vector unsigned short, vector bool short);
10119 int vec_any_gt (vector unsigned short, vector unsigned short);
10120 int vec_any_gt (vector bool short, vector signed short);
10121 int vec_any_gt (vector signed short, vector bool short);
10122 int vec_any_gt (vector signed short, vector signed short);
10123 int vec_any_gt (vector bool int, vector unsigned int);
10124 int vec_any_gt (vector unsigned int, vector bool int);
10125 int vec_any_gt (vector unsigned int, vector unsigned int);
10126 int vec_any_gt (vector bool int, vector signed int);
10127 int vec_any_gt (vector signed int, vector bool int);
10128 int vec_any_gt (vector signed int, vector signed int);
10129 int vec_any_gt (vector float, vector float);
10131 int vec_any_le (vector bool char, vector unsigned char);
10132 int vec_any_le (vector unsigned char, vector bool char);
10133 int vec_any_le (vector unsigned char, vector unsigned char);
10134 int vec_any_le (vector bool char, vector signed char);
10135 int vec_any_le (vector signed char, vector bool char);
10136 int vec_any_le (vector signed char, vector signed char);
10137 int vec_any_le (vector bool short, vector unsigned short);
10138 int vec_any_le (vector unsigned short, vector bool short);
10139 int vec_any_le (vector unsigned short, vector unsigned short);
10140 int vec_any_le (vector bool short, vector signed short);
10141 int vec_any_le (vector signed short, vector bool short);
10142 int vec_any_le (vector signed short, vector signed short);
10143 int vec_any_le (vector bool int, vector unsigned int);
10144 int vec_any_le (vector unsigned int, vector bool int);
10145 int vec_any_le (vector unsigned int, vector unsigned int);
10146 int vec_any_le (vector bool int, vector signed int);
10147 int vec_any_le (vector signed int, vector bool int);
10148 int vec_any_le (vector signed int, vector signed int);
10149 int vec_any_le (vector float, vector float);
10151 int vec_any_lt (vector bool char, vector unsigned char);
10152 int vec_any_lt (vector unsigned char, vector bool char);
10153 int vec_any_lt (vector unsigned char, vector unsigned char);
10154 int vec_any_lt (vector bool char, vector signed char);
10155 int vec_any_lt (vector signed char, vector bool char);
10156 int vec_any_lt (vector signed char, vector signed char);
10157 int vec_any_lt (vector bool short, vector unsigned short);
10158 int vec_any_lt (vector unsigned short, vector bool short);
10159 int vec_any_lt (vector unsigned short, vector unsigned short);
10160 int vec_any_lt (vector bool short, vector signed short);
10161 int vec_any_lt (vector signed short, vector bool short);
10162 int vec_any_lt (vector signed short, vector signed short);
10163 int vec_any_lt (vector bool int, vector unsigned int);
10164 int vec_any_lt (vector unsigned int, vector bool int);
10165 int vec_any_lt (vector unsigned int, vector unsigned int);
10166 int vec_any_lt (vector bool int, vector signed int);
10167 int vec_any_lt (vector signed int, vector bool int);
10168 int vec_any_lt (vector signed int, vector signed int);
10169 int vec_any_lt (vector float, vector float);
10171 int vec_any_nan (vector float);
10173 int vec_any_ne (vector signed char, vector bool char);
10174 int vec_any_ne (vector signed char, vector signed char);
10175 int vec_any_ne (vector unsigned char, vector bool char);
10176 int vec_any_ne (vector unsigned char, vector unsigned char);
10177 int vec_any_ne (vector bool char, vector bool char);
10178 int vec_any_ne (vector bool char, vector unsigned char);
10179 int vec_any_ne (vector bool char, vector signed char);
10180 int vec_any_ne (vector signed short, vector bool short);
10181 int vec_any_ne (vector signed short, vector signed short);
10182 int vec_any_ne (vector unsigned short, vector bool short);
10183 int vec_any_ne (vector unsigned short, vector unsigned short);
10184 int vec_any_ne (vector bool short, vector bool short);
10185 int vec_any_ne (vector bool short, vector unsigned short);
10186 int vec_any_ne (vector bool short, vector signed short);
10187 int vec_any_ne (vector pixel, vector pixel);
10188 int vec_any_ne (vector signed int, vector bool int);
10189 int vec_any_ne (vector signed int, vector signed int);
10190 int vec_any_ne (vector unsigned int, vector bool int);
10191 int vec_any_ne (vector unsigned int, vector unsigned int);
10192 int vec_any_ne (vector bool int, vector bool int);
10193 int vec_any_ne (vector bool int, vector unsigned int);
10194 int vec_any_ne (vector bool int, vector signed int);
10195 int vec_any_ne (vector float, vector float);
10197 int vec_any_nge (vector float, vector float);
10199 int vec_any_ngt (vector float, vector float);
10201 int vec_any_nle (vector float, vector float);
10203 int vec_any_nlt (vector float, vector float);
10205 int vec_any_numeric (vector float);
10207 int vec_any_out (vector float, vector float);
10210 @node SPARC VIS Built-in Functions
10211 @subsection SPARC VIS Built-in Functions
10213 GCC supports SIMD operations on the SPARC using both the generic vector
10214 extensions (@pxref{Vector Extensions}) as well as built-in functions for
10215 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
10216 switch, the VIS extension is exposed as the following built-in functions:
10219 typedef int v2si __attribute__ ((vector_size (8)));
10220 typedef short v4hi __attribute__ ((vector_size (8)));
10221 typedef short v2hi __attribute__ ((vector_size (4)));
10222 typedef char v8qi __attribute__ ((vector_size (8)));
10223 typedef char v4qi __attribute__ ((vector_size (4)));
10225 void * __builtin_vis_alignaddr (void *, long);
10226 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
10227 v2si __builtin_vis_faligndatav2si (v2si, v2si);
10228 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
10229 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
10231 v4hi __builtin_vis_fexpand (v4qi);
10233 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
10234 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
10235 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
10236 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
10237 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
10238 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
10239 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
10241 v4qi __builtin_vis_fpack16 (v4hi);
10242 v8qi __builtin_vis_fpack32 (v2si, v2si);
10243 v2hi __builtin_vis_fpackfix (v2si);
10244 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
10246 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
10249 @node SPU Built-in Functions
10250 @subsection SPU Built-in Functions
10252 GCC provides extensions for the SPU processor as described in the
10253 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
10254 found at @uref{http://cell.scei.co.jp/} or
10255 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
10256 implementation differs in several ways.
10261 The optional extension of specifying vector constants in parentheses is
10265 A vector initializer requires no cast if the vector constant is of the
10266 same type as the variable it is initializing.
10269 If @code{signed} or @code{unsigned} is omitted, the signedness of the
10270 vector type is the default signedness of the base type. The default
10271 varies depending on the operating system, so a portable program should
10272 always specify the signedness.
10275 By default, the keyword @code{__vector} is added. The macro
10276 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
10280 GCC allows using a @code{typedef} name as the type specifier for a
10284 For C, overloaded functions are implemented with macros so the following
10288 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
10291 Since @code{spu_add} is a macro, the vector constant in the example
10292 is treated as four separate arguments. Wrap the entire argument in
10293 parentheses for this to work.
10296 The extended version of @code{__builtin_expect} is not supported.
10300 @emph{Note:} Only the interface described in the aforementioned
10301 specification is supported. Internally, GCC uses built-in functions to
10302 implement the required functionality, but these are not supported and
10303 are subject to change without notice.
10305 @node Target Format Checks
10306 @section Format Checks Specific to Particular Target Machines
10308 For some target machines, GCC supports additional options to the
10310 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
10313 * Solaris Format Checks::
10316 @node Solaris Format Checks
10317 @subsection Solaris Format Checks
10319 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
10320 check. @code{cmn_err} accepts a subset of the standard @code{printf}
10321 conversions, and the two-argument @code{%b} conversion for displaying
10322 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
10325 @section Pragmas Accepted by GCC
10329 GCC supports several types of pragmas, primarily in order to compile
10330 code originally written for other compilers. Note that in general
10331 we do not recommend the use of pragmas; @xref{Function Attributes},
10332 for further explanation.
10337 * RS/6000 and PowerPC Pragmas::
10339 * Solaris Pragmas::
10340 * Symbol-Renaming Pragmas::
10341 * Structure-Packing Pragmas::
10343 * Diagnostic Pragmas::
10344 * Visibility Pragmas::
10348 @subsection ARM Pragmas
10350 The ARM target defines pragmas for controlling the default addition of
10351 @code{long_call} and @code{short_call} attributes to functions.
10352 @xref{Function Attributes}, for information about the effects of these
10357 @cindex pragma, long_calls
10358 Set all subsequent functions to have the @code{long_call} attribute.
10360 @item no_long_calls
10361 @cindex pragma, no_long_calls
10362 Set all subsequent functions to have the @code{short_call} attribute.
10364 @item long_calls_off
10365 @cindex pragma, long_calls_off
10366 Do not affect the @code{long_call} or @code{short_call} attributes of
10367 subsequent functions.
10371 @subsection M32C Pragmas
10374 @item memregs @var{number}
10375 @cindex pragma, memregs
10376 Overrides the command line option @code{-memregs=} for the current
10377 file. Use with care! This pragma must be before any function in the
10378 file, and mixing different memregs values in different objects may
10379 make them incompatible. This pragma is useful when a
10380 performance-critical function uses a memreg for temporary values,
10381 as it may allow you to reduce the number of memregs used.
10385 @node RS/6000 and PowerPC Pragmas
10386 @subsection RS/6000 and PowerPC Pragmas
10388 The RS/6000 and PowerPC targets define one pragma for controlling
10389 whether or not the @code{longcall} attribute is added to function
10390 declarations by default. This pragma overrides the @option{-mlongcall}
10391 option, but not the @code{longcall} and @code{shortcall} attributes.
10392 @xref{RS/6000 and PowerPC Options}, for more information about when long
10393 calls are and are not necessary.
10397 @cindex pragma, longcall
10398 Apply the @code{longcall} attribute to all subsequent function
10402 Do not apply the @code{longcall} attribute to subsequent function
10406 @c Describe c4x pragmas here.
10407 @c Describe h8300 pragmas here.
10408 @c Describe sh pragmas here.
10409 @c Describe v850 pragmas here.
10411 @node Darwin Pragmas
10412 @subsection Darwin Pragmas
10414 The following pragmas are available for all architectures running the
10415 Darwin operating system. These are useful for compatibility with other
10419 @item mark @var{tokens}@dots{}
10420 @cindex pragma, mark
10421 This pragma is accepted, but has no effect.
10423 @item options align=@var{alignment}
10424 @cindex pragma, options align
10425 This pragma sets the alignment of fields in structures. The values of
10426 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
10427 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
10428 properly; to restore the previous setting, use @code{reset} for the
10431 @item segment @var{tokens}@dots{}
10432 @cindex pragma, segment
10433 This pragma is accepted, but has no effect.
10435 @item unused (@var{var} [, @var{var}]@dots{})
10436 @cindex pragma, unused
10437 This pragma declares variables to be possibly unused. GCC will not
10438 produce warnings for the listed variables. The effect is similar to
10439 that of the @code{unused} attribute, except that this pragma may appear
10440 anywhere within the variables' scopes.
10443 @node Solaris Pragmas
10444 @subsection Solaris Pragmas
10446 The Solaris target supports @code{#pragma redefine_extname}
10447 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
10448 @code{#pragma} directives for compatibility with the system compiler.
10451 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
10452 @cindex pragma, align
10454 Increase the minimum alignment of each @var{variable} to @var{alignment}.
10455 This is the same as GCC's @code{aligned} attribute @pxref{Variable
10456 Attributes}). Macro expansion occurs on the arguments to this pragma
10457 when compiling C and Objective-C. It does not currently occur when
10458 compiling C++, but this is a bug which may be fixed in a future
10461 @item fini (@var{function} [, @var{function}]...)
10462 @cindex pragma, fini
10464 This pragma causes each listed @var{function} to be called after
10465 main, or during shared module unloading, by adding a call to the
10466 @code{.fini} section.
10468 @item init (@var{function} [, @var{function}]...)
10469 @cindex pragma, init
10471 This pragma causes each listed @var{function} to be called during
10472 initialization (before @code{main}) or during shared module loading, by
10473 adding a call to the @code{.init} section.
10477 @node Symbol-Renaming Pragmas
10478 @subsection Symbol-Renaming Pragmas
10480 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
10481 supports two @code{#pragma} directives which change the name used in
10482 assembly for a given declaration. These pragmas are only available on
10483 platforms whose system headers need them. To get this effect on all
10484 platforms supported by GCC, use the asm labels extension (@pxref{Asm
10488 @item redefine_extname @var{oldname} @var{newname}
10489 @cindex pragma, redefine_extname
10491 This pragma gives the C function @var{oldname} the assembly symbol
10492 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
10493 will be defined if this pragma is available (currently only on
10496 @item extern_prefix @var{string}
10497 @cindex pragma, extern_prefix
10499 This pragma causes all subsequent external function and variable
10500 declarations to have @var{string} prepended to their assembly symbols.
10501 This effect may be terminated with another @code{extern_prefix} pragma
10502 whose argument is an empty string. The preprocessor macro
10503 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
10504 available (currently only on Tru64 UNIX)@.
10507 These pragmas and the asm labels extension interact in a complicated
10508 manner. Here are some corner cases you may want to be aware of.
10511 @item Both pragmas silently apply only to declarations with external
10512 linkage. Asm labels do not have this restriction.
10514 @item In C++, both pragmas silently apply only to declarations with
10515 ``C'' linkage. Again, asm labels do not have this restriction.
10517 @item If any of the three ways of changing the assembly name of a
10518 declaration is applied to a declaration whose assembly name has
10519 already been determined (either by a previous use of one of these
10520 features, or because the compiler needed the assembly name in order to
10521 generate code), and the new name is different, a warning issues and
10522 the name does not change.
10524 @item The @var{oldname} used by @code{#pragma redefine_extname} is
10525 always the C-language name.
10527 @item If @code{#pragma extern_prefix} is in effect, and a declaration
10528 occurs with an asm label attached, the prefix is silently ignored for
10531 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
10532 apply to the same declaration, whichever triggered first wins, and a
10533 warning issues if they contradict each other. (We would like to have
10534 @code{#pragma redefine_extname} always win, for consistency with asm
10535 labels, but if @code{#pragma extern_prefix} triggers first we have no
10536 way of knowing that that happened.)
10539 @node Structure-Packing Pragmas
10540 @subsection Structure-Packing Pragmas
10542 For compatibility with Win32, GCC supports a set of @code{#pragma}
10543 directives which change the maximum alignment of members of structures
10544 (other than zero-width bitfields), unions, and classes subsequently
10545 defined. The @var{n} value below always is required to be a small power
10546 of two and specifies the new alignment in bytes.
10549 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
10550 @item @code{#pragma pack()} sets the alignment to the one that was in
10551 effect when compilation started (see also command line option
10552 @option{-fpack-struct[=<n>]} @pxref{Code Gen Options}).
10553 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
10554 setting on an internal stack and then optionally sets the new alignment.
10555 @item @code{#pragma pack(pop)} restores the alignment setting to the one
10556 saved at the top of the internal stack (and removes that stack entry).
10557 Note that @code{#pragma pack([@var{n}])} does not influence this internal
10558 stack; thus it is possible to have @code{#pragma pack(push)} followed by
10559 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
10560 @code{#pragma pack(pop)}.
10563 Some targets, e.g. i386 and powerpc, support the @code{ms_struct}
10564 @code{#pragma} which lays out a structure as the documented
10565 @code{__attribute__ ((ms_struct))}.
10567 @item @code{#pragma ms_struct on} turns on the layout for structures
10569 @item @code{#pragma ms_struct off} turns off the layout for structures
10571 @item @code{#pragma ms_struct reset} goes back to the default layout.
10575 @subsection Weak Pragmas
10577 For compatibility with SVR4, GCC supports a set of @code{#pragma}
10578 directives for declaring symbols to be weak, and defining weak
10582 @item #pragma weak @var{symbol}
10583 @cindex pragma, weak
10584 This pragma declares @var{symbol} to be weak, as if the declaration
10585 had the attribute of the same name. The pragma may appear before
10586 or after the declaration of @var{symbol}, but must appear before
10587 either its first use or its definition. It is not an error for
10588 @var{symbol} to never be defined at all.
10590 @item #pragma weak @var{symbol1} = @var{symbol2}
10591 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
10592 It is an error if @var{symbol2} is not defined in the current
10596 @node Diagnostic Pragmas
10597 @subsection Diagnostic Pragmas
10599 GCC allows the user to selectively enable or disable certain types of
10600 diagnostics, and change the kind of the diagnostic. For example, a
10601 project's policy might require that all sources compile with
10602 @option{-Werror} but certain files might have exceptions allowing
10603 specific types of warnings. Or, a project might selectively enable
10604 diagnostics and treat them as errors depending on which preprocessor
10605 macros are defined.
10608 @item #pragma GCC diagnostic @var{kind} @var{option}
10609 @cindex pragma, diagnostic
10611 Modifies the disposition of a diagnostic. Note that not all
10612 diagnostics are modifiable; at the moment only warnings (normally
10613 controlled by @samp{-W...}) can be controlled, and not all of them.
10614 Use @option{-fdiagnostics-show-option} to determine which diagnostics
10615 are controllable and which option controls them.
10617 @var{kind} is @samp{error} to treat this diagnostic as an error,
10618 @samp{warning} to treat it like a warning (even if @option{-Werror} is
10619 in effect), or @samp{ignored} if the diagnostic is to be ignored.
10620 @var{option} is a double quoted string which matches the command line
10624 #pragma GCC diagnostic warning "-Wformat"
10625 #pragma GCC diagnostic error "-Wformat"
10626 #pragma GCC diagnostic ignored "-Wformat"
10629 Note that these pragmas override any command line options. Also,
10630 while it is syntactically valid to put these pragmas anywhere in your
10631 sources, the only supported location for them is before any data or
10632 functions are defined. Doing otherwise may result in unpredictable
10633 results depending on how the optimizer manages your sources. If the
10634 same option is listed multiple times, the last one specified is the
10635 one that is in effect. This pragma is not intended to be a general
10636 purpose replacement for command line options, but for implementing
10637 strict control over project policies.
10641 @node Visibility Pragmas
10642 @subsection Visibility Pragmas
10645 @item #pragma GCC visibility push(@var{visibility})
10646 @itemx #pragma GCC visibility pop
10647 @cindex pragma, visibility
10649 This pragma allows the user to set the visibility for multiple
10650 declarations without having to give each a visibility attribute
10651 @xref{Function Attributes}, for more information about visibility and
10652 the attribute syntax.
10654 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
10655 declarations. Class members and template specializations are not
10656 affected; if you want to override the visibility for a particular
10657 member or instantiation, you must use an attribute.
10661 @node Unnamed Fields
10662 @section Unnamed struct/union fields within structs/unions
10666 For compatibility with other compilers, GCC allows you to define
10667 a structure or union that contains, as fields, structures and unions
10668 without names. For example:
10681 In this example, the user would be able to access members of the unnamed
10682 union with code like @samp{foo.b}. Note that only unnamed structs and
10683 unions are allowed, you may not have, for example, an unnamed
10686 You must never create such structures that cause ambiguous field definitions.
10687 For example, this structure:
10698 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
10699 Such constructs are not supported and must be avoided. In the future,
10700 such constructs may be detected and treated as compilation errors.
10702 @opindex fms-extensions
10703 Unless @option{-fms-extensions} is used, the unnamed field must be a
10704 structure or union definition without a tag (for example, @samp{struct
10705 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
10706 also be a definition with a tag such as @samp{struct foo @{ int a;
10707 @};}, a reference to a previously defined structure or union such as
10708 @samp{struct foo;}, or a reference to a @code{typedef} name for a
10709 previously defined structure or union type.
10712 @section Thread-Local Storage
10713 @cindex Thread-Local Storage
10714 @cindex @acronym{TLS}
10717 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
10718 are allocated such that there is one instance of the variable per extant
10719 thread. The run-time model GCC uses to implement this originates
10720 in the IA-64 processor-specific ABI, but has since been migrated
10721 to other processors as well. It requires significant support from
10722 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
10723 system libraries (@file{libc.so} and @file{libpthread.so}), so it
10724 is not available everywhere.
10726 At the user level, the extension is visible with a new storage
10727 class keyword: @code{__thread}. For example:
10731 extern __thread struct state s;
10732 static __thread char *p;
10735 The @code{__thread} specifier may be used alone, with the @code{extern}
10736 or @code{static} specifiers, but with no other storage class specifier.
10737 When used with @code{extern} or @code{static}, @code{__thread} must appear
10738 immediately after the other storage class specifier.
10740 The @code{__thread} specifier may be applied to any global, file-scoped
10741 static, function-scoped static, or static data member of a class. It may
10742 not be applied to block-scoped automatic or non-static data member.
10744 When the address-of operator is applied to a thread-local variable, it is
10745 evaluated at run-time and returns the address of the current thread's
10746 instance of that variable. An address so obtained may be used by any
10747 thread. When a thread terminates, any pointers to thread-local variables
10748 in that thread become invalid.
10750 No static initialization may refer to the address of a thread-local variable.
10752 In C++, if an initializer is present for a thread-local variable, it must
10753 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
10756 See @uref{http://people.redhat.com/drepper/tls.pdf,
10757 ELF Handling For Thread-Local Storage} for a detailed explanation of
10758 the four thread-local storage addressing models, and how the run-time
10759 is expected to function.
10762 * C99 Thread-Local Edits::
10763 * C++98 Thread-Local Edits::
10766 @node C99 Thread-Local Edits
10767 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
10769 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
10770 that document the exact semantics of the language extension.
10774 @cite{5.1.2 Execution environments}
10776 Add new text after paragraph 1
10779 Within either execution environment, a @dfn{thread} is a flow of
10780 control within a program. It is implementation defined whether
10781 or not there may be more than one thread associated with a program.
10782 It is implementation defined how threads beyond the first are
10783 created, the name and type of the function called at thread
10784 startup, and how threads may be terminated. However, objects
10785 with thread storage duration shall be initialized before thread
10790 @cite{6.2.4 Storage durations of objects}
10792 Add new text before paragraph 3
10795 An object whose identifier is declared with the storage-class
10796 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
10797 Its lifetime is the entire execution of the thread, and its
10798 stored value is initialized only once, prior to thread startup.
10802 @cite{6.4.1 Keywords}
10804 Add @code{__thread}.
10807 @cite{6.7.1 Storage-class specifiers}
10809 Add @code{__thread} to the list of storage class specifiers in
10812 Change paragraph 2 to
10815 With the exception of @code{__thread}, at most one storage-class
10816 specifier may be given [@dots{}]. The @code{__thread} specifier may
10817 be used alone, or immediately following @code{extern} or
10821 Add new text after paragraph 6
10824 The declaration of an identifier for a variable that has
10825 block scope that specifies @code{__thread} shall also
10826 specify either @code{extern} or @code{static}.
10828 The @code{__thread} specifier shall be used only with
10833 @node C++98 Thread-Local Edits
10834 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
10836 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
10837 that document the exact semantics of the language extension.
10841 @b{[intro.execution]}
10843 New text after paragraph 4
10846 A @dfn{thread} is a flow of control within the abstract machine.
10847 It is implementation defined whether or not there may be more than
10851 New text after paragraph 7
10854 It is unspecified whether additional action must be taken to
10855 ensure when and whether side effects are visible to other threads.
10861 Add @code{__thread}.
10864 @b{[basic.start.main]}
10866 Add after paragraph 5
10869 The thread that begins execution at the @code{main} function is called
10870 the @dfn{main thread}. It is implementation defined how functions
10871 beginning threads other than the main thread are designated or typed.
10872 A function so designated, as well as the @code{main} function, is called
10873 a @dfn{thread startup function}. It is implementation defined what
10874 happens if a thread startup function returns. It is implementation
10875 defined what happens to other threads when any thread calls @code{exit}.
10879 @b{[basic.start.init]}
10881 Add after paragraph 4
10884 The storage for an object of thread storage duration shall be
10885 statically initialized before the first statement of the thread startup
10886 function. An object of thread storage duration shall not require
10887 dynamic initialization.
10891 @b{[basic.start.term]}
10893 Add after paragraph 3
10896 The type of an object with thread storage duration shall not have a
10897 non-trivial destructor, nor shall it be an array type whose elements
10898 (directly or indirectly) have non-trivial destructors.
10904 Add ``thread storage duration'' to the list in paragraph 1.
10909 Thread, static, and automatic storage durations are associated with
10910 objects introduced by declarations [@dots{}].
10913 Add @code{__thread} to the list of specifiers in paragraph 3.
10916 @b{[basic.stc.thread]}
10918 New section before @b{[basic.stc.static]}
10921 The keyword @code{__thread} applied to a non-local object gives the
10922 object thread storage duration.
10924 A local variable or class data member declared both @code{static}
10925 and @code{__thread} gives the variable or member thread storage
10930 @b{[basic.stc.static]}
10935 All objects which have neither thread storage duration, dynamic
10936 storage duration nor are local [@dots{}].
10942 Add @code{__thread} to the list in paragraph 1.
10947 With the exception of @code{__thread}, at most one
10948 @var{storage-class-specifier} shall appear in a given
10949 @var{decl-specifier-seq}. The @code{__thread} specifier may
10950 be used alone, or immediately following the @code{extern} or
10951 @code{static} specifiers. [@dots{}]
10954 Add after paragraph 5
10957 The @code{__thread} specifier can be applied only to the names of objects
10958 and to anonymous unions.
10964 Add after paragraph 6
10967 Non-@code{static} members shall not be @code{__thread}.
10971 @node Binary constants
10972 @section Binary constants using the @samp{0b} prefix
10973 @cindex Binary constants using the @samp{0b} prefix
10975 Integer constants can be written as binary constants, consisting of a
10976 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
10977 @samp{0B}. This is particularly useful in environments that operate a
10978 lot on the bit-level (like microcontrollers).
10980 The following statements are identical:
10989 The type of these constants follows the same rules as for octal or
10990 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
10993 @node C++ Extensions
10994 @chapter Extensions to the C++ Language
10995 @cindex extensions, C++ language
10996 @cindex C++ language extensions
10998 The GNU compiler provides these extensions to the C++ language (and you
10999 can also use most of the C language extensions in your C++ programs). If you
11000 want to write code that checks whether these features are available, you can
11001 test for the GNU compiler the same way as for C programs: check for a
11002 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
11003 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
11004 Predefined Macros,cpp,The GNU C Preprocessor}).
11007 * Volatiles:: What constitutes an access to a volatile object.
11008 * Restricted Pointers:: C99 restricted pointers and references.
11009 * Vague Linkage:: Where G++ puts inlines, vtables and such.
11010 * C++ Interface:: You can use a single C++ header file for both
11011 declarations and definitions.
11012 * Template Instantiation:: Methods for ensuring that exactly one copy of
11013 each needed template instantiation is emitted.
11014 * Bound member functions:: You can extract a function pointer to the
11015 method denoted by a @samp{->*} or @samp{.*} expression.
11016 * C++ Attributes:: Variable, function, and type attributes for C++ only.
11017 * Namespace Association:: Strong using-directives for namespace association.
11018 * Type Traits:: Compiler support for type traits
11019 * Java Exceptions:: Tweaking exception handling to work with Java.
11020 * Deprecated Features:: Things will disappear from g++.
11021 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
11025 @section When is a Volatile Object Accessed?
11026 @cindex accessing volatiles
11027 @cindex volatile read
11028 @cindex volatile write
11029 @cindex volatile access
11031 Both the C and C++ standard have the concept of volatile objects. These
11032 are normally accessed by pointers and used for accessing hardware. The
11033 standards encourage compilers to refrain from optimizations concerning
11034 accesses to volatile objects. The C standard leaves it implementation
11035 defined as to what constitutes a volatile access. The C++ standard omits
11036 to specify this, except to say that C++ should behave in a similar manner
11037 to C with respect to volatiles, where possible. The minimum either
11038 standard specifies is that at a sequence point all previous accesses to
11039 volatile objects have stabilized and no subsequent accesses have
11040 occurred. Thus an implementation is free to reorder and combine
11041 volatile accesses which occur between sequence points, but cannot do so
11042 for accesses across a sequence point. The use of volatiles does not
11043 allow you to violate the restriction on updating objects multiple times
11044 within a sequence point.
11046 @xref{Qualifiers implementation, , Volatile qualifier and the C compiler}.
11048 The behavior differs slightly between C and C++ in the non-obvious cases:
11051 volatile int *src = @var{somevalue};
11055 With C, such expressions are rvalues, and GCC interprets this either as a
11056 read of the volatile object being pointed to or only as request to evaluate
11057 the side-effects. The C++ standard specifies that such expressions do not
11058 undergo lvalue to rvalue conversion, and that the type of the dereferenced
11059 object may be incomplete. The C++ standard does not specify explicitly
11060 that it is this lvalue to rvalue conversion which may be responsible for
11061 causing an access. However, there is reason to believe that it is,
11062 because otherwise certain simple expressions become undefined. However,
11063 because it would surprise most programmers, G++ treats dereferencing a
11064 pointer to volatile object of complete type when the value is unused as
11065 GCC would do for an equivalent type in C. When the object has incomplete
11066 type, G++ issues a warning; if you wish to force an error, you must
11067 force a conversion to rvalue with, for instance, a static cast.
11069 When using a reference to volatile, G++ does not treat equivalent
11070 expressions as accesses to volatiles, but instead issues a warning that
11071 no volatile is accessed. The rationale for this is that otherwise it
11072 becomes difficult to determine where volatile access occur, and not
11073 possible to ignore the return value from functions returning volatile
11074 references. Again, if you wish to force a read, cast the reference to
11077 @node Restricted Pointers
11078 @section Restricting Pointer Aliasing
11079 @cindex restricted pointers
11080 @cindex restricted references
11081 @cindex restricted this pointer
11083 As with the C front end, G++ understands the C99 feature of restricted pointers,
11084 specified with the @code{__restrict__}, or @code{__restrict} type
11085 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
11086 language flag, @code{restrict} is not a keyword in C++.
11088 In addition to allowing restricted pointers, you can specify restricted
11089 references, which indicate that the reference is not aliased in the local
11093 void fn (int *__restrict__ rptr, int &__restrict__ rref)
11100 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
11101 @var{rref} refers to a (different) unaliased integer.
11103 You may also specify whether a member function's @var{this} pointer is
11104 unaliased by using @code{__restrict__} as a member function qualifier.
11107 void T::fn () __restrict__
11114 Within the body of @code{T::fn}, @var{this} will have the effective
11115 definition @code{T *__restrict__ const this}. Notice that the
11116 interpretation of a @code{__restrict__} member function qualifier is
11117 different to that of @code{const} or @code{volatile} qualifier, in that it
11118 is applied to the pointer rather than the object. This is consistent with
11119 other compilers which implement restricted pointers.
11121 As with all outermost parameter qualifiers, @code{__restrict__} is
11122 ignored in function definition matching. This means you only need to
11123 specify @code{__restrict__} in a function definition, rather than
11124 in a function prototype as well.
11126 @node Vague Linkage
11127 @section Vague Linkage
11128 @cindex vague linkage
11130 There are several constructs in C++ which require space in the object
11131 file but are not clearly tied to a single translation unit. We say that
11132 these constructs have ``vague linkage''. Typically such constructs are
11133 emitted wherever they are needed, though sometimes we can be more
11137 @item Inline Functions
11138 Inline functions are typically defined in a header file which can be
11139 included in many different compilations. Hopefully they can usually be
11140 inlined, but sometimes an out-of-line copy is necessary, if the address
11141 of the function is taken or if inlining fails. In general, we emit an
11142 out-of-line copy in all translation units where one is needed. As an
11143 exception, we only emit inline virtual functions with the vtable, since
11144 it will always require a copy.
11146 Local static variables and string constants used in an inline function
11147 are also considered to have vague linkage, since they must be shared
11148 between all inlined and out-of-line instances of the function.
11152 C++ virtual functions are implemented in most compilers using a lookup
11153 table, known as a vtable. The vtable contains pointers to the virtual
11154 functions provided by a class, and each object of the class contains a
11155 pointer to its vtable (or vtables, in some multiple-inheritance
11156 situations). If the class declares any non-inline, non-pure virtual
11157 functions, the first one is chosen as the ``key method'' for the class,
11158 and the vtable is only emitted in the translation unit where the key
11161 @emph{Note:} If the chosen key method is later defined as inline, the
11162 vtable will still be emitted in every translation unit which defines it.
11163 Make sure that any inline virtuals are declared inline in the class
11164 body, even if they are not defined there.
11166 @item type_info objects
11169 C++ requires information about types to be written out in order to
11170 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
11171 For polymorphic classes (classes with virtual functions), the type_info
11172 object is written out along with the vtable so that @samp{dynamic_cast}
11173 can determine the dynamic type of a class object at runtime. For all
11174 other types, we write out the type_info object when it is used: when
11175 applying @samp{typeid} to an expression, throwing an object, or
11176 referring to a type in a catch clause or exception specification.
11178 @item Template Instantiations
11179 Most everything in this section also applies to template instantiations,
11180 but there are other options as well.
11181 @xref{Template Instantiation,,Where's the Template?}.
11185 When used with GNU ld version 2.8 or later on an ELF system such as
11186 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
11187 these constructs will be discarded at link time. This is known as
11190 On targets that don't support COMDAT, but do support weak symbols, GCC
11191 will use them. This way one copy will override all the others, but
11192 the unused copies will still take up space in the executable.
11194 For targets which do not support either COMDAT or weak symbols,
11195 most entities with vague linkage will be emitted as local symbols to
11196 avoid duplicate definition errors from the linker. This will not happen
11197 for local statics in inlines, however, as having multiple copies will
11198 almost certainly break things.
11200 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
11201 another way to control placement of these constructs.
11203 @node C++ Interface
11204 @section #pragma interface and implementation
11206 @cindex interface and implementation headers, C++
11207 @cindex C++ interface and implementation headers
11208 @cindex pragmas, interface and implementation
11210 @code{#pragma interface} and @code{#pragma implementation} provide the
11211 user with a way of explicitly directing the compiler to emit entities
11212 with vague linkage (and debugging information) in a particular
11215 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
11216 most cases, because of COMDAT support and the ``key method'' heuristic
11217 mentioned in @ref{Vague Linkage}. Using them can actually cause your
11218 program to grow due to unnecessary out-of-line copies of inline
11219 functions. Currently (3.4) the only benefit of these
11220 @code{#pragma}s is reduced duplication of debugging information, and
11221 that should be addressed soon on DWARF 2 targets with the use of
11225 @item #pragma interface
11226 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
11227 @kindex #pragma interface
11228 Use this directive in @emph{header files} that define object classes, to save
11229 space in most of the object files that use those classes. Normally,
11230 local copies of certain information (backup copies of inline member
11231 functions, debugging information, and the internal tables that implement
11232 virtual functions) must be kept in each object file that includes class
11233 definitions. You can use this pragma to avoid such duplication. When a
11234 header file containing @samp{#pragma interface} is included in a
11235 compilation, this auxiliary information will not be generated (unless
11236 the main input source file itself uses @samp{#pragma implementation}).
11237 Instead, the object files will contain references to be resolved at link
11240 The second form of this directive is useful for the case where you have
11241 multiple headers with the same name in different directories. If you
11242 use this form, you must specify the same string to @samp{#pragma
11245 @item #pragma implementation
11246 @itemx #pragma implementation "@var{objects}.h"
11247 @kindex #pragma implementation
11248 Use this pragma in a @emph{main input file}, when you want full output from
11249 included header files to be generated (and made globally visible). The
11250 included header file, in turn, should use @samp{#pragma interface}.
11251 Backup copies of inline member functions, debugging information, and the
11252 internal tables used to implement virtual functions are all generated in
11253 implementation files.
11255 @cindex implied @code{#pragma implementation}
11256 @cindex @code{#pragma implementation}, implied
11257 @cindex naming convention, implementation headers
11258 If you use @samp{#pragma implementation} with no argument, it applies to
11259 an include file with the same basename@footnote{A file's @dfn{basename}
11260 was the name stripped of all leading path information and of trailing
11261 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
11262 file. For example, in @file{allclass.cc}, giving just
11263 @samp{#pragma implementation}
11264 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
11266 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
11267 an implementation file whenever you would include it from
11268 @file{allclass.cc} even if you never specified @samp{#pragma
11269 implementation}. This was deemed to be more trouble than it was worth,
11270 however, and disabled.
11272 Use the string argument if you want a single implementation file to
11273 include code from multiple header files. (You must also use
11274 @samp{#include} to include the header file; @samp{#pragma
11275 implementation} only specifies how to use the file---it doesn't actually
11278 There is no way to split up the contents of a single header file into
11279 multiple implementation files.
11282 @cindex inlining and C++ pragmas
11283 @cindex C++ pragmas, effect on inlining
11284 @cindex pragmas in C++, effect on inlining
11285 @samp{#pragma implementation} and @samp{#pragma interface} also have an
11286 effect on function inlining.
11288 If you define a class in a header file marked with @samp{#pragma
11289 interface}, the effect on an inline function defined in that class is
11290 similar to an explicit @code{extern} declaration---the compiler emits
11291 no code at all to define an independent version of the function. Its
11292 definition is used only for inlining with its callers.
11294 @opindex fno-implement-inlines
11295 Conversely, when you include the same header file in a main source file
11296 that declares it as @samp{#pragma implementation}, the compiler emits
11297 code for the function itself; this defines a version of the function
11298 that can be found via pointers (or by callers compiled without
11299 inlining). If all calls to the function can be inlined, you can avoid
11300 emitting the function by compiling with @option{-fno-implement-inlines}.
11301 If any calls were not inlined, you will get linker errors.
11303 @node Template Instantiation
11304 @section Where's the Template?
11305 @cindex template instantiation
11307 C++ templates are the first language feature to require more
11308 intelligence from the environment than one usually finds on a UNIX
11309 system. Somehow the compiler and linker have to make sure that each
11310 template instance occurs exactly once in the executable if it is needed,
11311 and not at all otherwise. There are two basic approaches to this
11312 problem, which are referred to as the Borland model and the Cfront model.
11315 @item Borland model
11316 Borland C++ solved the template instantiation problem by adding the code
11317 equivalent of common blocks to their linker; the compiler emits template
11318 instances in each translation unit that uses them, and the linker
11319 collapses them together. The advantage of this model is that the linker
11320 only has to consider the object files themselves; there is no external
11321 complexity to worry about. This disadvantage is that compilation time
11322 is increased because the template code is being compiled repeatedly.
11323 Code written for this model tends to include definitions of all
11324 templates in the header file, since they must be seen to be
11328 The AT&T C++ translator, Cfront, solved the template instantiation
11329 problem by creating the notion of a template repository, an
11330 automatically maintained place where template instances are stored. A
11331 more modern version of the repository works as follows: As individual
11332 object files are built, the compiler places any template definitions and
11333 instantiations encountered in the repository. At link time, the link
11334 wrapper adds in the objects in the repository and compiles any needed
11335 instances that were not previously emitted. The advantages of this
11336 model are more optimal compilation speed and the ability to use the
11337 system linker; to implement the Borland model a compiler vendor also
11338 needs to replace the linker. The disadvantages are vastly increased
11339 complexity, and thus potential for error; for some code this can be
11340 just as transparent, but in practice it can been very difficult to build
11341 multiple programs in one directory and one program in multiple
11342 directories. Code written for this model tends to separate definitions
11343 of non-inline member templates into a separate file, which should be
11344 compiled separately.
11347 When used with GNU ld version 2.8 or later on an ELF system such as
11348 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
11349 Borland model. On other systems, G++ implements neither automatic
11352 A future version of G++ will support a hybrid model whereby the compiler
11353 will emit any instantiations for which the template definition is
11354 included in the compile, and store template definitions and
11355 instantiation context information into the object file for the rest.
11356 The link wrapper will extract that information as necessary and invoke
11357 the compiler to produce the remaining instantiations. The linker will
11358 then combine duplicate instantiations.
11360 In the mean time, you have the following options for dealing with
11361 template instantiations:
11366 Compile your template-using code with @option{-frepo}. The compiler will
11367 generate files with the extension @samp{.rpo} listing all of the
11368 template instantiations used in the corresponding object files which
11369 could be instantiated there; the link wrapper, @samp{collect2}, will
11370 then update the @samp{.rpo} files to tell the compiler where to place
11371 those instantiations and rebuild any affected object files. The
11372 link-time overhead is negligible after the first pass, as the compiler
11373 will continue to place the instantiations in the same files.
11375 This is your best option for application code written for the Borland
11376 model, as it will just work. Code written for the Cfront model will
11377 need to be modified so that the template definitions are available at
11378 one or more points of instantiation; usually this is as simple as adding
11379 @code{#include <tmethods.cc>} to the end of each template header.
11381 For library code, if you want the library to provide all of the template
11382 instantiations it needs, just try to link all of its object files
11383 together; the link will fail, but cause the instantiations to be
11384 generated as a side effect. Be warned, however, that this may cause
11385 conflicts if multiple libraries try to provide the same instantiations.
11386 For greater control, use explicit instantiation as described in the next
11390 @opindex fno-implicit-templates
11391 Compile your code with @option{-fno-implicit-templates} to disable the
11392 implicit generation of template instances, and explicitly instantiate
11393 all the ones you use. This approach requires more knowledge of exactly
11394 which instances you need than do the others, but it's less
11395 mysterious and allows greater control. You can scatter the explicit
11396 instantiations throughout your program, perhaps putting them in the
11397 translation units where the instances are used or the translation units
11398 that define the templates themselves; you can put all of the explicit
11399 instantiations you need into one big file; or you can create small files
11406 template class Foo<int>;
11407 template ostream& operator <<
11408 (ostream&, const Foo<int>&);
11411 for each of the instances you need, and create a template instantiation
11412 library from those.
11414 If you are using Cfront-model code, you can probably get away with not
11415 using @option{-fno-implicit-templates} when compiling files that don't
11416 @samp{#include} the member template definitions.
11418 If you use one big file to do the instantiations, you may want to
11419 compile it without @option{-fno-implicit-templates} so you get all of the
11420 instances required by your explicit instantiations (but not by any
11421 other files) without having to specify them as well.
11423 G++ has extended the template instantiation syntax given in the ISO
11424 standard to allow forward declaration of explicit instantiations
11425 (with @code{extern}), instantiation of the compiler support data for a
11426 template class (i.e.@: the vtable) without instantiating any of its
11427 members (with @code{inline}), and instantiation of only the static data
11428 members of a template class, without the support data or member
11429 functions (with (@code{static}):
11432 extern template int max (int, int);
11433 inline template class Foo<int>;
11434 static template class Foo<int>;
11438 Do nothing. Pretend G++ does implement automatic instantiation
11439 management. Code written for the Borland model will work fine, but
11440 each translation unit will contain instances of each of the templates it
11441 uses. In a large program, this can lead to an unacceptable amount of code
11445 @node Bound member functions
11446 @section Extracting the function pointer from a bound pointer to member function
11448 @cindex pointer to member function
11449 @cindex bound pointer to member function
11451 In C++, pointer to member functions (PMFs) are implemented using a wide
11452 pointer of sorts to handle all the possible call mechanisms; the PMF
11453 needs to store information about how to adjust the @samp{this} pointer,
11454 and if the function pointed to is virtual, where to find the vtable, and
11455 where in the vtable to look for the member function. If you are using
11456 PMFs in an inner loop, you should really reconsider that decision. If
11457 that is not an option, you can extract the pointer to the function that
11458 would be called for a given object/PMF pair and call it directly inside
11459 the inner loop, to save a bit of time.
11461 Note that you will still be paying the penalty for the call through a
11462 function pointer; on most modern architectures, such a call defeats the
11463 branch prediction features of the CPU@. This is also true of normal
11464 virtual function calls.
11466 The syntax for this extension is
11470 extern int (A::*fp)();
11471 typedef int (*fptr)(A *);
11473 fptr p = (fptr)(a.*fp);
11476 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
11477 no object is needed to obtain the address of the function. They can be
11478 converted to function pointers directly:
11481 fptr p1 = (fptr)(&A::foo);
11484 @opindex Wno-pmf-conversions
11485 You must specify @option{-Wno-pmf-conversions} to use this extension.
11487 @node C++ Attributes
11488 @section C++-Specific Variable, Function, and Type Attributes
11490 Some attributes only make sense for C++ programs.
11493 @item init_priority (@var{priority})
11494 @cindex init_priority attribute
11497 In Standard C++, objects defined at namespace scope are guaranteed to be
11498 initialized in an order in strict accordance with that of their definitions
11499 @emph{in a given translation unit}. No guarantee is made for initializations
11500 across translation units. However, GNU C++ allows users to control the
11501 order of initialization of objects defined at namespace scope with the
11502 @code{init_priority} attribute by specifying a relative @var{priority},
11503 a constant integral expression currently bounded between 101 and 65535
11504 inclusive. Lower numbers indicate a higher priority.
11506 In the following example, @code{A} would normally be created before
11507 @code{B}, but the @code{init_priority} attribute has reversed that order:
11510 Some_Class A __attribute__ ((init_priority (2000)));
11511 Some_Class B __attribute__ ((init_priority (543)));
11515 Note that the particular values of @var{priority} do not matter; only their
11518 @item java_interface
11519 @cindex java_interface attribute
11521 This type attribute informs C++ that the class is a Java interface. It may
11522 only be applied to classes declared within an @code{extern "Java"} block.
11523 Calls to methods declared in this interface will be dispatched using GCJ's
11524 interface table mechanism, instead of regular virtual table dispatch.
11528 See also @xref{Namespace Association}.
11530 @node Namespace Association
11531 @section Namespace Association
11533 @strong{Caution:} The semantics of this extension are not fully
11534 defined. Users should refrain from using this extension as its
11535 semantics may change subtly over time. It is possible that this
11536 extension will be removed in future versions of G++.
11538 A using-directive with @code{__attribute ((strong))} is stronger
11539 than a normal using-directive in two ways:
11543 Templates from the used namespace can be specialized and explicitly
11544 instantiated as though they were members of the using namespace.
11547 The using namespace is considered an associated namespace of all
11548 templates in the used namespace for purposes of argument-dependent
11552 The used namespace must be nested within the using namespace so that
11553 normal unqualified lookup works properly.
11555 This is useful for composing a namespace transparently from
11556 implementation namespaces. For example:
11561 template <class T> struct A @{ @};
11563 using namespace debug __attribute ((__strong__));
11564 template <> struct A<int> @{ @}; // @r{ok to specialize}
11566 template <class T> void f (A<T>);
11571 f (std::A<float>()); // @r{lookup finds} std::f
11577 @section Type Traits
11579 The C++ front-end implements syntactic extensions that allow to
11580 determine at compile time various characteristics of a type (or of a
11584 @item __has_nothrow_assign (type)
11585 If @code{type} is const qualified or is a reference type then the trait is
11586 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
11587 is true, else if @code{type} is a cv class or union type with copy assignment
11588 operators that are known not to throw an exception then the trait is true,
11589 else it is false. Requires: @code{type} shall be a complete type, an array
11590 type of unknown bound, or is a @code{void} type.
11592 @item __has_nothrow_copy (type)
11593 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
11594 @code{type} is a cv class or union type with copy constructors that
11595 are known not to throw an exception then the trait is true, else it is false.
11596 Requires: @code{type} shall be a complete type, an array type of
11597 unknown bound, or is a @code{void} type.
11599 @item __has_nothrow_constructor (type)
11600 If @code{__has_trivial_constructor (type)} is true then the trait is
11601 true, else if @code{type} is a cv class or union type (or array
11602 thereof) with a default constructor that is known not to throw an
11603 exception then the trait is true, else it is false. Requires:
11604 @code{type} shall be a complete type, an array type of unknown bound,
11605 or is a @code{void} type.
11607 @item __has_trivial_assign (type)
11608 If @code{type} is const qualified or is a reference type then the trait is
11609 false. Otherwise if @code{__is_pod (type)} is true then the trait is
11610 true, else if @code{type} is a cv class or union type with a trivial
11611 copy assignment ([class.copy]) then the trait is true, else it is
11612 false. Requires: @code{type} shall be a complete type, an array type
11613 of unknown bound, or is a @code{void} type.
11615 @item __has_trivial_copy (type)
11616 If @code{__is_pod (type)} is true or @code{type} is a reference type
11617 then the trait is true, else if @code{type} is a cv class or union type
11618 with a trivial copy constructor ([class.copy]) then the trait
11619 is true, else it is false. Requires: @code{type} shall be a complete
11620 type, an array type of unknown bound, or is a @code{void} type.
11622 @item __has_trivial_constructor (type)
11623 If @code{__is_pod (type)} is true then the trait is true, else if
11624 @code{type} is a cv class or union type (or array thereof) with a
11625 trivial default constructor ([class.ctor]) then the trait is true,
11626 else it is false. Requires: @code{type} shall be a complete type, an
11627 array type of unknown bound, or is a @code{void} type.
11629 @item __has_trivial_destructor (type)
11630 If @code{__is_pod (type)} is true or @code{type} is a reference type then
11631 the trait is true, else if @code{type} is a cv class or union type (or
11632 array thereof) with a trivial destructor ([class.dtor]) then the trait
11633 is true, else it is false. Requires: @code{type} shall be a complete
11634 type, an array type of unknown bound, or is a @code{void} type.
11636 @item __has_virtual_destructor (type)
11637 If @code{type} is a class type with a virtual destructor
11638 ([class.dtor]) then the trait is true, else it is false. Requires:
11639 @code{type} shall be a complete type, an array type of unknown bound,
11640 or is a @code{void} type.
11642 @item __is_abstract (type)
11643 If @code{type} is an abstract class ([class.abstract]) then the trait
11644 is true, else it is false. Requires: @code{type} shall be a complete
11645 type, an array type of unknown bound, or is a @code{void} type.
11647 @item __is_base_of (base_type, derived_type)
11648 If @code{base_type} is a base class of @code{derived_type}
11649 ([class.derived]) then the trait is true, otherwise it is false.
11650 Top-level cv qualifications of @code{base_type} and
11651 @code{derived_type} are ignored. For the purposes of this trait, a
11652 class type is considered is own base. Requires: if @code{__is_class
11653 (base_type)} and @code{__is_class (derived_type)} are true and
11654 @code{base_type} and @code{derived_type} are not the same type
11655 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
11656 type. Diagnostic is produced if this requirement is not met.
11658 @item __is_class (type)
11659 If @code{type} is a cv class type, and not a union type
11660 ([basic.compound]) the the trait is true, else it is false.
11662 @item __is_empty (type)
11663 If @code{__is_class (type)} is false then the trait is false.
11664 Otherwise @code{type} is considered empty if and only if: @code{type}
11665 has no non-static data members, or all non-static data members, if
11666 any, are bit-fields of lenght 0, and @code{type} has no virtual
11667 members, and @code{type} has no virtual base classes, and @code{type}
11668 has no base classes @code{base_type} for which
11669 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
11670 be a complete type, an array type of unknown bound, or is a
11673 @item __is_enum (type)
11674 If @code{type} is a cv enumeration type ([basic.compound]) the the trait is
11675 true, else it is false.
11677 @item __is_pod (type)
11678 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
11679 else it is false. Requires: @code{type} shall be a complete type,
11680 an array type of unknown bound, or is a @code{void} type.
11682 @item __is_polymorphic (type)
11683 If @code{type} is a polymorphic class ([class.virtual]) then the trait
11684 is true, else it is false. Requires: @code{type} shall be a complete
11685 type, an array type of unknown bound, or is a @code{void} type.
11687 @item __is_union (type)
11688 If @code{type} is a cv union type ([basic.compound]) the the trait is
11689 true, else it is false.
11693 @node Java Exceptions
11694 @section Java Exceptions
11696 The Java language uses a slightly different exception handling model
11697 from C++. Normally, GNU C++ will automatically detect when you are
11698 writing C++ code that uses Java exceptions, and handle them
11699 appropriately. However, if C++ code only needs to execute destructors
11700 when Java exceptions are thrown through it, GCC will guess incorrectly.
11701 Sample problematic code is:
11704 struct S @{ ~S(); @};
11705 extern void bar(); // @r{is written in Java, and may throw exceptions}
11714 The usual effect of an incorrect guess is a link failure, complaining of
11715 a missing routine called @samp{__gxx_personality_v0}.
11717 You can inform the compiler that Java exceptions are to be used in a
11718 translation unit, irrespective of what it might think, by writing
11719 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
11720 @samp{#pragma} must appear before any functions that throw or catch
11721 exceptions, or run destructors when exceptions are thrown through them.
11723 You cannot mix Java and C++ exceptions in the same translation unit. It
11724 is believed to be safe to throw a C++ exception from one file through
11725 another file compiled for the Java exception model, or vice versa, but
11726 there may be bugs in this area.
11728 @node Deprecated Features
11729 @section Deprecated Features
11731 In the past, the GNU C++ compiler was extended to experiment with new
11732 features, at a time when the C++ language was still evolving. Now that
11733 the C++ standard is complete, some of those features are superseded by
11734 superior alternatives. Using the old features might cause a warning in
11735 some cases that the feature will be dropped in the future. In other
11736 cases, the feature might be gone already.
11738 While the list below is not exhaustive, it documents some of the options
11739 that are now deprecated:
11742 @item -fexternal-templates
11743 @itemx -falt-external-templates
11744 These are two of the many ways for G++ to implement template
11745 instantiation. @xref{Template Instantiation}. The C++ standard clearly
11746 defines how template definitions have to be organized across
11747 implementation units. G++ has an implicit instantiation mechanism that
11748 should work just fine for standard-conforming code.
11750 @item -fstrict-prototype
11751 @itemx -fno-strict-prototype
11752 Previously it was possible to use an empty prototype parameter list to
11753 indicate an unspecified number of parameters (like C), rather than no
11754 parameters, as C++ demands. This feature has been removed, except where
11755 it is required for backwards compatibility @xref{Backwards Compatibility}.
11758 G++ allows a virtual function returning @samp{void *} to be overridden
11759 by one returning a different pointer type. This extension to the
11760 covariant return type rules is now deprecated and will be removed from a
11763 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
11764 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
11765 and will be removed in a future version. Code using these operators
11766 should be modified to use @code{std::min} and @code{std::max} instead.
11768 The named return value extension has been deprecated, and is now
11771 The use of initializer lists with new expressions has been deprecated,
11772 and is now removed from G++.
11774 Floating and complex non-type template parameters have been deprecated,
11775 and are now removed from G++.
11777 The implicit typename extension has been deprecated and is now
11780 The use of default arguments in function pointers, function typedefs
11781 and other places where they are not permitted by the standard is
11782 deprecated and will be removed from a future version of G++.
11784 G++ allows floating-point literals to appear in integral constant expressions,
11785 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
11786 This extension is deprecated and will be removed from a future version.
11788 G++ allows static data members of const floating-point type to be declared
11789 with an initializer in a class definition. The standard only allows
11790 initializers for static members of const integral types and const
11791 enumeration types so this extension has been deprecated and will be removed
11792 from a future version.
11794 @node Backwards Compatibility
11795 @section Backwards Compatibility
11796 @cindex Backwards Compatibility
11797 @cindex ARM [Annotated C++ Reference Manual]
11799 Now that there is a definitive ISO standard C++, G++ has a specification
11800 to adhere to. The C++ language evolved over time, and features that
11801 used to be acceptable in previous drafts of the standard, such as the ARM
11802 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
11803 compilation of C++ written to such drafts, G++ contains some backwards
11804 compatibilities. @emph{All such backwards compatibility features are
11805 liable to disappear in future versions of G++.} They should be considered
11806 deprecated @xref{Deprecated Features}.
11810 If a variable is declared at for scope, it used to remain in scope until
11811 the end of the scope which contained the for statement (rather than just
11812 within the for scope). G++ retains this, but issues a warning, if such a
11813 variable is accessed outside the for scope.
11815 @item Implicit C language
11816 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
11817 scope to set the language. On such systems, all header files are
11818 implicitly scoped inside a C language scope. Also, an empty prototype
11819 @code{()} will be treated as an unspecified number of arguments, rather
11820 than no arguments, as C++ demands.