1 @c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1996, 1998, 1999, 2000,
2 @c 2001, 2002, 2003, 2004, 2005, 2006 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 * Decimal Float:: Decimal Floating Point.
37 * Hex Floats:: Hexadecimal floating-point constants.
38 * Zero Length:: Zero-length arrays.
39 * Variable Length:: Arrays whose length is computed at run time.
40 * Empty Structures:: Structures with no members.
41 * Variadic Macros:: Macros with a variable number of arguments.
42 * Escaped Newlines:: Slightly looser rules for escaped newlines.
43 * Subscripting:: Any array can be subscripted, even if not an lvalue.
44 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
45 * Initializers:: Non-constant initializers.
46 * Compound Literals:: Compound literals give structures, unions
48 * Designated Inits:: Labeling elements of initializers.
49 * Cast to Union:: Casting to union type from any member of the union.
50 * Case Ranges:: `case 1 ... 9' and such.
51 * Mixed Declarations:: Mixing declarations and code.
52 * Function Attributes:: Declaring that functions have no side effects,
53 or that they can never return.
54 * Attribute Syntax:: Formal syntax for attributes.
55 * Function Prototypes:: Prototype declarations and old-style definitions.
56 * C++ Comments:: C++ comments are recognized.
57 * Dollar Signs:: Dollar sign is allowed in identifiers.
58 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
59 * Variable Attributes:: Specifying attributes of variables.
60 * Type Attributes:: Specifying attributes of types.
61 * Alignment:: Inquiring about the alignment of a type or variable.
62 * Inline:: Defining inline functions (as fast as macros).
63 * Extended Asm:: Assembler instructions with C expressions as operands.
64 (With them you can define ``built-in'' functions.)
65 * Constraints:: Constraints for asm operands
66 * Asm Labels:: Specifying the assembler name to use for a C symbol.
67 * Explicit Reg Vars:: Defining variables residing in specified registers.
68 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
69 * Incomplete Enums:: @code{enum foo;}, with details to follow.
70 * Function Names:: Printable strings which are the name of the current
72 * Return Address:: Getting the return or frame address of a function.
73 * Vector Extensions:: Using vector instructions through built-in functions.
74 * Offsetof:: Special syntax for implementing @code{offsetof}.
75 * Atomic Builtins:: Built-in functions for atomic memory access.
76 * Object Size Checking:: Built-in functions for limited buffer overflow
78 * Other Builtins:: Other built-in functions.
79 * Target Builtins:: Built-in functions specific to particular targets.
80 * Target Format Checks:: Format checks specific to particular targets.
81 * Pragmas:: Pragmas accepted by GCC.
82 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
83 * Thread-Local:: Per-thread variables.
87 @section Statements and Declarations in Expressions
88 @cindex statements inside expressions
89 @cindex declarations inside expressions
90 @cindex expressions containing statements
91 @cindex macros, statements in expressions
93 @c the above section title wrapped and causes an underfull hbox.. i
94 @c changed it from "within" to "in". --mew 4feb93
95 A compound statement enclosed in parentheses may appear as an expression
96 in GNU C@. This allows you to use loops, switches, and local variables
99 Recall that a compound statement is a sequence of statements surrounded
100 by braces; in this construct, parentheses go around the braces. For
104 (@{ int y = foo (); int z;
111 is a valid (though slightly more complex than necessary) expression
112 for the absolute value of @code{foo ()}.
114 The last thing in the compound statement should be an expression
115 followed by a semicolon; the value of this subexpression serves as the
116 value of the entire construct. (If you use some other kind of statement
117 last within the braces, the construct has type @code{void}, and thus
118 effectively no value.)
120 This feature is especially useful in making macro definitions ``safe'' (so
121 that they evaluate each operand exactly once). For example, the
122 ``maximum'' function is commonly defined as a macro in standard C as
126 #define max(a,b) ((a) > (b) ? (a) : (b))
130 @cindex side effects, macro argument
131 But this definition computes either @var{a} or @var{b} twice, with bad
132 results if the operand has side effects. In GNU C, if you know the
133 type of the operands (here taken as @code{int}), you can define
134 the macro safely as follows:
137 #define maxint(a,b) \
138 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
141 Embedded statements are not allowed in constant expressions, such as
142 the value of an enumeration constant, the width of a bit-field, or
143 the initial value of a static variable.
145 If you don't know the type of the operand, you can still do this, but you
146 must use @code{typeof} (@pxref{Typeof}).
148 In G++, the result value of a statement expression undergoes array and
149 function pointer decay, and is returned by value to the enclosing
150 expression. For instance, if @code{A} is a class, then
159 will construct a temporary @code{A} object to hold the result of the
160 statement expression, and that will be used to invoke @code{Foo}.
161 Therefore the @code{this} pointer observed by @code{Foo} will not be the
164 Any temporaries created within a statement within a statement expression
165 will be destroyed at the statement's end. This makes statement
166 expressions inside macros slightly different from function calls. In
167 the latter case temporaries introduced during argument evaluation will
168 be destroyed at the end of the statement that includes the function
169 call. In the statement expression case they will be destroyed during
170 the statement expression. For instance,
173 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
174 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
184 will have different places where temporaries are destroyed. For the
185 @code{macro} case, the temporary @code{X} will be destroyed just after
186 the initialization of @code{b}. In the @code{function} case that
187 temporary will be destroyed when the function returns.
189 These considerations mean that it is probably a bad idea to use
190 statement-expressions of this form in header files that are designed to
191 work with C++. (Note that some versions of the GNU C Library contained
192 header files using statement-expression that lead to precisely this
195 Jumping into a statement expression with @code{goto} or using a
196 @code{switch} statement outside the statement expression with a
197 @code{case} or @code{default} label inside the statement expression is
198 not permitted. Jumping into a statement expression with a computed
199 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
200 Jumping out of a statement expression is permitted, but if the
201 statement expression is part of a larger expression then it is
202 unspecified which other subexpressions of that expression have been
203 evaluated except where the language definition requires certain
204 subexpressions to be evaluated before or after the statement
205 expression. In any case, as with a function call the evaluation of a
206 statement expression is not interleaved with the evaluation of other
207 parts of the containing expression. For example,
210 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
214 will call @code{foo} and @code{bar1} and will not call @code{baz} but
215 may or may not call @code{bar2}. If @code{bar2} is called, it will be
216 called after @code{foo} and before @code{bar1}
219 @section Locally Declared Labels
221 @cindex macros, local labels
223 GCC allows you to declare @dfn{local labels} in any nested block
224 scope. A local label is just like an ordinary label, but you can
225 only reference it (with a @code{goto} statement, or by taking its
226 address) within the block in which it was declared.
228 A local label declaration looks like this:
231 __label__ @var{label};
238 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
241 Local label declarations must come at the beginning of the block,
242 before any ordinary declarations or statements.
244 The label declaration defines the label @emph{name}, but does not define
245 the label itself. You must do this in the usual way, with
246 @code{@var{label}:}, within the statements of the statement expression.
248 The local label feature is useful for complex macros. If a macro
249 contains nested loops, a @code{goto} can be useful for breaking out of
250 them. However, an ordinary label whose scope is the whole function
251 cannot be used: if the macro can be expanded several times in one
252 function, the label will be multiply defined in that function. A
253 local label avoids this problem. For example:
256 #define SEARCH(value, array, target) \
259 typeof (target) _SEARCH_target = (target); \
260 typeof (*(array)) *_SEARCH_array = (array); \
263 for (i = 0; i < max; i++) \
264 for (j = 0; j < max; j++) \
265 if (_SEARCH_array[i][j] == _SEARCH_target) \
266 @{ (value) = i; goto found; @} \
272 This could also be written using a statement-expression:
275 #define SEARCH(array, target) \
278 typeof (target) _SEARCH_target = (target); \
279 typeof (*(array)) *_SEARCH_array = (array); \
282 for (i = 0; i < max; i++) \
283 for (j = 0; j < max; j++) \
284 if (_SEARCH_array[i][j] == _SEARCH_target) \
285 @{ value = i; goto found; @} \
292 Local label declarations also make the labels they declare visible to
293 nested functions, if there are any. @xref{Nested Functions}, for details.
295 @node Labels as Values
296 @section Labels as Values
297 @cindex labels as values
298 @cindex computed gotos
299 @cindex goto with computed label
300 @cindex address of a label
302 You can get the address of a label defined in the current function
303 (or a containing function) with the unary operator @samp{&&}. The
304 value has type @code{void *}. This value is a constant and can be used
305 wherever a constant of that type is valid. For example:
313 To use these values, you need to be able to jump to one. This is done
314 with the computed goto statement@footnote{The analogous feature in
315 Fortran is called an assigned goto, but that name seems inappropriate in
316 C, where one can do more than simply store label addresses in label
317 variables.}, @code{goto *@var{exp};}. For example,
324 Any expression of type @code{void *} is allowed.
326 One way of using these constants is in initializing a static array that
327 will serve as a jump table:
330 static void *array[] = @{ &&foo, &&bar, &&hack @};
333 Then you can select a label with indexing, like this:
340 Note that this does not check whether the subscript is in bounds---array
341 indexing in C never does that.
343 Such an array of label values serves a purpose much like that of the
344 @code{switch} statement. The @code{switch} statement is cleaner, so
345 use that rather than an array unless the problem does not fit a
346 @code{switch} statement very well.
348 Another use of label values is in an interpreter for threaded code.
349 The labels within the interpreter function can be stored in the
350 threaded code for super-fast dispatching.
352 You may not use this mechanism to jump to code in a different function.
353 If you do that, totally unpredictable things will happen. The best way to
354 avoid this is to store the label address only in automatic variables and
355 never pass it as an argument.
357 An alternate way to write the above example is
360 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
362 goto *(&&foo + array[i]);
366 This is more friendly to code living in shared libraries, as it reduces
367 the number of dynamic relocations that are needed, and by consequence,
368 allows the data to be read-only.
370 @node Nested Functions
371 @section Nested Functions
372 @cindex nested functions
373 @cindex downward funargs
376 A @dfn{nested function} is a function defined inside another function.
377 (Nested functions are not supported for GNU C++.) The nested function's
378 name is local to the block where it is defined. For example, here we
379 define a nested function named @code{square}, and call it twice:
383 foo (double a, double b)
385 double square (double z) @{ return z * z; @}
387 return square (a) + square (b);
392 The nested function can access all the variables of the containing
393 function that are visible at the point of its definition. This is
394 called @dfn{lexical scoping}. For example, here we show a nested
395 function which uses an inherited variable named @code{offset}:
399 bar (int *array, int offset, int size)
401 int access (int *array, int index)
402 @{ return array[index + offset]; @}
405 for (i = 0; i < size; i++)
406 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
411 Nested function definitions are permitted within functions in the places
412 where variable definitions are allowed; that is, in any block, mixed
413 with the other declarations and statements in the block.
415 It is possible to call the nested function from outside the scope of its
416 name by storing its address or passing the address to another function:
419 hack (int *array, int size)
421 void store (int index, int value)
422 @{ array[index] = value; @}
424 intermediate (store, size);
428 Here, the function @code{intermediate} receives the address of
429 @code{store} as an argument. If @code{intermediate} calls @code{store},
430 the arguments given to @code{store} are used to store into @code{array}.
431 But this technique works only so long as the containing function
432 (@code{hack}, in this example) does not exit.
434 If you try to call the nested function through its address after the
435 containing function has exited, all hell will break loose. If you try
436 to call it after a containing scope level has exited, and if it refers
437 to some of the variables that are no longer in scope, you may be lucky,
438 but it's not wise to take the risk. If, however, the nested function
439 does not refer to anything that has gone out of scope, you should be
442 GCC implements taking the address of a nested function using a technique
443 called @dfn{trampolines}. A paper describing them is available as
446 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
448 A nested function can jump to a label inherited from a containing
449 function, provided the label was explicitly declared in the containing
450 function (@pxref{Local Labels}). Such a jump returns instantly to the
451 containing function, exiting the nested function which did the
452 @code{goto} and any intermediate functions as well. Here is an example:
456 bar (int *array, int offset, int size)
459 int access (int *array, int index)
463 return array[index + offset];
467 for (i = 0; i < size; i++)
468 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
472 /* @r{Control comes here from @code{access}
473 if it detects an error.} */
480 A nested function always has no linkage. Declaring one with
481 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
482 before its definition, use @code{auto} (which is otherwise meaningless
483 for function declarations).
486 bar (int *array, int offset, int size)
489 auto int access (int *, int);
491 int access (int *array, int index)
495 return array[index + offset];
501 @node Constructing Calls
502 @section Constructing Function Calls
503 @cindex constructing calls
504 @cindex forwarding calls
506 Using the built-in functions described below, you can record
507 the arguments a function received, and call another function
508 with the same arguments, without knowing the number or types
511 You can also record the return value of that function call,
512 and later return that value, without knowing what data type
513 the function tried to return (as long as your caller expects
516 However, these built-in functions may interact badly with some
517 sophisticated features or other extensions of the language. It
518 is, therefore, not recommended to use them outside very simple
519 functions acting as mere forwarders for their arguments.
521 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
522 This built-in function returns a pointer to data
523 describing how to perform a call with the same arguments as were passed
524 to the current function.
526 The function saves the arg pointer register, structure value address,
527 and all registers that might be used to pass arguments to a function
528 into a block of memory allocated on the stack. Then it returns the
529 address of that block.
532 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
533 This built-in function invokes @var{function}
534 with a copy of the parameters described by @var{arguments}
537 The value of @var{arguments} should be the value returned by
538 @code{__builtin_apply_args}. The argument @var{size} specifies the size
539 of the stack argument data, in bytes.
541 This function returns a pointer to data describing
542 how to return whatever value was returned by @var{function}. The data
543 is saved in a block of memory allocated on the stack.
545 It is not always simple to compute the proper value for @var{size}. The
546 value is used by @code{__builtin_apply} to compute the amount of data
547 that should be pushed on the stack and copied from the incoming argument
551 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
552 This built-in function returns the value described by @var{result} from
553 the containing function. You should specify, for @var{result}, a value
554 returned by @code{__builtin_apply}.
558 @section Referring to a Type with @code{typeof}
561 @cindex macros, types of arguments
563 Another way to refer to the type of an expression is with @code{typeof}.
564 The syntax of using of this keyword looks like @code{sizeof}, but the
565 construct acts semantically like a type name defined with @code{typedef}.
567 There are two ways of writing the argument to @code{typeof}: with an
568 expression or with a type. Here is an example with an expression:
575 This assumes that @code{x} is an array of pointers to functions;
576 the type described is that of the values of the functions.
578 Here is an example with a typename as the argument:
585 Here the type described is that of pointers to @code{int}.
587 If you are writing a header file that must work when included in ISO C
588 programs, write @code{__typeof__} instead of @code{typeof}.
589 @xref{Alternate Keywords}.
591 A @code{typeof}-construct can be used anywhere a typedef name could be
592 used. For example, you can use it in a declaration, in a cast, or inside
593 of @code{sizeof} or @code{typeof}.
595 @code{typeof} is often useful in conjunction with the
596 statements-within-expressions feature. Here is how the two together can
597 be used to define a safe ``maximum'' macro that operates on any
598 arithmetic type and evaluates each of its arguments exactly once:
602 (@{ typeof (a) _a = (a); \
603 typeof (b) _b = (b); \
604 _a > _b ? _a : _b; @})
607 @cindex underscores in variables in macros
608 @cindex @samp{_} in variables in macros
609 @cindex local variables in macros
610 @cindex variables, local, in macros
611 @cindex macros, local variables in
613 The reason for using names that start with underscores for the local
614 variables is to avoid conflicts with variable names that occur within the
615 expressions that are substituted for @code{a} and @code{b}. Eventually we
616 hope to design a new form of declaration syntax that allows you to declare
617 variables whose scopes start only after their initializers; this will be a
618 more reliable way to prevent such conflicts.
621 Some more examples of the use of @code{typeof}:
625 This declares @code{y} with the type of what @code{x} points to.
632 This declares @code{y} as an array of such values.
639 This declares @code{y} as an array of pointers to characters:
642 typeof (typeof (char *)[4]) y;
646 It is equivalent to the following traditional C declaration:
652 To see the meaning of the declaration using @code{typeof}, and why it
653 might be a useful way to write, rewrite it with these macros:
656 #define pointer(T) typeof(T *)
657 #define array(T, N) typeof(T [N])
661 Now the declaration can be rewritten this way:
664 array (pointer (char), 4) y;
668 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
669 pointers to @code{char}.
672 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
673 a more limited extension which permitted one to write
676 typedef @var{T} = @var{expr};
680 with the effect of declaring @var{T} to have the type of the expression
681 @var{expr}. This extension does not work with GCC 3 (versions between
682 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
683 relies on it should be rewritten to use @code{typeof}:
686 typedef typeof(@var{expr}) @var{T};
690 This will work with all versions of GCC@.
693 @section Conditionals with Omitted Operands
694 @cindex conditional expressions, extensions
695 @cindex omitted middle-operands
696 @cindex middle-operands, omitted
697 @cindex extensions, @code{?:}
698 @cindex @code{?:} extensions
700 The middle operand in a conditional expression may be omitted. Then
701 if the first operand is nonzero, its value is the value of the conditional
704 Therefore, the expression
711 has the value of @code{x} if that is nonzero; otherwise, the value of
714 This example is perfectly equivalent to
720 @cindex side effect in ?:
721 @cindex ?: side effect
723 In this simple case, the ability to omit the middle operand is not
724 especially useful. When it becomes useful is when the first operand does,
725 or may (if it is a macro argument), contain a side effect. Then repeating
726 the operand in the middle would perform the side effect twice. Omitting
727 the middle operand uses the value already computed without the undesirable
728 effects of recomputing it.
731 @section Double-Word Integers
732 @cindex @code{long long} data types
733 @cindex double-word arithmetic
734 @cindex multiprecision arithmetic
735 @cindex @code{LL} integer suffix
736 @cindex @code{ULL} integer suffix
738 ISO C99 supports data types for integers that are at least 64 bits wide,
739 and as an extension GCC supports them in C89 mode and in C++.
740 Simply write @code{long long int} for a signed integer, or
741 @code{unsigned long long int} for an unsigned integer. To make an
742 integer constant of type @code{long long int}, add the suffix @samp{LL}
743 to the integer. To make an integer constant of type @code{unsigned long
744 long int}, add the suffix @samp{ULL} to the integer.
746 You can use these types in arithmetic like any other integer types.
747 Addition, subtraction, and bitwise boolean operations on these types
748 are open-coded on all types of machines. Multiplication is open-coded
749 if the machine supports fullword-to-doubleword a widening multiply
750 instruction. Division and shifts are open-coded only on machines that
751 provide special support. The operations that are not open-coded use
752 special library routines that come with GCC@.
754 There may be pitfalls when you use @code{long long} types for function
755 arguments, unless you declare function prototypes. If a function
756 expects type @code{int} for its argument, and you pass a value of type
757 @code{long long int}, confusion will result because the caller and the
758 subroutine will disagree about the number of bytes for the argument.
759 Likewise, if the function expects @code{long long int} and you pass
760 @code{int}. The best way to avoid such problems is to use prototypes.
763 @section Complex Numbers
764 @cindex complex numbers
765 @cindex @code{_Complex} keyword
766 @cindex @code{__complex__} keyword
768 ISO C99 supports complex floating data types, and as an extension GCC
769 supports them in C89 mode and in C++, and supports complex integer data
770 types which are not part of ISO C99. You can declare complex types
771 using the keyword @code{_Complex}. As an extension, the older GNU
772 keyword @code{__complex__} is also supported.
774 For example, @samp{_Complex double x;} declares @code{x} as a
775 variable whose real part and imaginary part are both of type
776 @code{double}. @samp{_Complex short int y;} declares @code{y} to
777 have real and imaginary parts of type @code{short int}; this is not
778 likely to be useful, but it shows that the set of complex types is
781 To write a constant with a complex data type, use the suffix @samp{i} or
782 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
783 has type @code{_Complex float} and @code{3i} has type
784 @code{_Complex int}. Such a constant always has a pure imaginary
785 value, but you can form any complex value you like by adding one to a
786 real constant. This is a GNU extension; if you have an ISO C99
787 conforming C library (such as GNU libc), and want to construct complex
788 constants of floating type, you should include @code{<complex.h>} and
789 use the macros @code{I} or @code{_Complex_I} instead.
791 @cindex @code{__real__} keyword
792 @cindex @code{__imag__} keyword
793 To extract the real part of a complex-valued expression @var{exp}, write
794 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
795 extract the imaginary part. This is a GNU extension; for values of
796 floating type, you should use the ISO C99 functions @code{crealf},
797 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
798 @code{cimagl}, declared in @code{<complex.h>} and also provided as
799 built-in functions by GCC@.
801 @cindex complex conjugation
802 The operator @samp{~} performs complex conjugation when used on a value
803 with a complex type. This is a GNU extension; for values of
804 floating type, you should use the ISO C99 functions @code{conjf},
805 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
806 provided as built-in functions by GCC@.
808 GCC can allocate complex automatic variables in a noncontiguous
809 fashion; it's even possible for the real part to be in a register while
810 the imaginary part is on the stack (or vice-versa). Only the DWARF2
811 debug info format can represent this, so use of DWARF2 is recommended.
812 If you are using the stabs debug info format, GCC describes a noncontiguous
813 complex variable as if it were two separate variables of noncomplex type.
814 If the variable's actual name is @code{foo}, the two fictitious
815 variables are named @code{foo$real} and @code{foo$imag}. You can
816 examine and set these two fictitious variables with your debugger.
819 @section Decimal Floating Point
820 @cindex decimal floating point
821 @cindex @code{_Decimal32} data type
822 @cindex @code{_Decimal64} data type
823 @cindex @code{_Decimal128} data type
824 @cindex @code{df} integer suffix
825 @cindex @code{dd} integer suffix
826 @cindex @code{dl} integer suffix
827 @cindex @code{DF} integer suffix
828 @cindex @code{DD} integer suffix
829 @cindex @code{DL} integer suffix
831 GNU C supports decimal floating point types in addition to the
832 standard floating-point types. This extension supports decimal
833 floating-point arithmetic as defined in IEEE-754R, the proposed
834 revision of IEEE-754. The C language extension is defined in ISO/IEC
835 DTR 24732, Draft 5. Support for this functionality will change when
836 it is accepted into the C standard and might change for new drafts
837 of the proposal. Calling conventions for any target might also change.
838 Not all targets support decimal floating point.
840 Support for decimal floating point includes the arithmetic operators
841 add, subtract, multiply, divide; unary arithmetic operators;
842 relational operators; equality operators; and conversions to and from
843 integer and other floating-point types. Use a suffix @samp{df} or
844 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
845 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
848 Passing a decimal floating-point value as an argument to a function
849 without a prototype is undefined.
851 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
852 are supported by the DWARF2 debug information format.
858 ISO C99 supports floating-point numbers written not only in the usual
859 decimal notation, such as @code{1.55e1}, but also numbers such as
860 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
861 supports this in C89 mode (except in some cases when strictly
862 conforming) and in C++. In that format the
863 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
864 mandatory. The exponent is a decimal number that indicates the power of
865 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
872 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
873 is the same as @code{1.55e1}.
875 Unlike for floating-point numbers in the decimal notation the exponent
876 is always required in the hexadecimal notation. Otherwise the compiler
877 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
878 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
879 extension for floating-point constants of type @code{float}.
882 @section Arrays of Length Zero
883 @cindex arrays of length zero
884 @cindex zero-length arrays
885 @cindex length-zero arrays
886 @cindex flexible array members
888 Zero-length arrays are allowed in GNU C@. They are very useful as the
889 last element of a structure which is really a header for a variable-length
898 struct line *thisline = (struct line *)
899 malloc (sizeof (struct line) + this_length);
900 thisline->length = this_length;
903 In ISO C90, you would have to give @code{contents} a length of 1, which
904 means either you waste space or complicate the argument to @code{malloc}.
906 In ISO C99, you would use a @dfn{flexible array member}, which is
907 slightly different in syntax and semantics:
911 Flexible array members are written as @code{contents[]} without
915 Flexible array members have incomplete type, and so the @code{sizeof}
916 operator may not be applied. As a quirk of the original implementation
917 of zero-length arrays, @code{sizeof} evaluates to zero.
920 Flexible array members may only appear as the last member of a
921 @code{struct} that is otherwise non-empty.
924 A structure containing a flexible array member, or a union containing
925 such a structure (possibly recursively), may not be a member of a
926 structure or an element of an array. (However, these uses are
927 permitted by GCC as extensions.)
930 GCC versions before 3.0 allowed zero-length arrays to be statically
931 initialized, as if they were flexible arrays. In addition to those
932 cases that were useful, it also allowed initializations in situations
933 that would corrupt later data. Non-empty initialization of zero-length
934 arrays is now treated like any case where there are more initializer
935 elements than the array holds, in that a suitable warning about "excess
936 elements in array" is given, and the excess elements (all of them, in
937 this case) are ignored.
939 Instead GCC allows static initialization of flexible array members.
940 This is equivalent to defining a new structure containing the original
941 structure followed by an array of sufficient size to contain the data.
942 I.e.@: in the following, @code{f1} is constructed as if it were declared
948 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
951 struct f1 f1; int data[3];
952 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
956 The convenience of this extension is that @code{f1} has the desired
957 type, eliminating the need to consistently refer to @code{f2.f1}.
959 This has symmetry with normal static arrays, in that an array of
960 unknown size is also written with @code{[]}.
962 Of course, this extension only makes sense if the extra data comes at
963 the end of a top-level object, as otherwise we would be overwriting
964 data at subsequent offsets. To avoid undue complication and confusion
965 with initialization of deeply nested arrays, we simply disallow any
966 non-empty initialization except when the structure is the top-level
970 struct foo @{ int x; int y[]; @};
971 struct bar @{ struct foo z; @};
973 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
974 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
975 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
976 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
979 @node Empty Structures
980 @section Structures With No Members
981 @cindex empty structures
982 @cindex zero-size structures
984 GCC permits a C structure to have no members:
991 The structure will have size zero. In C++, empty structures are part
992 of the language. G++ treats empty structures as if they had a single
993 member of type @code{char}.
995 @node Variable Length
996 @section Arrays of Variable Length
997 @cindex variable-length arrays
998 @cindex arrays of variable length
1001 Variable-length automatic arrays are allowed in ISO C99, and as an
1002 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1003 implementation of variable-length arrays does not yet conform in detail
1004 to the ISO C99 standard.) These arrays are
1005 declared like any other automatic arrays, but with a length that is not
1006 a constant expression. The storage is allocated at the point of
1007 declaration and deallocated when the brace-level is exited. For
1012 concat_fopen (char *s1, char *s2, char *mode)
1014 char str[strlen (s1) + strlen (s2) + 1];
1017 return fopen (str, mode);
1021 @cindex scope of a variable length array
1022 @cindex variable-length array scope
1023 @cindex deallocating variable length arrays
1024 Jumping or breaking out of the scope of the array name deallocates the
1025 storage. Jumping into the scope is not allowed; you get an error
1028 @cindex @code{alloca} vs variable-length arrays
1029 You can use the function @code{alloca} to get an effect much like
1030 variable-length arrays. The function @code{alloca} is available in
1031 many other C implementations (but not in all). On the other hand,
1032 variable-length arrays are more elegant.
1034 There are other differences between these two methods. Space allocated
1035 with @code{alloca} exists until the containing @emph{function} returns.
1036 The space for a variable-length array is deallocated as soon as the array
1037 name's scope ends. (If you use both variable-length arrays and
1038 @code{alloca} in the same function, deallocation of a variable-length array
1039 will also deallocate anything more recently allocated with @code{alloca}.)
1041 You can also use variable-length arrays as arguments to functions:
1045 tester (int len, char data[len][len])
1051 The length of an array is computed once when the storage is allocated
1052 and is remembered for the scope of the array in case you access it with
1055 If you want to pass the array first and the length afterward, you can
1056 use a forward declaration in the parameter list---another GNU extension.
1060 tester (int len; char data[len][len], int len)
1066 @cindex parameter forward declaration
1067 The @samp{int len} before the semicolon is a @dfn{parameter forward
1068 declaration}, and it serves the purpose of making the name @code{len}
1069 known when the declaration of @code{data} is parsed.
1071 You can write any number of such parameter forward declarations in the
1072 parameter list. They can be separated by commas or semicolons, but the
1073 last one must end with a semicolon, which is followed by the ``real''
1074 parameter declarations. Each forward declaration must match a ``real''
1075 declaration in parameter name and data type. ISO C99 does not support
1076 parameter forward declarations.
1078 @node Variadic Macros
1079 @section Macros with a Variable Number of Arguments.
1080 @cindex variable number of arguments
1081 @cindex macro with variable arguments
1082 @cindex rest argument (in macro)
1083 @cindex variadic macros
1085 In the ISO C standard of 1999, a macro can be declared to accept a
1086 variable number of arguments much as a function can. The syntax for
1087 defining the macro is similar to that of a function. Here is an
1091 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1094 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1095 such a macro, it represents the zero or more tokens until the closing
1096 parenthesis that ends the invocation, including any commas. This set of
1097 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1098 wherever it appears. See the CPP manual for more information.
1100 GCC has long supported variadic macros, and used a different syntax that
1101 allowed you to give a name to the variable arguments just like any other
1102 argument. Here is an example:
1105 #define debug(format, args...) fprintf (stderr, format, args)
1108 This is in all ways equivalent to the ISO C example above, but arguably
1109 more readable and descriptive.
1111 GNU CPP has two further variadic macro extensions, and permits them to
1112 be used with either of the above forms of macro definition.
1114 In standard C, you are not allowed to leave the variable argument out
1115 entirely; but you are allowed to pass an empty argument. For example,
1116 this invocation is invalid in ISO C, because there is no comma after
1123 GNU CPP permits you to completely omit the variable arguments in this
1124 way. In the above examples, the compiler would complain, though since
1125 the expansion of the macro still has the extra comma after the format
1128 To help solve this problem, CPP behaves specially for variable arguments
1129 used with the token paste operator, @samp{##}. If instead you write
1132 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1135 and if the variable arguments are omitted or empty, the @samp{##}
1136 operator causes the preprocessor to remove the comma before it. If you
1137 do provide some variable arguments in your macro invocation, GNU CPP
1138 does not complain about the paste operation and instead places the
1139 variable arguments after the comma. Just like any other pasted macro
1140 argument, these arguments are not macro expanded.
1142 @node Escaped Newlines
1143 @section Slightly Looser Rules for Escaped Newlines
1144 @cindex escaped newlines
1145 @cindex newlines (escaped)
1147 Recently, the preprocessor has relaxed its treatment of escaped
1148 newlines. Previously, the newline had to immediately follow a
1149 backslash. The current implementation allows whitespace in the form
1150 of spaces, horizontal and vertical tabs, and form feeds between the
1151 backslash and the subsequent newline. The preprocessor issues a
1152 warning, but treats it as a valid escaped newline and combines the two
1153 lines to form a single logical line. This works within comments and
1154 tokens, as well as between tokens. Comments are @emph{not} treated as
1155 whitespace for the purposes of this relaxation, since they have not
1156 yet been replaced with spaces.
1159 @section Non-Lvalue Arrays May Have Subscripts
1160 @cindex subscripting
1161 @cindex arrays, non-lvalue
1163 @cindex subscripting and function values
1164 In ISO C99, arrays that are not lvalues still decay to pointers, and
1165 may be subscripted, although they may not be modified or used after
1166 the next sequence point and the unary @samp{&} operator may not be
1167 applied to them. As an extension, GCC allows such arrays to be
1168 subscripted in C89 mode, though otherwise they do not decay to
1169 pointers outside C99 mode. For example,
1170 this is valid in GNU C though not valid in C89:
1174 struct foo @{int a[4];@};
1180 return f().a[index];
1186 @section Arithmetic on @code{void}- and Function-Pointers
1187 @cindex void pointers, arithmetic
1188 @cindex void, size of pointer to
1189 @cindex function pointers, arithmetic
1190 @cindex function, size of pointer to
1192 In GNU C, addition and subtraction operations are supported on pointers to
1193 @code{void} and on pointers to functions. This is done by treating the
1194 size of a @code{void} or of a function as 1.
1196 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1197 and on function types, and returns 1.
1199 @opindex Wpointer-arith
1200 The option @option{-Wpointer-arith} requests a warning if these extensions
1204 @section Non-Constant Initializers
1205 @cindex initializers, non-constant
1206 @cindex non-constant initializers
1208 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1209 automatic variable are not required to be constant expressions in GNU C@.
1210 Here is an example of an initializer with run-time varying elements:
1213 foo (float f, float g)
1215 float beat_freqs[2] = @{ f-g, f+g @};
1220 @node Compound Literals
1221 @section Compound Literals
1222 @cindex constructor expressions
1223 @cindex initializations in expressions
1224 @cindex structures, constructor expression
1225 @cindex expressions, constructor
1226 @cindex compound literals
1227 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1229 ISO C99 supports compound literals. A compound literal looks like
1230 a cast containing an initializer. Its value is an object of the
1231 type specified in the cast, containing the elements specified in
1232 the initializer; it is an lvalue. As an extension, GCC supports
1233 compound literals in C89 mode and in C++.
1235 Usually, the specified type is a structure. Assume that
1236 @code{struct foo} and @code{structure} are declared as shown:
1239 struct foo @{int a; char b[2];@} structure;
1243 Here is an example of constructing a @code{struct foo} with a compound literal:
1246 structure = ((struct foo) @{x + y, 'a', 0@});
1250 This is equivalent to writing the following:
1254 struct foo temp = @{x + y, 'a', 0@};
1259 You can also construct an array. If all the elements of the compound literal
1260 are (made up of) simple constant expressions, suitable for use in
1261 initializers of objects of static storage duration, then the compound
1262 literal can be coerced to a pointer to its first element and used in
1263 such an initializer, as shown here:
1266 char **foo = (char *[]) @{ "x", "y", "z" @};
1269 Compound literals for scalar types and union types are is
1270 also allowed, but then the compound literal is equivalent
1273 As a GNU extension, GCC allows initialization of objects with static storage
1274 duration by compound literals (which is not possible in ISO C99, because
1275 the initializer is not a constant).
1276 It is handled as if the object was initialized only with the bracket
1277 enclosed list if compound literal's and object types match.
1278 The initializer list of the compound literal must be constant.
1279 If the object being initialized has array type of unknown size, the size is
1280 determined by compound literal size.
1283 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1284 static int y[] = (int []) @{1, 2, 3@};
1285 static int z[] = (int [3]) @{1@};
1289 The above lines are equivalent to the following:
1291 static struct foo x = @{1, 'a', 'b'@};
1292 static int y[] = @{1, 2, 3@};
1293 static int z[] = @{1, 0, 0@};
1296 @node Designated Inits
1297 @section Designated Initializers
1298 @cindex initializers with labeled elements
1299 @cindex labeled elements in initializers
1300 @cindex case labels in initializers
1301 @cindex designated initializers
1303 Standard C89 requires the elements of an initializer to appear in a fixed
1304 order, the same as the order of the elements in the array or structure
1307 In ISO C99 you can give the elements in any order, specifying the array
1308 indices or structure field names they apply to, and GNU C allows this as
1309 an extension in C89 mode as well. This extension is not
1310 implemented in GNU C++.
1312 To specify an array index, write
1313 @samp{[@var{index}] =} before the element value. For example,
1316 int a[6] = @{ [4] = 29, [2] = 15 @};
1323 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1327 The index values must be constant expressions, even if the array being
1328 initialized is automatic.
1330 An alternative syntax for this which has been obsolete since GCC 2.5 but
1331 GCC still accepts is to write @samp{[@var{index}]} before the element
1332 value, with no @samp{=}.
1334 To initialize a range of elements to the same value, write
1335 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1336 extension. For example,
1339 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1343 If the value in it has side-effects, the side-effects will happen only once,
1344 not for each initialized field by the range initializer.
1347 Note that the length of the array is the highest value specified
1350 In a structure initializer, specify the name of a field to initialize
1351 with @samp{.@var{fieldname} =} before the element value. For example,
1352 given the following structure,
1355 struct point @{ int x, y; @};
1359 the following initialization
1362 struct point p = @{ .y = yvalue, .x = xvalue @};
1369 struct point p = @{ xvalue, yvalue @};
1372 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1373 @samp{@var{fieldname}:}, as shown here:
1376 struct point p = @{ y: yvalue, x: xvalue @};
1380 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1381 @dfn{designator}. You can also use a designator (or the obsolete colon
1382 syntax) when initializing a union, to specify which element of the union
1383 should be used. For example,
1386 union foo @{ int i; double d; @};
1388 union foo f = @{ .d = 4 @};
1392 will convert 4 to a @code{double} to store it in the union using
1393 the second element. By contrast, casting 4 to type @code{union foo}
1394 would store it into the union as the integer @code{i}, since it is
1395 an integer. (@xref{Cast to Union}.)
1397 You can combine this technique of naming elements with ordinary C
1398 initialization of successive elements. Each initializer element that
1399 does not have a designator applies to the next consecutive element of the
1400 array or structure. For example,
1403 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1410 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1413 Labeling the elements of an array initializer is especially useful
1414 when the indices are characters or belong to an @code{enum} type.
1419 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1420 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1423 @cindex designator lists
1424 You can also write a series of @samp{.@var{fieldname}} and
1425 @samp{[@var{index}]} designators before an @samp{=} to specify a
1426 nested subobject to initialize; the list is taken relative to the
1427 subobject corresponding to the closest surrounding brace pair. For
1428 example, with the @samp{struct point} declaration above:
1431 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1435 If the same field is initialized multiple times, it will have value from
1436 the last initialization. If any such overridden initialization has
1437 side-effect, it is unspecified whether the side-effect happens or not.
1438 Currently, GCC will discard them and issue a warning.
1441 @section Case Ranges
1443 @cindex ranges in case statements
1445 You can specify a range of consecutive values in a single @code{case} label,
1449 case @var{low} ... @var{high}:
1453 This has the same effect as the proper number of individual @code{case}
1454 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1456 This feature is especially useful for ranges of ASCII character codes:
1462 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1463 it may be parsed wrong when you use it with integer values. For example,
1478 @section Cast to a Union Type
1479 @cindex cast to a union
1480 @cindex union, casting to a
1482 A cast to union type is similar to other casts, except that the type
1483 specified is a union type. You can specify the type either with
1484 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1485 a constructor though, not a cast, and hence does not yield an lvalue like
1486 normal casts. (@xref{Compound Literals}.)
1488 The types that may be cast to the union type are those of the members
1489 of the union. Thus, given the following union and variables:
1492 union foo @{ int i; double d; @};
1498 both @code{x} and @code{y} can be cast to type @code{union foo}.
1500 Using the cast as the right-hand side of an assignment to a variable of
1501 union type is equivalent to storing in a member of the union:
1506 u = (union foo) x @equiv{} u.i = x
1507 u = (union foo) y @equiv{} u.d = y
1510 You can also use the union cast as a function argument:
1513 void hack (union foo);
1515 hack ((union foo) x);
1518 @node Mixed Declarations
1519 @section Mixed Declarations and Code
1520 @cindex mixed declarations and code
1521 @cindex declarations, mixed with code
1522 @cindex code, mixed with declarations
1524 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1525 within compound statements. As an extension, GCC also allows this in
1526 C89 mode. For example, you could do:
1535 Each identifier is visible from where it is declared until the end of
1536 the enclosing block.
1538 @node Function Attributes
1539 @section Declaring Attributes of Functions
1540 @cindex function attributes
1541 @cindex declaring attributes of functions
1542 @cindex functions that never return
1543 @cindex functions that return more than once
1544 @cindex functions that have no side effects
1545 @cindex functions in arbitrary sections
1546 @cindex functions that behave like malloc
1547 @cindex @code{volatile} applied to function
1548 @cindex @code{const} applied to function
1549 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1550 @cindex functions with non-null pointer arguments
1551 @cindex functions that are passed arguments in registers on the 386
1552 @cindex functions that pop the argument stack on the 386
1553 @cindex functions that do not pop the argument stack on the 386
1555 In GNU C, you declare certain things about functions called in your program
1556 which help the compiler optimize function calls and check your code more
1559 The keyword @code{__attribute__} allows you to specify special
1560 attributes when making a declaration. This keyword is followed by an
1561 attribute specification inside double parentheses. The following
1562 attributes are currently defined for functions on all targets:
1563 @code{noreturn}, @code{returns_twice}, @code{noinline}, @code{always_inline},
1564 @code{flatten}, @code{pure}, @code{const}, @code{nothrow}, @code{sentinel},
1565 @code{format}, @code{format_arg}, @code{no_instrument_function},
1566 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1567 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1568 @code{alias}, @code{warn_unused_result}, @code{nonnull}
1569 and @code{externally_visible}. Several other
1570 attributes are defined for functions on particular target systems. Other
1571 attributes, including @code{section} are supported for variables declarations
1572 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1574 You may also specify attributes with @samp{__} preceding and following
1575 each keyword. This allows you to use them in header files without
1576 being concerned about a possible macro of the same name. For example,
1577 you may use @code{__noreturn__} instead of @code{noreturn}.
1579 @xref{Attribute Syntax}, for details of the exact syntax for using
1583 @c Keep this table alphabetized by attribute name. Treat _ as space.
1585 @item alias ("@var{target}")
1586 @cindex @code{alias} attribute
1587 The @code{alias} attribute causes the declaration to be emitted as an
1588 alias for another symbol, which must be specified. For instance,
1591 void __f () @{ /* @r{Do something.} */; @}
1592 void f () __attribute__ ((weak, alias ("__f")));
1595 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
1596 mangled name for the target must be used. It is an error if @samp{__f}
1597 is not defined in the same translation unit.
1599 Not all target machines support this attribute.
1602 @cindex @code{always_inline} function attribute
1603 Generally, functions are not inlined unless optimization is specified.
1604 For functions declared inline, this attribute inlines the function even
1605 if no optimization level was specified.
1607 @cindex @code{flatten} function attribute
1609 Generally, inlining into a function is limited. For a function marked with
1610 this attribute, every call inside this function will be inlined, if possible.
1611 Whether the function itself is considered for inlining depends on its size and
1612 the current inlining parameters. The @code{flatten} attribute only works
1613 reliably in unit-at-a-time mode.
1616 @cindex functions that do pop the argument stack on the 386
1618 On the Intel 386, the @code{cdecl} attribute causes the compiler to
1619 assume that the calling function will pop off the stack space used to
1620 pass arguments. This is
1621 useful to override the effects of the @option{-mrtd} switch.
1624 @cindex @code{const} function attribute
1625 Many functions do not examine any values except their arguments, and
1626 have no effects except the return value. Basically this is just slightly
1627 more strict class than the @code{pure} attribute below, since function is not
1628 allowed to read global memory.
1630 @cindex pointer arguments
1631 Note that a function that has pointer arguments and examines the data
1632 pointed to must @emph{not} be declared @code{const}. Likewise, a
1633 function that calls a non-@code{const} function usually must not be
1634 @code{const}. It does not make sense for a @code{const} function to
1637 The attribute @code{const} is not implemented in GCC versions earlier
1638 than 2.5. An alternative way to declare that a function has no side
1639 effects, which works in the current version and in some older versions,
1643 typedef int intfn ();
1645 extern const intfn square;
1648 This approach does not work in GNU C++ from 2.6.0 on, since the language
1649 specifies that the @samp{const} must be attached to the return value.
1653 @cindex @code{constructor} function attribute
1654 @cindex @code{destructor} function attribute
1655 The @code{constructor} attribute causes the function to be called
1656 automatically before execution enters @code{main ()}. Similarly, the
1657 @code{destructor} attribute causes the function to be called
1658 automatically after @code{main ()} has completed or @code{exit ()} has
1659 been called. Functions with these attributes are useful for
1660 initializing data that will be used implicitly during the execution of
1663 These attributes are not currently implemented for Objective-C@.
1666 @cindex @code{deprecated} attribute.
1667 The @code{deprecated} attribute results in a warning if the function
1668 is used anywhere in the source file. This is useful when identifying
1669 functions that are expected to be removed in a future version of a
1670 program. The warning also includes the location of the declaration
1671 of the deprecated function, to enable users to easily find further
1672 information about why the function is deprecated, or what they should
1673 do instead. Note that the warnings only occurs for uses:
1676 int old_fn () __attribute__ ((deprecated));
1678 int (*fn_ptr)() = old_fn;
1681 results in a warning on line 3 but not line 2.
1683 The @code{deprecated} attribute can also be used for variables and
1684 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
1687 @cindex @code{__declspec(dllexport)}
1688 On Microsoft Windows targets and Symbian OS targets the
1689 @code{dllexport} attribute causes the compiler to provide a global
1690 pointer to a pointer in a DLL, so that it can be referenced with the
1691 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
1692 name is formed by combining @code{_imp__} and the function or variable
1695 You can use @code{__declspec(dllexport)} as a synonym for
1696 @code{__attribute__ ((dllexport))} for compatibility with other
1699 On systems that support the @code{visibility} attribute, this
1700 attribute also implies ``default'' visibility, unless a
1701 @code{visibility} attribute is explicitly specified. You should avoid
1702 the use of @code{dllexport} with ``hidden'' or ``internal''
1703 visibility; in the future GCC may issue an error for those cases.
1705 Currently, the @code{dllexport} attribute is ignored for inlined
1706 functions, unless the @option{-fkeep-inline-functions} flag has been
1707 used. The attribute is also ignored for undefined symbols.
1709 When applied to C++ classes, the attribute marks defined non-inlined
1710 member functions and static data members as exports. Static consts
1711 initialized in-class are not marked unless they are also defined
1714 For Microsoft Windows targets there are alternative methods for
1715 including the symbol in the DLL's export table such as using a
1716 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
1717 the @option{--export-all} linker flag.
1720 @cindex @code{__declspec(dllimport)}
1721 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
1722 attribute causes the compiler to reference a function or variable via
1723 a global pointer to a pointer that is set up by the DLL exporting the
1724 symbol. The attribute implies @code{extern} storage. On Microsoft
1725 Windows targets, the pointer name is formed by combining @code{_imp__}
1726 and the function or variable name.
1728 You can use @code{__declspec(dllimport)} as a synonym for
1729 @code{__attribute__ ((dllimport))} for compatibility with other
1732 Currently, the attribute is ignored for inlined functions. If the
1733 attribute is applied to a symbol @emph{definition}, an error is reported.
1734 If a symbol previously declared @code{dllimport} is later defined, the
1735 attribute is ignored in subsequent references, and a warning is emitted.
1736 The attribute is also overridden by a subsequent declaration as
1739 When applied to C++ classes, the attribute marks non-inlined
1740 member functions and static data members as imports. However, the
1741 attribute is ignored for virtual methods to allow creation of vtables
1744 On the SH Symbian OS target the @code{dllimport} attribute also has
1745 another affect---it can cause the vtable and run-time type information
1746 for a class to be exported. This happens when the class has a
1747 dllimport'ed constructor or a non-inline, non-pure virtual function
1748 and, for either of those two conditions, the class also has a inline
1749 constructor or destructor and has a key function that is defined in
1750 the current translation unit.
1752 For Microsoft Windows based targets the use of the @code{dllimport}
1753 attribute on functions is not necessary, but provides a small
1754 performance benefit by eliminating a thunk in the DLL@. The use of the
1755 @code{dllimport} attribute on imported variables was required on older
1756 versions of the GNU linker, but can now be avoided by passing the
1757 @option{--enable-auto-import} switch to the GNU linker. As with
1758 functions, using the attribute for a variable eliminates a thunk in
1761 One drawback to using this attribute is that a pointer to a function
1762 or variable marked as @code{dllimport} cannot be used as a constant
1763 address. On Microsoft Windows targets, the attribute can be disabled
1764 for functions by setting the @option{-mnop-fun-dllimport} flag.
1767 @cindex eight bit data on the H8/300, H8/300H, and H8S
1768 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1769 variable should be placed into the eight bit data section.
1770 The compiler will generate more efficient code for certain operations
1771 on data in the eight bit data area. Note the eight bit data area is limited to
1774 You must use GAS and GLD from GNU binutils version 2.7 or later for
1775 this attribute to work correctly.
1777 @item exception_handler
1778 @cindex exception handler functions on the Blackfin processor
1779 Use this attribute on the Blackfin to indicate that the specified function
1780 is an exception handler. The compiler will generate function entry and
1781 exit sequences suitable for use in an exception handler when this
1782 attribute is present.
1785 @cindex functions which handle memory bank switching
1786 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
1787 use a calling convention that takes care of switching memory banks when
1788 entering and leaving a function. This calling convention is also the
1789 default when using the @option{-mlong-calls} option.
1791 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
1792 to call and return from a function.
1794 On 68HC11 the compiler will generate a sequence of instructions
1795 to invoke a board-specific routine to switch the memory bank and call the
1796 real function. The board-specific routine simulates a @code{call}.
1797 At the end of a function, it will jump to a board-specific routine
1798 instead of using @code{rts}. The board-specific return routine simulates
1802 @cindex functions that pop the argument stack on the 386
1803 On the Intel 386, the @code{fastcall} attribute causes the compiler to
1804 pass the first argument (if of integral type) in the register ECX and
1805 the second argument (if of integral type) in the register EDX@. Subsequent
1806 and other typed arguments are passed on the stack. The called function will
1807 pop the arguments off the stack. If the number of arguments is variable all
1808 arguments are pushed on the stack.
1810 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
1811 @cindex @code{format} function attribute
1813 The @code{format} attribute specifies that a function takes @code{printf},
1814 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
1815 should be type-checked against a format string. For example, the
1820 my_printf (void *my_object, const char *my_format, ...)
1821 __attribute__ ((format (printf, 2, 3)));
1825 causes the compiler to check the arguments in calls to @code{my_printf}
1826 for consistency with the @code{printf} style format string argument
1829 The parameter @var{archetype} determines how the format string is
1830 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
1831 or @code{strfmon}. (You can also use @code{__printf__},
1832 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
1833 parameter @var{string-index} specifies which argument is the format
1834 string argument (starting from 1), while @var{first-to-check} is the
1835 number of the first argument to check against the format string. For
1836 functions where the arguments are not available to be checked (such as
1837 @code{vprintf}), specify the third parameter as zero. In this case the
1838 compiler only checks the format string for consistency. For
1839 @code{strftime} formats, the third parameter is required to be zero.
1840 Since non-static C++ methods have an implicit @code{this} argument, the
1841 arguments of such methods should be counted from two, not one, when
1842 giving values for @var{string-index} and @var{first-to-check}.
1844 In the example above, the format string (@code{my_format}) is the second
1845 argument of the function @code{my_print}, and the arguments to check
1846 start with the third argument, so the correct parameters for the format
1847 attribute are 2 and 3.
1849 @opindex ffreestanding
1850 @opindex fno-builtin
1851 The @code{format} attribute allows you to identify your own functions
1852 which take format strings as arguments, so that GCC can check the
1853 calls to these functions for errors. The compiler always (unless
1854 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
1855 for the standard library functions @code{printf}, @code{fprintf},
1856 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
1857 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
1858 warnings are requested (using @option{-Wformat}), so there is no need to
1859 modify the header file @file{stdio.h}. In C99 mode, the functions
1860 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
1861 @code{vsscanf} are also checked. Except in strictly conforming C
1862 standard modes, the X/Open function @code{strfmon} is also checked as
1863 are @code{printf_unlocked} and @code{fprintf_unlocked}.
1864 @xref{C Dialect Options,,Options Controlling C Dialect}.
1866 The target may provide additional types of format checks.
1867 @xref{Target Format Checks,,Format Checks Specific to Particular
1870 @item format_arg (@var{string-index})
1871 @cindex @code{format_arg} function attribute
1872 @opindex Wformat-nonliteral
1873 The @code{format_arg} attribute specifies that a function takes a format
1874 string for a @code{printf}, @code{scanf}, @code{strftime} or
1875 @code{strfmon} style function and modifies it (for example, to translate
1876 it into another language), so the result can be passed to a
1877 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
1878 function (with the remaining arguments to the format function the same
1879 as they would have been for the unmodified string). For example, the
1884 my_dgettext (char *my_domain, const char *my_format)
1885 __attribute__ ((format_arg (2)));
1889 causes the compiler to check the arguments in calls to a @code{printf},
1890 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
1891 format string argument is a call to the @code{my_dgettext} function, for
1892 consistency with the format string argument @code{my_format}. If the
1893 @code{format_arg} attribute had not been specified, all the compiler
1894 could tell in such calls to format functions would be that the format
1895 string argument is not constant; this would generate a warning when
1896 @option{-Wformat-nonliteral} is used, but the calls could not be checked
1897 without the attribute.
1899 The parameter @var{string-index} specifies which argument is the format
1900 string argument (starting from one). Since non-static C++ methods have
1901 an implicit @code{this} argument, the arguments of such methods should
1902 be counted from two.
1904 The @code{format-arg} attribute allows you to identify your own
1905 functions which modify format strings, so that GCC can check the
1906 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
1907 type function whose operands are a call to one of your own function.
1908 The compiler always treats @code{gettext}, @code{dgettext}, and
1909 @code{dcgettext} in this manner except when strict ISO C support is
1910 requested by @option{-ansi} or an appropriate @option{-std} option, or
1911 @option{-ffreestanding} or @option{-fno-builtin}
1912 is used. @xref{C Dialect Options,,Options
1913 Controlling C Dialect}.
1915 @item function_vector
1916 @cindex calling functions through the function vector on the H8/300 processors
1917 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1918 function should be called through the function vector. Calling a
1919 function through the function vector will reduce code size, however;
1920 the function vector has a limited size (maximum 128 entries on the H8/300
1921 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
1923 You must use GAS and GLD from GNU binutils version 2.7 or later for
1924 this attribute to work correctly.
1927 @cindex interrupt handler functions
1928 Use this attribute on the ARM, AVR, C4x, CRX, M32C, M32R/D, MS1, and Xstormy16
1929 ports to indicate that the specified function is an interrupt handler.
1930 The compiler will generate function entry and exit sequences suitable
1931 for use in an interrupt handler when this attribute is present.
1933 Note, interrupt handlers for the Blackfin, m68k, H8/300, H8/300H, H8S, and
1934 SH processors can be specified via the @code{interrupt_handler} attribute.
1936 Note, on the AVR, interrupts will be enabled inside the function.
1938 Note, for the ARM, you can specify the kind of interrupt to be handled by
1939 adding an optional parameter to the interrupt attribute like this:
1942 void f () __attribute__ ((interrupt ("IRQ")));
1945 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
1947 @item interrupt_handler
1948 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
1949 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
1950 indicate that the specified function is an interrupt handler. The compiler
1951 will generate function entry and exit sequences suitable for use in an
1952 interrupt handler when this attribute is present.
1955 @cindex User stack pointer in interrupts on the Blackfin
1956 When used together with @code{interrupt_handler}, @code{exception_handler}
1957 or @code{nmi_handler}, code will be generated to load the stack pointer
1958 from the USP register in the function prologue.
1960 @item long_call/short_call
1961 @cindex indirect calls on ARM
1962 This attribute specifies how a particular function is called on
1963 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
1964 command line switch and @code{#pragma long_calls} settings. The
1965 @code{long_call} attribute causes the compiler to always call the
1966 function by first loading its address into a register and then using the
1967 contents of that register. The @code{short_call} attribute always places
1968 the offset to the function from the call site into the @samp{BL}
1969 instruction directly.
1971 @item longcall/shortcall
1972 @cindex functions called via pointer on the RS/6000 and PowerPC
1973 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute causes
1974 the compiler to always call this function via a pointer, just as it would if
1975 the @option{-mlongcall} option had been specified. The @code{shortcall}
1976 attribute causes the compiler not to do this. These attributes override
1977 both the @option{-mlongcall} switch and, on the RS/6000 and PowerPC, the
1978 @code{#pragma longcall} setting.
1980 @xref{RS/6000 and PowerPC Options}, for more information on whether long
1981 calls are necessary.
1984 @cindex indirect calls on MIPS
1985 This attribute specifies how a particular function is called on MIPS@.
1986 The attribute overrides the @option{-mlong-calls} (@pxref{MIPS Options})
1987 command line switch. This attribute causes the compiler to always call
1988 the function by first loading its address into a register, and then using
1989 the contents of that register.
1992 @cindex @code{malloc} attribute
1993 The @code{malloc} attribute is used to tell the compiler that a function
1994 may be treated as if any non-@code{NULL} pointer it returns cannot
1995 alias any other pointer valid when the function returns.
1996 This will often improve optimization.
1997 Standard functions with this property include @code{malloc} and
1998 @code{calloc}. @code{realloc}-like functions have this property as
1999 long as the old pointer is never referred to (including comparing it
2000 to the new pointer) after the function returns a non-@code{NULL}
2003 @item model (@var{model-name})
2004 @cindex function addressability on the M32R/D
2005 @cindex variable addressability on the IA-64
2007 On the M32R/D, use this attribute to set the addressability of an
2008 object, and of the code generated for a function. The identifier
2009 @var{model-name} is one of @code{small}, @code{medium}, or
2010 @code{large}, representing each of the code models.
2012 Small model objects live in the lower 16MB of memory (so that their
2013 addresses can be loaded with the @code{ld24} instruction), and are
2014 callable with the @code{bl} instruction.
2016 Medium model objects may live anywhere in the 32-bit address space (the
2017 compiler will generate @code{seth/add3} instructions to load their addresses),
2018 and are callable with the @code{bl} instruction.
2020 Large model objects may live anywhere in the 32-bit address space (the
2021 compiler will generate @code{seth/add3} instructions to load their addresses),
2022 and may not be reachable with the @code{bl} instruction (the compiler will
2023 generate the much slower @code{seth/add3/jl} instruction sequence).
2025 On IA-64, use this attribute to set the addressability of an object.
2026 At present, the only supported identifier for @var{model-name} is
2027 @code{small}, indicating addressability via ``small'' (22-bit)
2028 addresses (so that their addresses can be loaded with the @code{addl}
2029 instruction). Caveat: such addressing is by definition not position
2030 independent and hence this attribute must not be used for objects
2031 defined by shared libraries.
2034 @cindex function without a prologue/epilogue code
2035 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate that the
2036 specified function does not need prologue/epilogue sequences generated by
2037 the compiler. It is up to the programmer to provide these sequences.
2040 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2041 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2042 use the normal calling convention based on @code{jsr} and @code{rts}.
2043 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2047 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2048 Use this attribute together with @code{interrupt_handler},
2049 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2050 entry code should enable nested interrupts or exceptions.
2053 @cindex NMI handler functions on the Blackfin processor
2054 Use this attribute on the Blackfin to indicate that the specified function
2055 is an NMI handler. The compiler will generate function entry and
2056 exit sequences suitable for use in an NMI handler when this
2057 attribute is present.
2059 @item no_instrument_function
2060 @cindex @code{no_instrument_function} function attribute
2061 @opindex finstrument-functions
2062 If @option{-finstrument-functions} is given, profiling function calls will
2063 be generated at entry and exit of most user-compiled functions.
2064 Functions with this attribute will not be so instrumented.
2067 @cindex @code{noinline} function attribute
2068 This function attribute prevents a function from being considered for
2071 @item nonnull (@var{arg-index}, @dots{})
2072 @cindex @code{nonnull} function attribute
2073 The @code{nonnull} attribute specifies that some function parameters should
2074 be non-null pointers. For instance, the declaration:
2078 my_memcpy (void *dest, const void *src, size_t len)
2079 __attribute__((nonnull (1, 2)));
2083 causes the compiler to check that, in calls to @code{my_memcpy},
2084 arguments @var{dest} and @var{src} are non-null. If the compiler
2085 determines that a null pointer is passed in an argument slot marked
2086 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2087 is issued. The compiler may also choose to make optimizations based
2088 on the knowledge that certain function arguments will not be null.
2090 If no argument index list is given to the @code{nonnull} attribute,
2091 all pointer arguments are marked as non-null. To illustrate, the
2092 following declaration is equivalent to the previous example:
2096 my_memcpy (void *dest, const void *src, size_t len)
2097 __attribute__((nonnull));
2101 @cindex @code{noreturn} function attribute
2102 A few standard library functions, such as @code{abort} and @code{exit},
2103 cannot return. GCC knows this automatically. Some programs define
2104 their own functions that never return. You can declare them
2105 @code{noreturn} to tell the compiler this fact. For example,
2109 void fatal () __attribute__ ((noreturn));
2112 fatal (/* @r{@dots{}} */)
2114 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2120 The @code{noreturn} keyword tells the compiler to assume that
2121 @code{fatal} cannot return. It can then optimize without regard to what
2122 would happen if @code{fatal} ever did return. This makes slightly
2123 better code. More importantly, it helps avoid spurious warnings of
2124 uninitialized variables.
2126 The @code{noreturn} keyword does not affect the exceptional path when that
2127 applies: a @code{noreturn}-marked function may still return to the caller
2128 by throwing an exception or calling @code{longjmp}.
2130 Do not assume that registers saved by the calling function are
2131 restored before calling the @code{noreturn} function.
2133 It does not make sense for a @code{noreturn} function to have a return
2134 type other than @code{void}.
2136 The attribute @code{noreturn} is not implemented in GCC versions
2137 earlier than 2.5. An alternative way to declare that a function does
2138 not return, which works in the current version and in some older
2139 versions, is as follows:
2142 typedef void voidfn ();
2144 volatile voidfn fatal;
2147 This approach does not work in GNU C++.
2150 @cindex @code{nothrow} function attribute
2151 The @code{nothrow} attribute is used to inform the compiler that a
2152 function cannot throw an exception. For example, most functions in
2153 the standard C library can be guaranteed not to throw an exception
2154 with the notable exceptions of @code{qsort} and @code{bsearch} that
2155 take function pointer arguments. The @code{nothrow} attribute is not
2156 implemented in GCC versions earlier than 3.3.
2159 @cindex @code{pure} function attribute
2160 Many functions have no effects except the return value and their
2161 return value depends only on the parameters and/or global variables.
2162 Such a function can be subject
2163 to common subexpression elimination and loop optimization just as an
2164 arithmetic operator would be. These functions should be declared
2165 with the attribute @code{pure}. For example,
2168 int square (int) __attribute__ ((pure));
2172 says that the hypothetical function @code{square} is safe to call
2173 fewer times than the program says.
2175 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2176 Interesting non-pure functions are functions with infinite loops or those
2177 depending on volatile memory or other system resource, that may change between
2178 two consecutive calls (such as @code{feof} in a multithreading environment).
2180 The attribute @code{pure} is not implemented in GCC versions earlier
2183 @item regparm (@var{number})
2184 @cindex @code{regparm} attribute
2185 @cindex functions that are passed arguments in registers on the 386
2186 On the Intel 386, the @code{regparm} attribute causes the compiler to
2187 pass arguments number one to @var{number} if they are of integral type
2188 in registers EAX, EDX, and ECX instead of on the stack. Functions that
2189 take a variable number of arguments will continue to be passed all of their
2190 arguments on the stack.
2192 Beware that on some ELF systems this attribute is unsuitable for
2193 global functions in shared libraries with lazy binding (which is the
2194 default). Lazy binding will send the first call via resolving code in
2195 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2196 per the standard calling conventions. Solaris 8 is affected by this.
2197 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2198 safe since the loaders there save all registers. (Lazy binding can be
2199 disabled with the linker or the loader if desired, to avoid the
2203 @cindex @code{sseregparm} attribute
2204 On the Intel 386 with SSE support, the @code{sseregparm} attribute
2205 causes the compiler to pass up to 8 floating point arguments in
2206 SSE registers instead of on the stack. Functions that take a
2207 variable number of arguments will continue to pass all of their
2208 floating point arguments on the stack.
2211 @cindex @code{returns_twice} attribute
2212 The @code{returns_twice} attribute tells the compiler that a function may
2213 return more than one time. The compiler will ensure that all registers
2214 are dead before calling such a function and will emit a warning about
2215 the variables that may be clobbered after the second return from the
2216 function. Examples of such functions are @code{setjmp} and @code{vfork}.
2217 The @code{longjmp}-like counterpart of such function, if any, might need
2218 to be marked with the @code{noreturn} attribute.
2221 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
2222 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
2223 all registers except the stack pointer should be saved in the prologue
2224 regardless of whether they are used or not.
2226 @item section ("@var{section-name}")
2227 @cindex @code{section} function attribute
2228 Normally, the compiler places the code it generates in the @code{text} section.
2229 Sometimes, however, you need additional sections, or you need certain
2230 particular functions to appear in special sections. The @code{section}
2231 attribute specifies that a function lives in a particular section.
2232 For example, the declaration:
2235 extern void foobar (void) __attribute__ ((section ("bar")));
2239 puts the function @code{foobar} in the @code{bar} section.
2241 Some file formats do not support arbitrary sections so the @code{section}
2242 attribute is not available on all platforms.
2243 If you need to map the entire contents of a module to a particular
2244 section, consider using the facilities of the linker instead.
2247 @cindex @code{sentinel} function attribute
2248 This function attribute ensures that a parameter in a function call is
2249 an explicit @code{NULL}. The attribute is only valid on variadic
2250 functions. By default, the sentinel is located at position zero, the
2251 last parameter of the function call. If an optional integer position
2252 argument P is supplied to the attribute, the sentinel must be located at
2253 position P counting backwards from the end of the argument list.
2256 __attribute__ ((sentinel))
2258 __attribute__ ((sentinel(0)))
2261 The attribute is automatically set with a position of 0 for the built-in
2262 functions @code{execl} and @code{execlp}. The built-in function
2263 @code{execle} has the attribute set with a position of 1.
2265 A valid @code{NULL} in this context is defined as zero with any pointer
2266 type. If your system defines the @code{NULL} macro with an integer type
2267 then you need to add an explicit cast. GCC replaces @code{stddef.h}
2268 with a copy that redefines NULL appropriately.
2270 The warnings for missing or incorrect sentinels are enabled with
2274 See long_call/short_call.
2277 See longcall/shortcall.
2280 @cindex signal handler functions on the AVR processors
2281 Use this attribute on the AVR to indicate that the specified
2282 function is a signal handler. The compiler will generate function
2283 entry and exit sequences suitable for use in a signal handler when this
2284 attribute is present. Interrupts will be disabled inside the function.
2287 Use this attribute on the SH to indicate an @code{interrupt_handler}
2288 function should switch to an alternate stack. It expects a string
2289 argument that names a global variable holding the address of the
2294 void f () __attribute__ ((interrupt_handler,
2295 sp_switch ("alt_stack")));
2299 @cindex functions that pop the argument stack on the 386
2300 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2301 assume that the called function will pop off the stack space used to
2302 pass arguments, unless it takes a variable number of arguments.
2305 @cindex tiny data section on the H8/300H and H8S
2306 Use this attribute on the H8/300H and H8S to indicate that the specified
2307 variable should be placed into the tiny data section.
2308 The compiler will generate more efficient code for loads and stores
2309 on data in the tiny data section. Note the tiny data area is limited to
2310 slightly under 32kbytes of data.
2313 Use this attribute on the SH for an @code{interrupt_handler} to return using
2314 @code{trapa} instead of @code{rte}. This attribute expects an integer
2315 argument specifying the trap number to be used.
2318 @cindex @code{unused} attribute.
2319 This attribute, attached to a function, means that the function is meant
2320 to be possibly unused. GCC will not produce a warning for this
2324 @cindex @code{used} attribute.
2325 This attribute, attached to a function, means that code must be emitted
2326 for the function even if it appears that the function is not referenced.
2327 This is useful, for example, when the function is referenced only in
2330 @item visibility ("@var{visibility_type}")
2331 @cindex @code{visibility} attribute
2332 The @code{visibility} attribute on ELF targets causes the declaration
2333 to be emitted with default, hidden, protected or internal visibility.
2336 void __attribute__ ((visibility ("protected")))
2337 f () @{ /* @r{Do something.} */; @}
2338 int i __attribute__ ((visibility ("hidden")));
2341 See the ELF gABI for complete details, but the short story is:
2344 @c keep this list of visibilities in alphabetical order.
2347 Default visibility is the normal case for ELF@. This value is
2348 available for the visibility attribute to override other options
2349 that may change the assumed visibility of symbols.
2352 Hidden visibility indicates that the symbol will not be placed into
2353 the dynamic symbol table, so no other @dfn{module} (executable or
2354 shared library) can reference it directly.
2357 Internal visibility is like hidden visibility, but with additional
2358 processor specific semantics. Unless otherwise specified by the psABI,
2359 GCC defines internal visibility to mean that the function is @emph{never}
2360 called from another module. Note that hidden symbols, while they cannot
2361 be referenced directly by other modules, can be referenced indirectly via
2362 function pointers. By indicating that a symbol cannot be called from
2363 outside the module, GCC may for instance omit the load of a PIC register
2364 since it is known that the calling function loaded the correct value.
2367 Protected visibility indicates that the symbol will be placed in the
2368 dynamic symbol table, but that references within the defining module
2369 will bind to the local symbol. That is, the symbol cannot be overridden
2374 Not all ELF targets support this attribute.
2376 @item warn_unused_result
2377 @cindex @code{warn_unused_result} attribute
2378 The @code{warn_unused_result} attribute causes a warning to be emitted
2379 if a caller of the function with this attribute does not use its
2380 return value. This is useful for functions where not checking
2381 the result is either a security problem or always a bug, such as
2385 int fn () __attribute__ ((warn_unused_result));
2388 if (fn () < 0) return -1;
2394 results in warning on line 5.
2397 @cindex @code{weak} attribute
2398 The @code{weak} attribute causes the declaration to be emitted as a weak
2399 symbol rather than a global. This is primarily useful in defining
2400 library functions which can be overridden in user code, though it can
2401 also be used with non-function declarations. Weak symbols are supported
2402 for ELF targets, and also for a.out targets when using the GNU assembler
2406 @itemx weakref ("@var{target}")
2407 @cindex @code{weakref} attribute
2408 The @code{weakref} attribute marks a declaration as a weak reference.
2409 Without arguments, it should be accompanied by an @code{alias} attribute
2410 naming the target symbol. Optionally, the @var{target} may be given as
2411 an argument to @code{weakref} itself. In either case, @code{weakref}
2412 implicitly marks the declaration as @code{weak}. Without a
2413 @var{target}, given as an argument to @code{weakref} or to @code{alias},
2414 @code{weakref} is equivalent to @code{weak}.
2417 static int x() __attribute__ ((weakref ("y")));
2418 /* is equivalent to... */
2419 static int x() __attribute__ ((weak, weakref, alias ("y")));
2421 static int x() __attribute__ ((weakref));
2422 static int x() __attribute__ ((alias ("y")));
2425 A weak reference is an alias that does not by itself require a
2426 definition to be given for the target symbol. If the target symbol is
2427 only referenced through weak references, then the becomes a @code{weak}
2428 undefined symbol. If it is directly referenced, however, then such
2429 strong references prevail, and a definition will be required for the
2430 symbol, not necessarily in the same translation unit.
2432 The effect is equivalent to moving all references to the alias to a
2433 separate translation unit, renaming the alias to the aliased symbol,
2434 declaring it as weak, compiling the two separate translation units and
2435 performing a reloadable link on them.
2437 At present, a declaration to which @code{weakref} is attached can
2438 only be @code{static}.
2440 @item externally_visible
2441 @cindex @code{externally_visible} attribute.
2442 This attribute, attached to a global variable or function nullify
2443 effect of @option{-fwhole-program} command line option, so the object
2444 remain visible outside the current compilation unit
2448 You can specify multiple attributes in a declaration by separating them
2449 by commas within the double parentheses or by immediately following an
2450 attribute declaration with another attribute declaration.
2452 @cindex @code{#pragma}, reason for not using
2453 @cindex pragma, reason for not using
2454 Some people object to the @code{__attribute__} feature, suggesting that
2455 ISO C's @code{#pragma} should be used instead. At the time
2456 @code{__attribute__} was designed, there were two reasons for not doing
2461 It is impossible to generate @code{#pragma} commands from a macro.
2464 There is no telling what the same @code{#pragma} might mean in another
2468 These two reasons applied to almost any application that might have been
2469 proposed for @code{#pragma}. It was basically a mistake to use
2470 @code{#pragma} for @emph{anything}.
2472 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2473 to be generated from macros. In addition, a @code{#pragma GCC}
2474 namespace is now in use for GCC-specific pragmas. However, it has been
2475 found convenient to use @code{__attribute__} to achieve a natural
2476 attachment of attributes to their corresponding declarations, whereas
2477 @code{#pragma GCC} is of use for constructs that do not naturally form
2478 part of the grammar. @xref{Other Directives,,Miscellaneous
2479 Preprocessing Directives, cpp, The GNU C Preprocessor}.
2481 @node Attribute Syntax
2482 @section Attribute Syntax
2483 @cindex attribute syntax
2485 This section describes the syntax with which @code{__attribute__} may be
2486 used, and the constructs to which attribute specifiers bind, for the C
2487 language. Some details may vary for C++ and Objective-C@. Because of
2488 infelicities in the grammar for attributes, some forms described here
2489 may not be successfully parsed in all cases.
2491 There are some problems with the semantics of attributes in C++. For
2492 example, there are no manglings for attributes, although they may affect
2493 code generation, so problems may arise when attributed types are used in
2494 conjunction with templates or overloading. Similarly, @code{typeid}
2495 does not distinguish between types with different attributes. Support
2496 for attributes in C++ may be restricted in future to attributes on
2497 declarations only, but not on nested declarators.
2499 @xref{Function Attributes}, for details of the semantics of attributes
2500 applying to functions. @xref{Variable Attributes}, for details of the
2501 semantics of attributes applying to variables. @xref{Type Attributes},
2502 for details of the semantics of attributes applying to structure, union
2503 and enumerated types.
2505 An @dfn{attribute specifier} is of the form
2506 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2507 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2508 each attribute is one of the following:
2512 Empty. Empty attributes are ignored.
2515 A word (which may be an identifier such as @code{unused}, or a reserved
2516 word such as @code{const}).
2519 A word, followed by, in parentheses, parameters for the attribute.
2520 These parameters take one of the following forms:
2524 An identifier. For example, @code{mode} attributes use this form.
2527 An identifier followed by a comma and a non-empty comma-separated list
2528 of expressions. For example, @code{format} attributes use this form.
2531 A possibly empty comma-separated list of expressions. For example,
2532 @code{format_arg} attributes use this form with the list being a single
2533 integer constant expression, and @code{alias} attributes use this form
2534 with the list being a single string constant.
2538 An @dfn{attribute specifier list} is a sequence of one or more attribute
2539 specifiers, not separated by any other tokens.
2541 In GNU C, an attribute specifier list may appear after the colon following a
2542 label, other than a @code{case} or @code{default} label. The only
2543 attribute it makes sense to use after a label is @code{unused}. This
2544 feature is intended for code generated by programs which contains labels
2545 that may be unused but which is compiled with @option{-Wall}. It would
2546 not normally be appropriate to use in it human-written code, though it
2547 could be useful in cases where the code that jumps to the label is
2548 contained within an @code{#ifdef} conditional. GNU C++ does not permit
2549 such placement of attribute lists, as it is permissible for a
2550 declaration, which could begin with an attribute list, to be labelled in
2551 C++. Declarations cannot be labelled in C90 or C99, so the ambiguity
2552 does not arise there.
2554 An attribute specifier list may appear as part of a @code{struct},
2555 @code{union} or @code{enum} specifier. It may go either immediately
2556 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2557 the closing brace. It is ignored if the content of the structure, union
2558 or enumerated type is not defined in the specifier in which the
2559 attribute specifier list is used---that is, in usages such as
2560 @code{struct __attribute__((foo)) bar} with no following opening brace.
2561 Where attribute specifiers follow the closing brace, they are considered
2562 to relate to the structure, union or enumerated type defined, not to any
2563 enclosing declaration the type specifier appears in, and the type
2564 defined is not complete until after the attribute specifiers.
2565 @c Otherwise, there would be the following problems: a shift/reduce
2566 @c conflict between attributes binding the struct/union/enum and
2567 @c binding to the list of specifiers/qualifiers; and "aligned"
2568 @c attributes could use sizeof for the structure, but the size could be
2569 @c changed later by "packed" attributes.
2571 Otherwise, an attribute specifier appears as part of a declaration,
2572 counting declarations of unnamed parameters and type names, and relates
2573 to that declaration (which may be nested in another declaration, for
2574 example in the case of a parameter declaration), or to a particular declarator
2575 within a declaration. Where an
2576 attribute specifier is applied to a parameter declared as a function or
2577 an array, it should apply to the function or array rather than the
2578 pointer to which the parameter is implicitly converted, but this is not
2579 yet correctly implemented.
2581 Any list of specifiers and qualifiers at the start of a declaration may
2582 contain attribute specifiers, whether or not such a list may in that
2583 context contain storage class specifiers. (Some attributes, however,
2584 are essentially in the nature of storage class specifiers, and only make
2585 sense where storage class specifiers may be used; for example,
2586 @code{section}.) There is one necessary limitation to this syntax: the
2587 first old-style parameter declaration in a function definition cannot
2588 begin with an attribute specifier, because such an attribute applies to
2589 the function instead by syntax described below (which, however, is not
2590 yet implemented in this case). In some other cases, attribute
2591 specifiers are permitted by this grammar but not yet supported by the
2592 compiler. All attribute specifiers in this place relate to the
2593 declaration as a whole. In the obsolescent usage where a type of
2594 @code{int} is implied by the absence of type specifiers, such a list of
2595 specifiers and qualifiers may be an attribute specifier list with no
2596 other specifiers or qualifiers.
2598 At present, the first parameter in a function prototype must have some
2599 type specifier which is not an attribute specifier; this resolves an
2600 ambiguity in the interpretation of @code{void f(int
2601 (__attribute__((foo)) x))}, but is subject to change. At present, if
2602 the parentheses of a function declarator contain only attributes then
2603 those attributes are ignored, rather than yielding an error or warning
2604 or implying a single parameter of type int, but this is subject to
2607 An attribute specifier list may appear immediately before a declarator
2608 (other than the first) in a comma-separated list of declarators in a
2609 declaration of more than one identifier using a single list of
2610 specifiers and qualifiers. Such attribute specifiers apply
2611 only to the identifier before whose declarator they appear. For
2615 __attribute__((noreturn)) void d0 (void),
2616 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2621 the @code{noreturn} attribute applies to all the functions
2622 declared; the @code{format} attribute only applies to @code{d1}.
2624 An attribute specifier list may appear immediately before the comma,
2625 @code{=} or semicolon terminating the declaration of an identifier other
2626 than a function definition. At present, such attribute specifiers apply
2627 to the declared object or function, but in future they may attach to the
2628 outermost adjacent declarator. In simple cases there is no difference,
2629 but, for example, in
2632 void (****f)(void) __attribute__((noreturn));
2636 at present the @code{noreturn} attribute applies to @code{f}, which
2637 causes a warning since @code{f} is not a function, but in future it may
2638 apply to the function @code{****f}. The precise semantics of what
2639 attributes in such cases will apply to are not yet specified. Where an
2640 assembler name for an object or function is specified (@pxref{Asm
2641 Labels}), at present the attribute must follow the @code{asm}
2642 specification; in future, attributes before the @code{asm} specification
2643 may apply to the adjacent declarator, and those after it to the declared
2646 An attribute specifier list may, in future, be permitted to appear after
2647 the declarator in a function definition (before any old-style parameter
2648 declarations or the function body).
2650 Attribute specifiers may be mixed with type qualifiers appearing inside
2651 the @code{[]} of a parameter array declarator, in the C99 construct by
2652 which such qualifiers are applied to the pointer to which the array is
2653 implicitly converted. Such attribute specifiers apply to the pointer,
2654 not to the array, but at present this is not implemented and they are
2657 An attribute specifier list may appear at the start of a nested
2658 declarator. At present, there are some limitations in this usage: the
2659 attributes correctly apply to the declarator, but for most individual
2660 attributes the semantics this implies are not implemented.
2661 When attribute specifiers follow the @code{*} of a pointer
2662 declarator, they may be mixed with any type qualifiers present.
2663 The following describes the formal semantics of this syntax. It will make the
2664 most sense if you are familiar with the formal specification of
2665 declarators in the ISO C standard.
2667 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2668 D1}, where @code{T} contains declaration specifiers that specify a type
2669 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2670 contains an identifier @var{ident}. The type specified for @var{ident}
2671 for derived declarators whose type does not include an attribute
2672 specifier is as in the ISO C standard.
2674 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2675 and the declaration @code{T D} specifies the type
2676 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2677 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2678 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2680 If @code{D1} has the form @code{*
2681 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2682 declaration @code{T D} specifies the type
2683 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2684 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2685 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2691 void (__attribute__((noreturn)) ****f) (void);
2695 specifies the type ``pointer to pointer to pointer to pointer to
2696 non-returning function returning @code{void}''. As another example,
2699 char *__attribute__((aligned(8))) *f;
2703 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2704 Note again that this does not work with most attributes; for example,
2705 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2706 is not yet supported.
2708 For compatibility with existing code written for compiler versions that
2709 did not implement attributes on nested declarators, some laxity is
2710 allowed in the placing of attributes. If an attribute that only applies
2711 to types is applied to a declaration, it will be treated as applying to
2712 the type of that declaration. If an attribute that only applies to
2713 declarations is applied to the type of a declaration, it will be treated
2714 as applying to that declaration; and, for compatibility with code
2715 placing the attributes immediately before the identifier declared, such
2716 an attribute applied to a function return type will be treated as
2717 applying to the function type, and such an attribute applied to an array
2718 element type will be treated as applying to the array type. If an
2719 attribute that only applies to function types is applied to a
2720 pointer-to-function type, it will be treated as applying to the pointer
2721 target type; if such an attribute is applied to a function return type
2722 that is not a pointer-to-function type, it will be treated as applying
2723 to the function type.
2725 @node Function Prototypes
2726 @section Prototypes and Old-Style Function Definitions
2727 @cindex function prototype declarations
2728 @cindex old-style function definitions
2729 @cindex promotion of formal parameters
2731 GNU C extends ISO C to allow a function prototype to override a later
2732 old-style non-prototype definition. Consider the following example:
2735 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2742 /* @r{Prototype function declaration.} */
2743 int isroot P((uid_t));
2745 /* @r{Old-style function definition.} */
2747 isroot (x) /* @r{??? lossage here ???} */
2754 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2755 not allow this example, because subword arguments in old-style
2756 non-prototype definitions are promoted. Therefore in this example the
2757 function definition's argument is really an @code{int}, which does not
2758 match the prototype argument type of @code{short}.
2760 This restriction of ISO C makes it hard to write code that is portable
2761 to traditional C compilers, because the programmer does not know
2762 whether the @code{uid_t} type is @code{short}, @code{int}, or
2763 @code{long}. Therefore, in cases like these GNU C allows a prototype
2764 to override a later old-style definition. More precisely, in GNU C, a
2765 function prototype argument type overrides the argument type specified
2766 by a later old-style definition if the former type is the same as the
2767 latter type before promotion. Thus in GNU C the above example is
2768 equivalent to the following:
2781 GNU C++ does not support old-style function definitions, so this
2782 extension is irrelevant.
2785 @section C++ Style Comments
2787 @cindex C++ comments
2788 @cindex comments, C++ style
2790 In GNU C, you may use C++ style comments, which start with @samp{//} and
2791 continue until the end of the line. Many other C implementations allow
2792 such comments, and they are included in the 1999 C standard. However,
2793 C++ style comments are not recognized if you specify an @option{-std}
2794 option specifying a version of ISO C before C99, or @option{-ansi}
2795 (equivalent to @option{-std=c89}).
2798 @section Dollar Signs in Identifier Names
2800 @cindex dollar signs in identifier names
2801 @cindex identifier names, dollar signs in
2803 In GNU C, you may normally use dollar signs in identifier names.
2804 This is because many traditional C implementations allow such identifiers.
2805 However, dollar signs in identifiers are not supported on a few target
2806 machines, typically because the target assembler does not allow them.
2808 @node Character Escapes
2809 @section The Character @key{ESC} in Constants
2811 You can use the sequence @samp{\e} in a string or character constant to
2812 stand for the ASCII character @key{ESC}.
2815 @section Inquiring on Alignment of Types or Variables
2817 @cindex type alignment
2818 @cindex variable alignment
2820 The keyword @code{__alignof__} allows you to inquire about how an object
2821 is aligned, or the minimum alignment usually required by a type. Its
2822 syntax is just like @code{sizeof}.
2824 For example, if the target machine requires a @code{double} value to be
2825 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2826 This is true on many RISC machines. On more traditional machine
2827 designs, @code{__alignof__ (double)} is 4 or even 2.
2829 Some machines never actually require alignment; they allow reference to any
2830 data type even at an odd address. For these machines, @code{__alignof__}
2831 reports the @emph{recommended} alignment of a type.
2833 If the operand of @code{__alignof__} is an lvalue rather than a type,
2834 its value is the required alignment for its type, taking into account
2835 any minimum alignment specified with GCC's @code{__attribute__}
2836 extension (@pxref{Variable Attributes}). For example, after this
2840 struct foo @{ int x; char y; @} foo1;
2844 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
2845 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
2847 It is an error to ask for the alignment of an incomplete type.
2849 @node Variable Attributes
2850 @section Specifying Attributes of Variables
2851 @cindex attribute of variables
2852 @cindex variable attributes
2854 The keyword @code{__attribute__} allows you to specify special
2855 attributes of variables or structure fields. This keyword is followed
2856 by an attribute specification inside double parentheses. Some
2857 attributes are currently defined generically for variables.
2858 Other attributes are defined for variables on particular target
2859 systems. Other attributes are available for functions
2860 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
2861 Other front ends might define more attributes
2862 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2864 You may also specify attributes with @samp{__} preceding and following
2865 each keyword. This allows you to use them in header files without
2866 being concerned about a possible macro of the same name. For example,
2867 you may use @code{__aligned__} instead of @code{aligned}.
2869 @xref{Attribute Syntax}, for details of the exact syntax for using
2873 @cindex @code{aligned} attribute
2874 @item aligned (@var{alignment})
2875 This attribute specifies a minimum alignment for the variable or
2876 structure field, measured in bytes. For example, the declaration:
2879 int x __attribute__ ((aligned (16))) = 0;
2883 causes the compiler to allocate the global variable @code{x} on a
2884 16-byte boundary. On a 68040, this could be used in conjunction with
2885 an @code{asm} expression to access the @code{move16} instruction which
2886 requires 16-byte aligned operands.
2888 You can also specify the alignment of structure fields. For example, to
2889 create a double-word aligned @code{int} pair, you could write:
2892 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2896 This is an alternative to creating a union with a @code{double} member
2897 that forces the union to be double-word aligned.
2899 As in the preceding examples, you can explicitly specify the alignment
2900 (in bytes) that you wish the compiler to use for a given variable or
2901 structure field. Alternatively, you can leave out the alignment factor
2902 and just ask the compiler to align a variable or field to the maximum
2903 useful alignment for the target machine you are compiling for. For
2904 example, you could write:
2907 short array[3] __attribute__ ((aligned));
2910 Whenever you leave out the alignment factor in an @code{aligned} attribute
2911 specification, the compiler automatically sets the alignment for the declared
2912 variable or field to the largest alignment which is ever used for any data
2913 type on the target machine you are compiling for. Doing this can often make
2914 copy operations more efficient, because the compiler can use whatever
2915 instructions copy the biggest chunks of memory when performing copies to
2916 or from the variables or fields that you have aligned this way.
2918 The @code{aligned} attribute can only increase the alignment; but you
2919 can decrease it by specifying @code{packed} as well. See below.
2921 Note that the effectiveness of @code{aligned} attributes may be limited
2922 by inherent limitations in your linker. On many systems, the linker is
2923 only able to arrange for variables to be aligned up to a certain maximum
2924 alignment. (For some linkers, the maximum supported alignment may
2925 be very very small.) If your linker is only able to align variables
2926 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2927 in an @code{__attribute__} will still only provide you with 8 byte
2928 alignment. See your linker documentation for further information.
2930 @item cleanup (@var{cleanup_function})
2931 @cindex @code{cleanup} attribute
2932 The @code{cleanup} attribute runs a function when the variable goes
2933 out of scope. This attribute can only be applied to auto function
2934 scope variables; it may not be applied to parameters or variables
2935 with static storage duration. The function must take one parameter,
2936 a pointer to a type compatible with the variable. The return value
2937 of the function (if any) is ignored.
2939 If @option{-fexceptions} is enabled, then @var{cleanup_function}
2940 will be run during the stack unwinding that happens during the
2941 processing of the exception. Note that the @code{cleanup} attribute
2942 does not allow the exception to be caught, only to perform an action.
2943 It is undefined what happens if @var{cleanup_function} does not
2948 @cindex @code{common} attribute
2949 @cindex @code{nocommon} attribute
2952 The @code{common} attribute requests GCC to place a variable in
2953 ``common'' storage. The @code{nocommon} attribute requests the
2954 opposite---to allocate space for it directly.
2956 These attributes override the default chosen by the
2957 @option{-fno-common} and @option{-fcommon} flags respectively.
2960 @cindex @code{deprecated} attribute
2961 The @code{deprecated} attribute results in a warning if the variable
2962 is used anywhere in the source file. This is useful when identifying
2963 variables that are expected to be removed in a future version of a
2964 program. The warning also includes the location of the declaration
2965 of the deprecated variable, to enable users to easily find further
2966 information about why the variable is deprecated, or what they should
2967 do instead. Note that the warning only occurs for uses:
2970 extern int old_var __attribute__ ((deprecated));
2972 int new_fn () @{ return old_var; @}
2975 results in a warning on line 3 but not line 2.
2977 The @code{deprecated} attribute can also be used for functions and
2978 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
2980 @item mode (@var{mode})
2981 @cindex @code{mode} attribute
2982 This attribute specifies the data type for the declaration---whichever
2983 type corresponds to the mode @var{mode}. This in effect lets you
2984 request an integer or floating point type according to its width.
2986 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2987 indicate the mode corresponding to a one-byte integer, @samp{word} or
2988 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2989 or @samp{__pointer__} for the mode used to represent pointers.
2992 @cindex @code{packed} attribute
2993 The @code{packed} attribute specifies that a variable or structure field
2994 should have the smallest possible alignment---one byte for a variable,
2995 and one bit for a field, unless you specify a larger value with the
2996 @code{aligned} attribute.
2998 Here is a structure in which the field @code{x} is packed, so that it
2999 immediately follows @code{a}:
3005 int x[2] __attribute__ ((packed));
3009 @item section ("@var{section-name}")
3010 @cindex @code{section} variable attribute
3011 Normally, the compiler places the objects it generates in sections like
3012 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
3013 or you need certain particular variables to appear in special sections,
3014 for example to map to special hardware. The @code{section}
3015 attribute specifies that a variable (or function) lives in a particular
3016 section. For example, this small program uses several specific section names:
3019 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
3020 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
3021 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
3022 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3026 /* @r{Initialize stack pointer} */
3027 init_sp (stack + sizeof (stack));
3029 /* @r{Initialize initialized data} */
3030 memcpy (&init_data, &data, &edata - &data);
3032 /* @r{Turn on the serial ports} */
3039 Use the @code{section} attribute with an @emph{initialized} definition
3040 of a @emph{global} variable, as shown in the example. GCC issues
3041 a warning and otherwise ignores the @code{section} attribute in
3042 uninitialized variable declarations.
3044 You may only use the @code{section} attribute with a fully initialized
3045 global definition because of the way linkers work. The linker requires
3046 each object be defined once, with the exception that uninitialized
3047 variables tentatively go in the @code{common} (or @code{bss}) section
3048 and can be multiply ``defined''. You can force a variable to be
3049 initialized with the @option{-fno-common} flag or the @code{nocommon}
3052 Some file formats do not support arbitrary sections so the @code{section}
3053 attribute is not available on all platforms.
3054 If you need to map the entire contents of a module to a particular
3055 section, consider using the facilities of the linker instead.
3058 @cindex @code{shared} variable attribute
3059 On Microsoft Windows, in addition to putting variable definitions in a named
3060 section, the section can also be shared among all running copies of an
3061 executable or DLL@. For example, this small program defines shared data
3062 by putting it in a named section @code{shared} and marking the section
3066 int foo __attribute__((section ("shared"), shared)) = 0;
3071 /* @r{Read and write foo. All running
3072 copies see the same value.} */
3078 You may only use the @code{shared} attribute along with @code{section}
3079 attribute with a fully initialized global definition because of the way
3080 linkers work. See @code{section} attribute for more information.
3082 The @code{shared} attribute is only available on Microsoft Windows@.
3084 @item tls_model ("@var{tls_model}")
3085 @cindex @code{tls_model} attribute
3086 The @code{tls_model} attribute sets thread-local storage model
3087 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
3088 overriding @option{-ftls-model=} command line switch on a per-variable
3090 The @var{tls_model} argument should be one of @code{global-dynamic},
3091 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
3093 Not all targets support this attribute.
3096 This attribute, attached to a variable, means that the variable is meant
3097 to be possibly unused. GCC will not produce a warning for this
3100 @item vector_size (@var{bytes})
3101 This attribute specifies the vector size for the variable, measured in
3102 bytes. For example, the declaration:
3105 int foo __attribute__ ((vector_size (16)));
3109 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3110 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3111 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3113 This attribute is only applicable to integral and float scalars,
3114 although arrays, pointers, and function return values are allowed in
3115 conjunction with this construct.
3117 Aggregates with this attribute are invalid, even if they are of the same
3118 size as a corresponding scalar. For example, the declaration:
3121 struct S @{ int a; @};
3122 struct S __attribute__ ((vector_size (16))) foo;
3126 is invalid even if the size of the structure is the same as the size of
3130 The @code{selectany} attribute causes an initialized global variable to
3131 have link-once semantics. When multiple definitions of the variable are
3132 encountered by the linker, the first is selected and the remainder are
3133 discarded. Following usage by the Microsoft compiler, the linker is told
3134 @emph{not} to warn about size or content differences of the multiple
3137 Although the primary usage of this attribute is for POD types, the
3138 attribute can also be applied to global C++ objects that are initialized
3139 by a constructor. In this case, the static initialization and destruction
3140 code for the object is emitted in each translation defining the object,
3141 but the calls to the constructor and destructor are protected by a
3142 link-once guard variable.
3144 The @code{selectany} attribute is only available on Microsoft Windows
3145 targets. You can use @code{__declspec (selectany)} as a synonym for
3146 @code{__attribute__ ((selectany))} for compatibility with other
3150 The @code{weak} attribute is described in @xref{Function Attributes}.
3153 The @code{dllimport} attribute is described in @xref{Function Attributes}.
3156 The @code{dllexport} attribute is described in @xref{Function Attributes}.
3160 @subsection M32R/D Variable Attributes
3162 One attribute is currently defined for the M32R/D@.
3165 @item model (@var{model-name})
3166 @cindex variable addressability on the M32R/D
3167 Use this attribute on the M32R/D to set the addressability of an object.
3168 The identifier @var{model-name} is one of @code{small}, @code{medium},
3169 or @code{large}, representing each of the code models.
3171 Small model objects live in the lower 16MB of memory (so that their
3172 addresses can be loaded with the @code{ld24} instruction).
3174 Medium and large model objects may live anywhere in the 32-bit address space
3175 (the compiler will generate @code{seth/add3} instructions to load their
3179 @subsection i386 Variable Attributes
3181 Two attributes are currently defined for i386 configurations:
3182 @code{ms_struct} and @code{gcc_struct}
3187 @cindex @code{ms_struct} attribute
3188 @cindex @code{gcc_struct} attribute
3190 If @code{packed} is used on a structure, or if bit-fields are used
3191 it may be that the Microsoft ABI packs them differently
3192 than GCC would normally pack them. Particularly when moving packed
3193 data between functions compiled with GCC and the native Microsoft compiler
3194 (either via function call or as data in a file), it may be necessary to access
3197 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3198 compilers to match the native Microsoft compiler.
3201 @subsection Xstormy16 Variable Attributes
3203 One attribute is currently defined for xstormy16 configurations:
3208 @cindex @code{below100} attribute
3210 If a variable has the @code{below100} attribute (@code{BELOW100} is
3211 allowed also), GCC will place the variable in the first 0x100 bytes of
3212 memory and use special opcodes to access it. Such variables will be
3213 placed in either the @code{.bss_below100} section or the
3214 @code{.data_below100} section.
3218 @node Type Attributes
3219 @section Specifying Attributes of Types
3220 @cindex attribute of types
3221 @cindex type attributes
3223 The keyword @code{__attribute__} allows you to specify special
3224 attributes of @code{struct} and @code{union} types when you define such
3225 types. This keyword is followed by an attribute specification inside
3226 double parentheses. Six attributes are currently defined for types:
3227 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
3228 @code{deprecated} and @code{may_alias}. Other attributes are defined for
3229 functions (@pxref{Function Attributes}) and for variables
3230 (@pxref{Variable Attributes}).
3232 You may also specify any one of these attributes with @samp{__}
3233 preceding and following its keyword. This allows you to use these
3234 attributes in header files without being concerned about a possible
3235 macro of the same name. For example, you may use @code{__aligned__}
3236 instead of @code{aligned}.
3238 You may specify the @code{aligned} and @code{transparent_union}
3239 attributes either in a @code{typedef} declaration or just past the
3240 closing curly brace of a complete enum, struct or union type
3241 @emph{definition} and the @code{packed} attribute only past the closing
3242 brace of a definition.
3244 You may also specify attributes between the enum, struct or union
3245 tag and the name of the type rather than after the closing brace.
3247 @xref{Attribute Syntax}, for details of the exact syntax for using
3251 @cindex @code{aligned} attribute
3252 @item aligned (@var{alignment})
3253 This attribute specifies a minimum alignment (in bytes) for variables
3254 of the specified type. For example, the declarations:
3257 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3258 typedef int more_aligned_int __attribute__ ((aligned (8)));
3262 force the compiler to insure (as far as it can) that each variable whose
3263 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3264 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
3265 variables of type @code{struct S} aligned to 8-byte boundaries allows
3266 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3267 store) instructions when copying one variable of type @code{struct S} to
3268 another, thus improving run-time efficiency.
3270 Note that the alignment of any given @code{struct} or @code{union} type
3271 is required by the ISO C standard to be at least a perfect multiple of
3272 the lowest common multiple of the alignments of all of the members of
3273 the @code{struct} or @code{union} in question. This means that you @emph{can}
3274 effectively adjust the alignment of a @code{struct} or @code{union}
3275 type by attaching an @code{aligned} attribute to any one of the members
3276 of such a type, but the notation illustrated in the example above is a
3277 more obvious, intuitive, and readable way to request the compiler to
3278 adjust the alignment of an entire @code{struct} or @code{union} type.
3280 As in the preceding example, you can explicitly specify the alignment
3281 (in bytes) that you wish the compiler to use for a given @code{struct}
3282 or @code{union} type. Alternatively, you can leave out the alignment factor
3283 and just ask the compiler to align a type to the maximum
3284 useful alignment for the target machine you are compiling for. For
3285 example, you could write:
3288 struct S @{ short f[3]; @} __attribute__ ((aligned));
3291 Whenever you leave out the alignment factor in an @code{aligned}
3292 attribute specification, the compiler automatically sets the alignment
3293 for the type to the largest alignment which is ever used for any data
3294 type on the target machine you are compiling for. Doing this can often
3295 make copy operations more efficient, because the compiler can use
3296 whatever instructions copy the biggest chunks of memory when performing
3297 copies to or from the variables which have types that you have aligned
3300 In the example above, if the size of each @code{short} is 2 bytes, then
3301 the size of the entire @code{struct S} type is 6 bytes. The smallest
3302 power of two which is greater than or equal to that is 8, so the
3303 compiler sets the alignment for the entire @code{struct S} type to 8
3306 Note that although you can ask the compiler to select a time-efficient
3307 alignment for a given type and then declare only individual stand-alone
3308 objects of that type, the compiler's ability to select a time-efficient
3309 alignment is primarily useful only when you plan to create arrays of
3310 variables having the relevant (efficiently aligned) type. If you
3311 declare or use arrays of variables of an efficiently-aligned type, then
3312 it is likely that your program will also be doing pointer arithmetic (or
3313 subscripting, which amounts to the same thing) on pointers to the
3314 relevant type, and the code that the compiler generates for these
3315 pointer arithmetic operations will often be more efficient for
3316 efficiently-aligned types than for other types.
3318 The @code{aligned} attribute can only increase the alignment; but you
3319 can decrease it by specifying @code{packed} as well. See below.
3321 Note that the effectiveness of @code{aligned} attributes may be limited
3322 by inherent limitations in your linker. On many systems, the linker is
3323 only able to arrange for variables to be aligned up to a certain maximum
3324 alignment. (For some linkers, the maximum supported alignment may
3325 be very very small.) If your linker is only able to align variables
3326 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3327 in an @code{__attribute__} will still only provide you with 8 byte
3328 alignment. See your linker documentation for further information.
3331 This attribute, attached to @code{struct} or @code{union} type
3332 definition, specifies that each member (other than zero-width bitfields)
3333 of the structure or union is placed to minimize the memory required. When
3334 attached to an @code{enum} definition, it indicates that the smallest
3335 integral type should be used.
3337 @opindex fshort-enums
3338 Specifying this attribute for @code{struct} and @code{union} types is
3339 equivalent to specifying the @code{packed} attribute on each of the
3340 structure or union members. Specifying the @option{-fshort-enums}
3341 flag on the line is equivalent to specifying the @code{packed}
3342 attribute on all @code{enum} definitions.
3344 In the following example @code{struct my_packed_struct}'s members are
3345 packed closely together, but the internal layout of its @code{s} member
3346 is not packed---to do that, @code{struct my_unpacked_struct} would need to
3350 struct my_unpacked_struct
3356 struct __attribute__ ((__packed__)) my_packed_struct
3360 struct my_unpacked_struct s;
3364 You may only specify this attribute on the definition of a @code{enum},
3365 @code{struct} or @code{union}, not on a @code{typedef} which does not
3366 also define the enumerated type, structure or union.
3368 @item transparent_union
3369 This attribute, attached to a @code{union} type definition, indicates
3370 that any function parameter having that union type causes calls to that
3371 function to be treated in a special way.
3373 First, the argument corresponding to a transparent union type can be of
3374 any type in the union; no cast is required. Also, if the union contains
3375 a pointer type, the corresponding argument can be a null pointer
3376 constant or a void pointer expression; and if the union contains a void
3377 pointer type, the corresponding argument can be any pointer expression.
3378 If the union member type is a pointer, qualifiers like @code{const} on
3379 the referenced type must be respected, just as with normal pointer
3382 Second, the argument is passed to the function using the calling
3383 conventions of the first member of the transparent union, not the calling
3384 conventions of the union itself. All members of the union must have the
3385 same machine representation; this is necessary for this argument passing
3388 Transparent unions are designed for library functions that have multiple
3389 interfaces for compatibility reasons. For example, suppose the
3390 @code{wait} function must accept either a value of type @code{int *} to
3391 comply with Posix, or a value of type @code{union wait *} to comply with
3392 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3393 @code{wait} would accept both kinds of arguments, but it would also
3394 accept any other pointer type and this would make argument type checking
3395 less useful. Instead, @code{<sys/wait.h>} might define the interface
3403 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3405 pid_t wait (wait_status_ptr_t);
3408 This interface allows either @code{int *} or @code{union wait *}
3409 arguments to be passed, using the @code{int *} calling convention.
3410 The program can call @code{wait} with arguments of either type:
3413 int w1 () @{ int w; return wait (&w); @}
3414 int w2 () @{ union wait w; return wait (&w); @}
3417 With this interface, @code{wait}'s implementation might look like this:
3420 pid_t wait (wait_status_ptr_t p)
3422 return waitpid (-1, p.__ip, 0);
3427 When attached to a type (including a @code{union} or a @code{struct}),
3428 this attribute means that variables of that type are meant to appear
3429 possibly unused. GCC will not produce a warning for any variables of
3430 that type, even if the variable appears to do nothing. This is often
3431 the case with lock or thread classes, which are usually defined and then
3432 not referenced, but contain constructors and destructors that have
3433 nontrivial bookkeeping functions.
3436 The @code{deprecated} attribute results in a warning if the type
3437 is used anywhere in the source file. This is useful when identifying
3438 types that are expected to be removed in a future version of a program.
3439 If possible, the warning also includes the location of the declaration
3440 of the deprecated type, to enable users to easily find further
3441 information about why the type is deprecated, or what they should do
3442 instead. Note that the warnings only occur for uses and then only
3443 if the type is being applied to an identifier that itself is not being
3444 declared as deprecated.
3447 typedef int T1 __attribute__ ((deprecated));
3451 typedef T1 T3 __attribute__ ((deprecated));
3452 T3 z __attribute__ ((deprecated));
3455 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3456 warning is issued for line 4 because T2 is not explicitly
3457 deprecated. Line 5 has no warning because T3 is explicitly
3458 deprecated. Similarly for line 6.
3460 The @code{deprecated} attribute can also be used for functions and
3461 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3464 Accesses to objects with types with this attribute are not subjected to
3465 type-based alias analysis, but are instead assumed to be able to alias
3466 any other type of objects, just like the @code{char} type. See
3467 @option{-fstrict-aliasing} for more information on aliasing issues.
3472 typedef short __attribute__((__may_alias__)) short_a;
3478 short_a *b = (short_a *) &a;
3482 if (a == 0x12345678)
3489 If you replaced @code{short_a} with @code{short} in the variable
3490 declaration, the above program would abort when compiled with
3491 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3492 above in recent GCC versions.
3494 @subsection ARM Type Attributes
3496 On those ARM targets that support @code{dllimport} (such as Symbian
3497 OS), you can use the @code{notshared} attribute to indicate that the
3498 virtual table and other similar data for a class should not be
3499 exported from a DLL@. For example:
3502 class __declspec(notshared) C @{
3504 __declspec(dllimport) C();
3508 __declspec(dllexport)
3512 In this code, @code{C::C} is exported from the current DLL, but the
3513 virtual table for @code{C} is not exported. (You can use
3514 @code{__attribute__} instead of @code{__declspec} if you prefer, but
3515 most Symbian OS code uses @code{__declspec}.)
3517 @subsection i386 Type Attributes
3519 Two attributes are currently defined for i386 configurations:
3520 @code{ms_struct} and @code{gcc_struct}
3524 @cindex @code{ms_struct}
3525 @cindex @code{gcc_struct}
3527 If @code{packed} is used on a structure, or if bit-fields are used
3528 it may be that the Microsoft ABI packs them differently
3529 than GCC would normally pack them. Particularly when moving packed
3530 data between functions compiled with GCC and the native Microsoft compiler
3531 (either via function call or as data in a file), it may be necessary to access
3534 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3535 compilers to match the native Microsoft compiler.
3538 To specify multiple attributes, separate them by commas within the
3539 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3543 @section An Inline Function is As Fast As a Macro
3544 @cindex inline functions
3545 @cindex integrating function code
3547 @cindex macros, inline alternative
3549 By declaring a function @code{inline}, you can direct GCC to
3550 integrate that function's code into the code for its callers. This
3551 makes execution faster by eliminating the function-call overhead; in
3552 addition, if any of the actual argument values are constant, their known
3553 values may permit simplifications at compile time so that not all of the
3554 inline function's code needs to be included. The effect on code size is
3555 less predictable; object code may be larger or smaller with function
3556 inlining, depending on the particular case. Inlining of functions is an
3557 optimization and it really ``works'' only in optimizing compilation. If
3558 you don't use @option{-O}, no function is really inline.
3560 Inline functions are included in the ISO C99 standard, but there are
3561 currently substantial differences between what GCC implements and what
3562 the ISO C99 standard requires.
3564 To declare a function inline, use the @code{inline} keyword in its
3565 declaration, like this:
3575 (If you are writing a header file to be included in ISO C programs, write
3576 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3577 You can also make all ``simple enough'' functions inline with the option
3578 @option{-finline-functions}.
3581 Note that certain usages in a function definition can make it unsuitable
3582 for inline substitution. Among these usages are: use of varargs, use of
3583 alloca, use of variable sized data types (@pxref{Variable Length}),
3584 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3585 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3586 will warn when a function marked @code{inline} could not be substituted,
3587 and will give the reason for the failure.
3589 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3590 does not affect the linkage of the function.
3592 @cindex automatic @code{inline} for C++ member fns
3593 @cindex @code{inline} automatic for C++ member fns
3594 @cindex member fns, automatically @code{inline}
3595 @cindex C++ member fns, automatically @code{inline}
3596 @opindex fno-default-inline
3597 GCC automatically inlines member functions defined within the class
3598 body of C++ programs even if they are not explicitly declared
3599 @code{inline}. (You can override this with @option{-fno-default-inline};
3600 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3602 @cindex inline functions, omission of
3603 @opindex fkeep-inline-functions
3604 When a function is both inline and @code{static}, if all calls to the
3605 function are integrated into the caller, and the function's address is
3606 never used, then the function's own assembler code is never referenced.
3607 In this case, GCC does not actually output assembler code for the
3608 function, unless you specify the option @option{-fkeep-inline-functions}.
3609 Some calls cannot be integrated for various reasons (in particular,
3610 calls that precede the function's definition cannot be integrated, and
3611 neither can recursive calls within the definition). If there is a
3612 nonintegrated call, then the function is compiled to assembler code as
3613 usual. The function must also be compiled as usual if the program
3614 refers to its address, because that can't be inlined.
3616 @cindex non-static inline function
3617 When an inline function is not @code{static}, then the compiler must assume
3618 that there may be calls from other source files; since a global symbol can
3619 be defined only once in any program, the function must not be defined in
3620 the other source files, so the calls therein cannot be integrated.
3621 Therefore, a non-@code{static} inline function is always compiled on its
3622 own in the usual fashion.
3624 If you specify both @code{inline} and @code{extern} in the function
3625 definition, then the definition is used only for inlining. In no case
3626 is the function compiled on its own, not even if you refer to its
3627 address explicitly. Such an address becomes an external reference, as
3628 if you had only declared the function, and had not defined it.
3630 This combination of @code{inline} and @code{extern} has almost the
3631 effect of a macro. The way to use it is to put a function definition in
3632 a header file with these keywords, and put another copy of the
3633 definition (lacking @code{inline} and @code{extern}) in a library file.
3634 The definition in the header file will cause most calls to the function
3635 to be inlined. If any uses of the function remain, they will refer to
3636 the single copy in the library.
3638 Since GCC eventually will implement ISO C99 semantics for
3639 inline functions, it is best to use @code{static inline} only
3640 to guarantee compatibility. (The
3641 existing semantics will remain available when @option{-std=gnu89} is
3642 specified, but eventually the default will be @option{-std=gnu99} and
3643 that will implement the C99 semantics, though it does not do so yet.)
3645 GCC does not inline any functions when not optimizing unless you specify
3646 the @samp{always_inline} attribute for the function, like this:
3649 /* @r{Prototype.} */
3650 inline void foo (const char) __attribute__((always_inline));
3654 @section Assembler Instructions with C Expression Operands
3655 @cindex extended @code{asm}
3656 @cindex @code{asm} expressions
3657 @cindex assembler instructions
3660 In an assembler instruction using @code{asm}, you can specify the
3661 operands of the instruction using C expressions. This means you need not
3662 guess which registers or memory locations will contain the data you want
3665 You must specify an assembler instruction template much like what
3666 appears in a machine description, plus an operand constraint string for
3669 For example, here is how to use the 68881's @code{fsinx} instruction:
3672 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3676 Here @code{angle} is the C expression for the input operand while
3677 @code{result} is that of the output operand. Each has @samp{"f"} as its
3678 operand constraint, saying that a floating point register is required.
3679 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3680 output operands' constraints must use @samp{=}. The constraints use the
3681 same language used in the machine description (@pxref{Constraints}).
3683 Each operand is described by an operand-constraint string followed by
3684 the C expression in parentheses. A colon separates the assembler
3685 template from the first output operand and another separates the last
3686 output operand from the first input, if any. Commas separate the
3687 operands within each group. The total number of operands is currently
3688 limited to 30; this limitation may be lifted in some future version of
3691 If there are no output operands but there are input operands, you must
3692 place two consecutive colons surrounding the place where the output
3695 As of GCC version 3.1, it is also possible to specify input and output
3696 operands using symbolic names which can be referenced within the
3697 assembler code. These names are specified inside square brackets
3698 preceding the constraint string, and can be referenced inside the
3699 assembler code using @code{%[@var{name}]} instead of a percentage sign
3700 followed by the operand number. Using named operands the above example
3704 asm ("fsinx %[angle],%[output]"
3705 : [output] "=f" (result)
3706 : [angle] "f" (angle));
3710 Note that the symbolic operand names have no relation whatsoever to
3711 other C identifiers. You may use any name you like, even those of
3712 existing C symbols, but you must ensure that no two operands within the same
3713 assembler construct use the same symbolic name.
3715 Output operand expressions must be lvalues; the compiler can check this.
3716 The input operands need not be lvalues. The compiler cannot check
3717 whether the operands have data types that are reasonable for the
3718 instruction being executed. It does not parse the assembler instruction
3719 template and does not know what it means or even whether it is valid
3720 assembler input. The extended @code{asm} feature is most often used for
3721 machine instructions the compiler itself does not know exist. If
3722 the output expression cannot be directly addressed (for example, it is a
3723 bit-field), your constraint must allow a register. In that case, GCC
3724 will use the register as the output of the @code{asm}, and then store
3725 that register into the output.
3727 The ordinary output operands must be write-only; GCC will assume that
3728 the values in these operands before the instruction are dead and need
3729 not be generated. Extended asm supports input-output or read-write
3730 operands. Use the constraint character @samp{+} to indicate such an
3731 operand and list it with the output operands. You should only use
3732 read-write operands when the constraints for the operand (or the
3733 operand in which only some of the bits are to be changed) allow a
3736 You may, as an alternative, logically split its function into two
3737 separate operands, one input operand and one write-only output
3738 operand. The connection between them is expressed by constraints
3739 which say they need to be in the same location when the instruction
3740 executes. You can use the same C expression for both operands, or
3741 different expressions. For example, here we write the (fictitious)
3742 @samp{combine} instruction with @code{bar} as its read-only source
3743 operand and @code{foo} as its read-write destination:
3746 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3750 The constraint @samp{"0"} for operand 1 says that it must occupy the
3751 same location as operand 0. A number in constraint is allowed only in
3752 an input operand and it must refer to an output operand.
3754 Only a number in the constraint can guarantee that one operand will be in
3755 the same place as another. The mere fact that @code{foo} is the value
3756 of both operands is not enough to guarantee that they will be in the
3757 same place in the generated assembler code. The following would not
3761 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3764 Various optimizations or reloading could cause operands 0 and 1 to be in
3765 different registers; GCC knows no reason not to do so. For example, the
3766 compiler might find a copy of the value of @code{foo} in one register and
3767 use it for operand 1, but generate the output operand 0 in a different
3768 register (copying it afterward to @code{foo}'s own address). Of course,
3769 since the register for operand 1 is not even mentioned in the assembler
3770 code, the result will not work, but GCC can't tell that.
3772 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3773 the operand number for a matching constraint. For example:
3776 asm ("cmoveq %1,%2,%[result]"
3777 : [result] "=r"(result)
3778 : "r" (test), "r"(new), "[result]"(old));
3781 Sometimes you need to make an @code{asm} operand be a specific register,
3782 but there's no matching constraint letter for that register @emph{by
3783 itself}. To force the operand into that register, use a local variable
3784 for the operand and specify the register in the variable declaration.
3785 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
3786 register constraint letter that matches the register:
3789 register int *p1 asm ("r0") = @dots{};
3790 register int *p2 asm ("r1") = @dots{};
3791 register int *result asm ("r0");
3792 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
3795 @anchor{Example of asm with clobbered asm reg}
3796 In the above example, beware that a register that is call-clobbered by
3797 the target ABI will be overwritten by any function call in the
3798 assignment, including library calls for arithmetic operators.
3799 Assuming it is a call-clobbered register, this may happen to @code{r0}
3800 above by the assignment to @code{p2}. If you have to use such a
3801 register, use temporary variables for expressions between the register
3806 register int *p1 asm ("r0") = @dots{};
3807 register int *p2 asm ("r1") = t1;
3808 register int *result asm ("r0");
3809 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
3812 Some instructions clobber specific hard registers. To describe this,
3813 write a third colon after the input operands, followed by the names of
3814 the clobbered hard registers (given as strings). Here is a realistic
3815 example for the VAX:
3818 asm volatile ("movc3 %0,%1,%2"
3819 : /* @r{no outputs} */
3820 : "g" (from), "g" (to), "g" (count)
3821 : "r0", "r1", "r2", "r3", "r4", "r5");
3824 You may not write a clobber description in a way that overlaps with an
3825 input or output operand. For example, you may not have an operand
3826 describing a register class with one member if you mention that register
3827 in the clobber list. Variables declared to live in specific registers
3828 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
3829 have no part mentioned in the clobber description.
3830 There is no way for you to specify that an input
3831 operand is modified without also specifying it as an output
3832 operand. Note that if all the output operands you specify are for this
3833 purpose (and hence unused), you will then also need to specify
3834 @code{volatile} for the @code{asm} construct, as described below, to
3835 prevent GCC from deleting the @code{asm} statement as unused.
3837 If you refer to a particular hardware register from the assembler code,
3838 you will probably have to list the register after the third colon to
3839 tell the compiler the register's value is modified. In some assemblers,
3840 the register names begin with @samp{%}; to produce one @samp{%} in the
3841 assembler code, you must write @samp{%%} in the input.
3843 If your assembler instruction can alter the condition code register, add
3844 @samp{cc} to the list of clobbered registers. GCC on some machines
3845 represents the condition codes as a specific hardware register;
3846 @samp{cc} serves to name this register. On other machines, the
3847 condition code is handled differently, and specifying @samp{cc} has no
3848 effect. But it is valid no matter what the machine.
3850 If your assembler instructions access memory in an unpredictable
3851 fashion, add @samp{memory} to the list of clobbered registers. This
3852 will cause GCC to not keep memory values cached in registers across the
3853 assembler instruction and not optimize stores or loads to that memory.
3854 You will also want to add the @code{volatile} keyword if the memory
3855 affected is not listed in the inputs or outputs of the @code{asm}, as
3856 the @samp{memory} clobber does not count as a side-effect of the
3857 @code{asm}. If you know how large the accessed memory is, you can add
3858 it as input or output but if this is not known, you should add
3859 @samp{memory}. As an example, if you access ten bytes of a string, you
3860 can use a memory input like:
3863 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
3866 Note that in the following example the memory input is necessary,
3867 otherwise GCC might optimize the store to @code{x} away:
3874 asm ("magic stuff accessing an 'int' pointed to by '%1'"
3875 "=&d" (r) : "a" (y), "m" (*y));
3880 You can put multiple assembler instructions together in a single
3881 @code{asm} template, separated by the characters normally used in assembly
3882 code for the system. A combination that works in most places is a newline
3883 to break the line, plus a tab character to move to the instruction field
3884 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3885 assembler allows semicolons as a line-breaking character. Note that some
3886 assembler dialects use semicolons to start a comment.
3887 The input operands are guaranteed not to use any of the clobbered
3888 registers, and neither will the output operands' addresses, so you can
3889 read and write the clobbered registers as many times as you like. Here
3890 is an example of multiple instructions in a template; it assumes the
3891 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3894 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3896 : "g" (from), "g" (to)
3900 Unless an output operand has the @samp{&} constraint modifier, GCC
3901 may allocate it in the same register as an unrelated input operand, on
3902 the assumption the inputs are consumed before the outputs are produced.
3903 This assumption may be false if the assembler code actually consists of
3904 more than one instruction. In such a case, use @samp{&} for each output
3905 operand that may not overlap an input. @xref{Modifiers}.
3907 If you want to test the condition code produced by an assembler
3908 instruction, you must include a branch and a label in the @code{asm}
3909 construct, as follows:
3912 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3918 This assumes your assembler supports local labels, as the GNU assembler
3919 and most Unix assemblers do.
3921 Speaking of labels, jumps from one @code{asm} to another are not
3922 supported. The compiler's optimizers do not know about these jumps, and
3923 therefore they cannot take account of them when deciding how to
3926 @cindex macros containing @code{asm}
3927 Usually the most convenient way to use these @code{asm} instructions is to
3928 encapsulate them in macros that look like functions. For example,
3932 (@{ double __value, __arg = (x); \
3933 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3938 Here the variable @code{__arg} is used to make sure that the instruction
3939 operates on a proper @code{double} value, and to accept only those
3940 arguments @code{x} which can convert automatically to a @code{double}.
3942 Another way to make sure the instruction operates on the correct data
3943 type is to use a cast in the @code{asm}. This is different from using a
3944 variable @code{__arg} in that it converts more different types. For
3945 example, if the desired type were @code{int}, casting the argument to
3946 @code{int} would accept a pointer with no complaint, while assigning the
3947 argument to an @code{int} variable named @code{__arg} would warn about
3948 using a pointer unless the caller explicitly casts it.
3950 If an @code{asm} has output operands, GCC assumes for optimization
3951 purposes the instruction has no side effects except to change the output
3952 operands. This does not mean instructions with a side effect cannot be
3953 used, but you must be careful, because the compiler may eliminate them
3954 if the output operands aren't used, or move them out of loops, or
3955 replace two with one if they constitute a common subexpression. Also,
3956 if your instruction does have a side effect on a variable that otherwise
3957 appears not to change, the old value of the variable may be reused later
3958 if it happens to be found in a register.
3960 You can prevent an @code{asm} instruction from being deleted
3961 by writing the keyword @code{volatile} after
3962 the @code{asm}. For example:
3965 #define get_and_set_priority(new) \
3967 asm volatile ("get_and_set_priority %0, %1" \
3968 : "=g" (__old) : "g" (new)); \
3973 The @code{volatile} keyword indicates that the instruction has
3974 important side-effects. GCC will not delete a volatile @code{asm} if
3975 it is reachable. (The instruction can still be deleted if GCC can
3976 prove that control-flow will never reach the location of the
3977 instruction.) Note that even a volatile @code{asm} instruction
3978 can be moved relative to other code, including across jump
3979 instructions. For example, on many targets there is a system
3980 register which can be set to control the rounding mode of
3981 floating point operations. You might try
3982 setting it with a volatile @code{asm}, like this PowerPC example:
3985 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
3990 This will not work reliably, as the compiler may move the addition back
3991 before the volatile @code{asm}. To make it work you need to add an
3992 artificial dependency to the @code{asm} referencing a variable in the code
3993 you don't want moved, for example:
3996 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
4000 Similarly, you can't expect a
4001 sequence of volatile @code{asm} instructions to remain perfectly
4002 consecutive. If you want consecutive output, use a single @code{asm}.
4003 Also, GCC will perform some optimizations across a volatile @code{asm}
4004 instruction; GCC does not ``forget everything'' when it encounters
4005 a volatile @code{asm} instruction the way some other compilers do.
4007 An @code{asm} instruction without any output operands will be treated
4008 identically to a volatile @code{asm} instruction.
4010 It is a natural idea to look for a way to give access to the condition
4011 code left by the assembler instruction. However, when we attempted to
4012 implement this, we found no way to make it work reliably. The problem
4013 is that output operands might need reloading, which would result in
4014 additional following ``store'' instructions. On most machines, these
4015 instructions would alter the condition code before there was time to
4016 test it. This problem doesn't arise for ordinary ``test'' and
4017 ``compare'' instructions because they don't have any output operands.
4019 For reasons similar to those described above, it is not possible to give
4020 an assembler instruction access to the condition code left by previous
4023 If you are writing a header file that should be includable in ISO C
4024 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
4027 @subsection Size of an @code{asm}
4029 Some targets require that GCC track the size of each instruction used in
4030 order to generate correct code. Because the final length of an
4031 @code{asm} is only known by the assembler, GCC must make an estimate as
4032 to how big it will be. The estimate is formed by counting the number of
4033 statements in the pattern of the @code{asm} and multiplying that by the
4034 length of the longest instruction on that processor. Statements in the
4035 @code{asm} are identified by newline characters and whatever statement
4036 separator characters are supported by the assembler; on most processors
4037 this is the `@code{;}' character.
4039 Normally, GCC's estimate is perfectly adequate to ensure that correct
4040 code is generated, but it is possible to confuse the compiler if you use
4041 pseudo instructions or assembler macros that expand into multiple real
4042 instructions or if you use assembler directives that expand to more
4043 space in the object file than would be needed for a single instruction.
4044 If this happens then the assembler will produce a diagnostic saying that
4045 a label is unreachable.
4047 @subsection i386 floating point asm operands
4049 There are several rules on the usage of stack-like regs in
4050 asm_operands insns. These rules apply only to the operands that are
4055 Given a set of input regs that die in an asm_operands, it is
4056 necessary to know which are implicitly popped by the asm, and
4057 which must be explicitly popped by gcc.
4059 An input reg that is implicitly popped by the asm must be
4060 explicitly clobbered, unless it is constrained to match an
4064 For any input reg that is implicitly popped by an asm, it is
4065 necessary to know how to adjust the stack to compensate for the pop.
4066 If any non-popped input is closer to the top of the reg-stack than
4067 the implicitly popped reg, it would not be possible to know what the
4068 stack looked like---it's not clear how the rest of the stack ``slides
4071 All implicitly popped input regs must be closer to the top of
4072 the reg-stack than any input that is not implicitly popped.
4074 It is possible that if an input dies in an insn, reload might
4075 use the input reg for an output reload. Consider this example:
4078 asm ("foo" : "=t" (a) : "f" (b));
4081 This asm says that input B is not popped by the asm, and that
4082 the asm pushes a result onto the reg-stack, i.e., the stack is one
4083 deeper after the asm than it was before. But, it is possible that
4084 reload will think that it can use the same reg for both the input and
4085 the output, if input B dies in this insn.
4087 If any input operand uses the @code{f} constraint, all output reg
4088 constraints must use the @code{&} earlyclobber.
4090 The asm above would be written as
4093 asm ("foo" : "=&t" (a) : "f" (b));
4097 Some operands need to be in particular places on the stack. All
4098 output operands fall in this category---there is no other way to
4099 know which regs the outputs appear in unless the user indicates
4100 this in the constraints.
4102 Output operands must specifically indicate which reg an output
4103 appears in after an asm. @code{=f} is not allowed: the operand
4104 constraints must select a class with a single reg.
4107 Output operands may not be ``inserted'' between existing stack regs.
4108 Since no 387 opcode uses a read/write operand, all output operands
4109 are dead before the asm_operands, and are pushed by the asm_operands.
4110 It makes no sense to push anywhere but the top of the reg-stack.
4112 Output operands must start at the top of the reg-stack: output
4113 operands may not ``skip'' a reg.
4116 Some asm statements may need extra stack space for internal
4117 calculations. This can be guaranteed by clobbering stack registers
4118 unrelated to the inputs and outputs.
4122 Here are a couple of reasonable asms to want to write. This asm
4123 takes one input, which is internally popped, and produces two outputs.
4126 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
4129 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
4130 and replaces them with one output. The user must code the @code{st(1)}
4131 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
4134 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
4140 @section Controlling Names Used in Assembler Code
4141 @cindex assembler names for identifiers
4142 @cindex names used in assembler code
4143 @cindex identifiers, names in assembler code
4145 You can specify the name to be used in the assembler code for a C
4146 function or variable by writing the @code{asm} (or @code{__asm__})
4147 keyword after the declarator as follows:
4150 int foo asm ("myfoo") = 2;
4154 This specifies that the name to be used for the variable @code{foo} in
4155 the assembler code should be @samp{myfoo} rather than the usual
4158 On systems where an underscore is normally prepended to the name of a C
4159 function or variable, this feature allows you to define names for the
4160 linker that do not start with an underscore.
4162 It does not make sense to use this feature with a non-static local
4163 variable since such variables do not have assembler names. If you are
4164 trying to put the variable in a particular register, see @ref{Explicit
4165 Reg Vars}. GCC presently accepts such code with a warning, but will
4166 probably be changed to issue an error, rather than a warning, in the
4169 You cannot use @code{asm} in this way in a function @emph{definition}; but
4170 you can get the same effect by writing a declaration for the function
4171 before its definition and putting @code{asm} there, like this:
4174 extern func () asm ("FUNC");
4181 It is up to you to make sure that the assembler names you choose do not
4182 conflict with any other assembler symbols. Also, you must not use a
4183 register name; that would produce completely invalid assembler code. GCC
4184 does not as yet have the ability to store static variables in registers.
4185 Perhaps that will be added.
4187 @node Explicit Reg Vars
4188 @section Variables in Specified Registers
4189 @cindex explicit register variables
4190 @cindex variables in specified registers
4191 @cindex specified registers
4192 @cindex registers, global allocation
4194 GNU C allows you to put a few global variables into specified hardware
4195 registers. You can also specify the register in which an ordinary
4196 register variable should be allocated.
4200 Global register variables reserve registers throughout the program.
4201 This may be useful in programs such as programming language
4202 interpreters which have a couple of global variables that are accessed
4206 Local register variables in specific registers do not reserve the
4207 registers, except at the point where they are used as input or output
4208 operands in an @code{asm} statement and the @code{asm} statement itself is
4209 not deleted. The compiler's data flow analysis is capable of determining
4210 where the specified registers contain live values, and where they are
4211 available for other uses. Stores into local register variables may be deleted
4212 when they appear to be dead according to dataflow analysis. References
4213 to local register variables may be deleted or moved or simplified.
4215 These local variables are sometimes convenient for use with the extended
4216 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
4217 output of the assembler instruction directly into a particular register.
4218 (This will work provided the register you specify fits the constraints
4219 specified for that operand in the @code{asm}.)
4227 @node Global Reg Vars
4228 @subsection Defining Global Register Variables
4229 @cindex global register variables
4230 @cindex registers, global variables in
4232 You can define a global register variable in GNU C like this:
4235 register int *foo asm ("a5");
4239 Here @code{a5} is the name of the register which should be used. Choose a
4240 register which is normally saved and restored by function calls on your
4241 machine, so that library routines will not clobber it.
4243 Naturally the register name is cpu-dependent, so you would need to
4244 conditionalize your program according to cpu type. The register
4245 @code{a5} would be a good choice on a 68000 for a variable of pointer
4246 type. On machines with register windows, be sure to choose a ``global''
4247 register that is not affected magically by the function call mechanism.
4249 In addition, operating systems on one type of cpu may differ in how they
4250 name the registers; then you would need additional conditionals. For
4251 example, some 68000 operating systems call this register @code{%a5}.
4253 Eventually there may be a way of asking the compiler to choose a register
4254 automatically, but first we need to figure out how it should choose and
4255 how to enable you to guide the choice. No solution is evident.
4257 Defining a global register variable in a certain register reserves that
4258 register entirely for this use, at least within the current compilation.
4259 The register will not be allocated for any other purpose in the functions
4260 in the current compilation. The register will not be saved and restored by
4261 these functions. Stores into this register are never deleted even if they
4262 would appear to be dead, but references may be deleted or moved or
4265 It is not safe to access the global register variables from signal
4266 handlers, or from more than one thread of control, because the system
4267 library routines may temporarily use the register for other things (unless
4268 you recompile them specially for the task at hand).
4270 @cindex @code{qsort}, and global register variables
4271 It is not safe for one function that uses a global register variable to
4272 call another such function @code{foo} by way of a third function
4273 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4274 different source file in which the variable wasn't declared). This is
4275 because @code{lose} might save the register and put some other value there.
4276 For example, you can't expect a global register variable to be available in
4277 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4278 might have put something else in that register. (If you are prepared to
4279 recompile @code{qsort} with the same global register variable, you can
4280 solve this problem.)
4282 If you want to recompile @code{qsort} or other source files which do not
4283 actually use your global register variable, so that they will not use that
4284 register for any other purpose, then it suffices to specify the compiler
4285 option @option{-ffixed-@var{reg}}. You need not actually add a global
4286 register declaration to their source code.
4288 A function which can alter the value of a global register variable cannot
4289 safely be called from a function compiled without this variable, because it
4290 could clobber the value the caller expects to find there on return.
4291 Therefore, the function which is the entry point into the part of the
4292 program that uses the global register variable must explicitly save and
4293 restore the value which belongs to its caller.
4295 @cindex register variable after @code{longjmp}
4296 @cindex global register after @code{longjmp}
4297 @cindex value after @code{longjmp}
4300 On most machines, @code{longjmp} will restore to each global register
4301 variable the value it had at the time of the @code{setjmp}. On some
4302 machines, however, @code{longjmp} will not change the value of global
4303 register variables. To be portable, the function that called @code{setjmp}
4304 should make other arrangements to save the values of the global register
4305 variables, and to restore them in a @code{longjmp}. This way, the same
4306 thing will happen regardless of what @code{longjmp} does.
4308 All global register variable declarations must precede all function
4309 definitions. If such a declaration could appear after function
4310 definitions, the declaration would be too late to prevent the register from
4311 being used for other purposes in the preceding functions.
4313 Global register variables may not have initial values, because an
4314 executable file has no means to supply initial contents for a register.
4316 On the SPARC, there are reports that g3 @dots{} g7 are suitable
4317 registers, but certain library functions, such as @code{getwd}, as well
4318 as the subroutines for division and remainder, modify g3 and g4. g1 and
4319 g2 are local temporaries.
4321 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4322 Of course, it will not do to use more than a few of those.
4324 @node Local Reg Vars
4325 @subsection Specifying Registers for Local Variables
4326 @cindex local variables, specifying registers
4327 @cindex specifying registers for local variables
4328 @cindex registers for local variables
4330 You can define a local register variable with a specified register
4334 register int *foo asm ("a5");
4338 Here @code{a5} is the name of the register which should be used. Note
4339 that this is the same syntax used for defining global register
4340 variables, but for a local variable it would appear within a function.
4342 Naturally the register name is cpu-dependent, but this is not a
4343 problem, since specific registers are most often useful with explicit
4344 assembler instructions (@pxref{Extended Asm}). Both of these things
4345 generally require that you conditionalize your program according to
4348 In addition, operating systems on one type of cpu may differ in how they
4349 name the registers; then you would need additional conditionals. For
4350 example, some 68000 operating systems call this register @code{%a5}.
4352 Defining such a register variable does not reserve the register; it
4353 remains available for other uses in places where flow control determines
4354 the variable's value is not live.
4356 This option does not guarantee that GCC will generate code that has
4357 this variable in the register you specify at all times. You may not
4358 code an explicit reference to this register in the @emph{assembler
4359 instruction template} part of an @code{asm} statement and assume it will
4360 always refer to this variable. However, using the variable as an
4361 @code{asm} @emph{operand} guarantees that the specified register is used
4364 Stores into local register variables may be deleted when they appear to be dead
4365 according to dataflow analysis. References to local register variables may
4366 be deleted or moved or simplified.
4368 As for global register variables, it's recommended that you choose a
4369 register which is normally saved and restored by function calls on
4370 your machine, so that library routines will not clobber it. A common
4371 pitfall is to initialize multiple call-clobbered registers with
4372 arbitrary expressions, where a function call or library call for an
4373 arithmetic operator will overwrite a register value from a previous
4374 assignment, for example @code{r0} below:
4376 register int *p1 asm ("r0") = @dots{};
4377 register int *p2 asm ("r1") = @dots{};
4379 In those cases, a solution is to use a temporary variable for
4380 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
4382 @node Alternate Keywords
4383 @section Alternate Keywords
4384 @cindex alternate keywords
4385 @cindex keywords, alternate
4387 @option{-ansi} and the various @option{-std} options disable certain
4388 keywords. This causes trouble when you want to use GNU C extensions, or
4389 a general-purpose header file that should be usable by all programs,
4390 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4391 @code{inline} are not available in programs compiled with
4392 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4393 program compiled with @option{-std=c99}). The ISO C99 keyword
4394 @code{restrict} is only available when @option{-std=gnu99} (which will
4395 eventually be the default) or @option{-std=c99} (or the equivalent
4396 @option{-std=iso9899:1999}) is used.
4398 The way to solve these problems is to put @samp{__} at the beginning and
4399 end of each problematical keyword. For example, use @code{__asm__}
4400 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4402 Other C compilers won't accept these alternative keywords; if you want to
4403 compile with another compiler, you can define the alternate keywords as
4404 macros to replace them with the customary keywords. It looks like this:
4412 @findex __extension__
4414 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4416 prevent such warnings within one expression by writing
4417 @code{__extension__} before the expression. @code{__extension__} has no
4418 effect aside from this.
4420 @node Incomplete Enums
4421 @section Incomplete @code{enum} Types
4423 You can define an @code{enum} tag without specifying its possible values.
4424 This results in an incomplete type, much like what you get if you write
4425 @code{struct foo} without describing the elements. A later declaration
4426 which does specify the possible values completes the type.
4428 You can't allocate variables or storage using the type while it is
4429 incomplete. However, you can work with pointers to that type.
4431 This extension may not be very useful, but it makes the handling of
4432 @code{enum} more consistent with the way @code{struct} and @code{union}
4435 This extension is not supported by GNU C++.
4437 @node Function Names
4438 @section Function Names as Strings
4439 @cindex @code{__func__} identifier
4440 @cindex @code{__FUNCTION__} identifier
4441 @cindex @code{__PRETTY_FUNCTION__} identifier
4443 GCC provides three magic variables which hold the name of the current
4444 function, as a string. The first of these is @code{__func__}, which
4445 is part of the C99 standard:
4448 The identifier @code{__func__} is implicitly declared by the translator
4449 as if, immediately following the opening brace of each function
4450 definition, the declaration
4453 static const char __func__[] = "function-name";
4456 appeared, where function-name is the name of the lexically-enclosing
4457 function. This name is the unadorned name of the function.
4460 @code{__FUNCTION__} is another name for @code{__func__}. Older
4461 versions of GCC recognize only this name. However, it is not
4462 standardized. For maximum portability, we recommend you use
4463 @code{__func__}, but provide a fallback definition with the
4467 #if __STDC_VERSION__ < 199901L
4469 # define __func__ __FUNCTION__
4471 # define __func__ "<unknown>"
4476 In C, @code{__PRETTY_FUNCTION__} is yet another name for
4477 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
4478 the type signature of the function as well as its bare name. For
4479 example, this program:
4483 extern int printf (char *, ...);
4490 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4491 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4509 __PRETTY_FUNCTION__ = void a::sub(int)
4512 These identifiers are not preprocessor macros. In GCC 3.3 and
4513 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
4514 were treated as string literals; they could be used to initialize
4515 @code{char} arrays, and they could be concatenated with other string
4516 literals. GCC 3.4 and later treat them as variables, like
4517 @code{__func__}. In C++, @code{__FUNCTION__} and
4518 @code{__PRETTY_FUNCTION__} have always been variables.
4520 @node Return Address
4521 @section Getting the Return or Frame Address of a Function
4523 These functions may be used to get information about the callers of a
4526 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4527 This function returns the return address of the current function, or of
4528 one of its callers. The @var{level} argument is number of frames to
4529 scan up the call stack. A value of @code{0} yields the return address
4530 of the current function, a value of @code{1} yields the return address
4531 of the caller of the current function, and so forth. When inlining
4532 the expected behavior is that the function will return the address of
4533 the function that will be returned to. To work around this behavior use
4534 the @code{noinline} function attribute.
4536 The @var{level} argument must be a constant integer.
4538 On some machines it may be impossible to determine the return address of
4539 any function other than the current one; in such cases, or when the top
4540 of the stack has been reached, this function will return @code{0} or a
4541 random value. In addition, @code{__builtin_frame_address} may be used
4542 to determine if the top of the stack has been reached.
4544 This function should only be used with a nonzero argument for debugging
4548 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4549 This function is similar to @code{__builtin_return_address}, but it
4550 returns the address of the function frame rather than the return address
4551 of the function. Calling @code{__builtin_frame_address} with a value of
4552 @code{0} yields the frame address of the current function, a value of
4553 @code{1} yields the frame address of the caller of the current function,
4556 The frame is the area on the stack which holds local variables and saved
4557 registers. The frame address is normally the address of the first word
4558 pushed on to the stack by the function. However, the exact definition
4559 depends upon the processor and the calling convention. If the processor
4560 has a dedicated frame pointer register, and the function has a frame,
4561 then @code{__builtin_frame_address} will return the value of the frame
4564 On some machines it may be impossible to determine the frame address of
4565 any function other than the current one; in such cases, or when the top
4566 of the stack has been reached, this function will return @code{0} if
4567 the first frame pointer is properly initialized by the startup code.
4569 This function should only be used with a nonzero argument for debugging
4573 @node Vector Extensions
4574 @section Using vector instructions through built-in functions
4576 On some targets, the instruction set contains SIMD vector instructions that
4577 operate on multiple values contained in one large register at the same time.
4578 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4581 The first step in using these extensions is to provide the necessary data
4582 types. This should be done using an appropriate @code{typedef}:
4585 typedef int v4si __attribute__ ((vector_size (16)));
4588 The @code{int} type specifies the base type, while the attribute specifies
4589 the vector size for the variable, measured in bytes. For example, the
4590 declaration above causes the compiler to set the mode for the @code{v4si}
4591 type to be 16 bytes wide and divided into @code{int} sized units. For
4592 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
4593 corresponding mode of @code{foo} will be @acronym{V4SI}.
4595 The @code{vector_size} attribute is only applicable to integral and
4596 float scalars, although arrays, pointers, and function return values
4597 are allowed in conjunction with this construct.
4599 All the basic integer types can be used as base types, both as signed
4600 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
4601 @code{long long}. In addition, @code{float} and @code{double} can be
4602 used to build floating-point vector types.
4604 Specifying a combination that is not valid for the current architecture
4605 will cause GCC to synthesize the instructions using a narrower mode.
4606 For example, if you specify a variable of type @code{V4SI} and your
4607 architecture does not allow for this specific SIMD type, GCC will
4608 produce code that uses 4 @code{SIs}.
4610 The types defined in this manner can be used with a subset of normal C
4611 operations. Currently, GCC will allow using the following operators
4612 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~}@.
4614 The operations behave like C++ @code{valarrays}. Addition is defined as
4615 the addition of the corresponding elements of the operands. For
4616 example, in the code below, each of the 4 elements in @var{a} will be
4617 added to the corresponding 4 elements in @var{b} and the resulting
4618 vector will be stored in @var{c}.
4621 typedef int v4si __attribute__ ((vector_size (16)));
4628 Subtraction, multiplication, division, and the logical operations
4629 operate in a similar manner. Likewise, the result of using the unary
4630 minus or complement operators on a vector type is a vector whose
4631 elements are the negative or complemented values of the corresponding
4632 elements in the operand.
4634 You can declare variables and use them in function calls and returns, as
4635 well as in assignments and some casts. You can specify a vector type as
4636 a return type for a function. Vector types can also be used as function
4637 arguments. It is possible to cast from one vector type to another,
4638 provided they are of the same size (in fact, you can also cast vectors
4639 to and from other datatypes of the same size).
4641 You cannot operate between vectors of different lengths or different
4642 signedness without a cast.
4644 A port that supports hardware vector operations, usually provides a set
4645 of built-in functions that can be used to operate on vectors. For
4646 example, a function to add two vectors and multiply the result by a
4647 third could look like this:
4650 v4si f (v4si a, v4si b, v4si c)
4652 v4si tmp = __builtin_addv4si (a, b);
4653 return __builtin_mulv4si (tmp, c);
4660 @findex __builtin_offsetof
4662 GCC implements for both C and C++ a syntactic extension to implement
4663 the @code{offsetof} macro.
4667 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
4669 offsetof_member_designator:
4671 | offsetof_member_designator "." @code{identifier}
4672 | offsetof_member_designator "[" @code{expr} "]"
4675 This extension is sufficient such that
4678 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
4681 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
4682 may be dependent. In either case, @var{member} may consist of a single
4683 identifier, or a sequence of member accesses and array references.
4685 @node Atomic Builtins
4686 @section Built-in functions for atomic memory access
4688 The following builtins are intended to be compatible with those described
4689 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
4690 section 7.4. As such, they depart from the normal GCC practice of using
4691 the ``__builtin_'' prefix, and further that they are overloaded such that
4692 they work on multiple types.
4694 The definition given in the Intel documentation allows only for the use of
4695 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
4696 counterparts. GCC will allow any integral scalar or pointer type that is
4697 1, 2, 4 or 8 bytes in length.
4699 Not all operations are supported by all target processors. If a particular
4700 operation cannot be implemented on the target processor, a warning will be
4701 generated and a call an external function will be generated. The external
4702 function will carry the same name as the builtin, with an additional suffix
4703 @samp{_@var{n}} where @var{n} is the size of the data type.
4705 @c ??? Should we have a mechanism to suppress this warning? This is almost
4706 @c useful for implementing the operation under the control of an external
4709 In most cases, these builtins are considered a @dfn{full barrier}. That is,
4710 no memory operand will be moved across the operation, either forward or
4711 backward. Further, instructions will be issued as necessary to prevent the
4712 processor from speculating loads across the operation and from queuing stores
4713 after the operation.
4715 All of the routines are are described in the Intel documentation to take
4716 ``an optional list of variables protected by the memory barrier''. It's
4717 not clear what is meant by that; it could mean that @emph{only} the
4718 following variables are protected, or it could mean that these variables
4719 should in addition be protected. At present GCC ignores this list and
4720 protects all variables which are globally accessible. If in the future
4721 we make some use of this list, an empty list will continue to mean all
4722 globally accessible variables.
4725 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
4726 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
4727 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
4728 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
4729 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
4730 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
4731 @findex __sync_fetch_and_add
4732 @findex __sync_fetch_and_sub
4733 @findex __sync_fetch_and_or
4734 @findex __sync_fetch_and_and
4735 @findex __sync_fetch_and_xor
4736 @findex __sync_fetch_and_nand
4737 These builtins perform the operation suggested by the name, and
4738 returns the value that had previously been in memory. That is,
4741 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
4742 @{ tmp = *ptr; *ptr = ~tmp & value; return tmp; @} // nand
4745 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
4746 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
4747 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
4748 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
4749 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
4750 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
4751 @findex __sync_add_and_fetch
4752 @findex __sync_sub_and_fetch
4753 @findex __sync_or_and_fetch
4754 @findex __sync_and_and_fetch
4755 @findex __sync_xor_and_fetch
4756 @findex __sync_nand_and_fetch
4757 These builtins perform the operation suggested by the name, and
4758 return the new value. That is,
4761 @{ *ptr @var{op}= value; return *ptr; @}
4762 @{ *ptr = ~*ptr & value; return *ptr; @} // nand
4765 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
4766 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
4767 @findex __sync_bool_compare_and_swap
4768 @findex __sync_val_compare_and_swap
4769 These builtins perform an atomic compare and swap. That is, if the current
4770 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
4773 The ``bool'' version returns true if the comparison is successful and
4774 @var{newval} was written. The ``val'' version returns the contents
4775 of @code{*@var{ptr}} before the operation.
4777 @item __sync_synchronize (...)
4778 @findex __sync_synchronize
4779 This builtin issues a full memory barrier.
4781 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
4782 @findex __sync_lock_test_and_set
4783 This builtin, as described by Intel, is not a traditional test-and-set
4784 operation, but rather an atomic exchange operation. It writes @var{value}
4785 into @code{*@var{ptr}}, and returns the previous contents of
4788 Many targets have only minimal support for such locks, and do not support
4789 a full exchange operation. In this case, a target may support reduced
4790 functionality here by which the @emph{only} valid value to store is the
4791 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
4792 is implementation defined.
4794 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
4795 This means that references after the builtin cannot move to (or be
4796 speculated to) before the builtin, but previous memory stores may not
4797 be globally visible yet, and previous memory loads may not yet be
4800 @item void __sync_lock_release (@var{type} *ptr, ...)
4801 @findex __sync_lock_release
4802 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
4803 Normally this means writing the constant 0 to @code{*@var{ptr}}.
4805 This builtin is not a full barrier, but rather a @dfn{release barrier}.
4806 This means that all previous memory stores are globally visible, and all
4807 previous memory loads have been satisfied, but following memory reads
4808 are not prevented from being speculated to before the barrier.
4811 @node Object Size Checking
4812 @section Object Size Checking Builtins
4813 @findex __builtin_object_size
4814 @findex __builtin___memcpy_chk
4815 @findex __builtin___mempcpy_chk
4816 @findex __builtin___memmove_chk
4817 @findex __builtin___memset_chk
4818 @findex __builtin___strcpy_chk
4819 @findex __builtin___stpcpy_chk
4820 @findex __builtin___strncpy_chk
4821 @findex __builtin___strcat_chk
4822 @findex __builtin___strncat_chk
4823 @findex __builtin___sprintf_chk
4824 @findex __builtin___snprintf_chk
4825 @findex __builtin___vsprintf_chk
4826 @findex __builtin___vsnprintf_chk
4827 @findex __builtin___printf_chk
4828 @findex __builtin___vprintf_chk
4829 @findex __builtin___fprintf_chk
4830 @findex __builtin___vfprintf_chk
4832 GCC implements a limited buffer overflow protection mechanism
4833 that can prevent some buffer overflow attacks.
4835 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
4836 is a built-in construct that returns a constant number of bytes from
4837 @var{ptr} to the end of the object @var{ptr} pointer points to
4838 (if known at compile time). @code{__builtin_object_size} never evaluates
4839 its arguments for side-effects. If there are any side-effects in them, it
4840 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
4841 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
4842 point to and all of them are known at compile time, the returned number
4843 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
4844 0 and minimum if nonzero. If it is not possible to determine which objects
4845 @var{ptr} points to at compile time, @code{__builtin_object_size} should
4846 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
4847 for @var{type} 2 or 3.
4849 @var{type} is an integer constant from 0 to 3. If the least significant
4850 bit is clear, objects are whole variables, if it is set, a closest
4851 surrounding subobject is considered the object a pointer points to.
4852 The second bit determines if maximum or minimum of remaining bytes
4856 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
4857 char *p = &var.buf1[1], *q = &var.b;
4859 /* Here the object p points to is var. */
4860 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
4861 /* The subobject p points to is var.buf1. */
4862 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
4863 /* The object q points to is var. */
4864 assert (__builtin_object_size (q, 0)
4865 == (char *) (&var + 1) - (char *) &var.b);
4866 /* The subobject q points to is var.b. */
4867 assert (__builtin_object_size (q, 1) == sizeof (var.b));
4871 There are built-in functions added for many common string operation
4872 functions, e.g. for @code{memcpy} @code{__builtin___memcpy_chk}
4873 built-in is provided. This built-in has an additional last argument,
4874 which is the number of bytes remaining in object the @var{dest}
4875 argument points to or @code{(size_t) -1} if the size is not known.
4877 The built-in functions are optimized into the normal string functions
4878 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
4879 it is known at compile time that the destination object will not
4880 be overflown. If the compiler can determine at compile time the
4881 object will be always overflown, it issues a warning.
4883 The intended use can be e.g.
4887 #define bos0(dest) __builtin_object_size (dest, 0)
4888 #define memcpy(dest, src, n) \
4889 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
4893 /* It is unknown what object p points to, so this is optimized
4894 into plain memcpy - no checking is possible. */
4895 memcpy (p, "abcde", n);
4896 /* Destination is known and length too. It is known at compile
4897 time there will be no overflow. */
4898 memcpy (&buf[5], "abcde", 5);
4899 /* Destination is known, but the length is not known at compile time.
4900 This will result in __memcpy_chk call that can check for overflow
4902 memcpy (&buf[5], "abcde", n);
4903 /* Destination is known and it is known at compile time there will
4904 be overflow. There will be a warning and __memcpy_chk call that
4905 will abort the program at runtime. */
4906 memcpy (&buf[6], "abcde", 5);
4909 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
4910 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
4911 @code{strcat} and @code{strncat}.
4913 There are also checking built-in functions for formatted output functions.
4915 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
4916 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
4917 const char *fmt, ...);
4918 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
4920 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
4921 const char *fmt, va_list ap);
4924 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
4925 etc. functions and can contain implementation specific flags on what
4926 additional security measures the checking function might take, such as
4927 handling @code{%n} differently.
4929 The @var{os} argument is the object size @var{s} points to, like in the
4930 other built-in functions. There is a small difference in the behavior
4931 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
4932 optimized into the non-checking functions only if @var{flag} is 0, otherwise
4933 the checking function is called with @var{os} argument set to
4936 In addition to this, there are checking built-in functions
4937 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
4938 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
4939 These have just one additional argument, @var{flag}, right before
4940 format string @var{fmt}. If the compiler is able to optimize them to
4941 @code{fputc} etc. functions, it will, otherwise the checking function
4942 should be called and the @var{flag} argument passed to it.
4944 @node Other Builtins
4945 @section Other built-in functions provided by GCC
4946 @cindex built-in functions
4947 @findex __builtin_isgreater
4948 @findex __builtin_isgreaterequal
4949 @findex __builtin_isless
4950 @findex __builtin_islessequal
4951 @findex __builtin_islessgreater
4952 @findex __builtin_isunordered
4953 @findex __builtin_powi
4954 @findex __builtin_powif
4955 @findex __builtin_powil
5113 @findex fprintf_unlocked
5115 @findex fputs_unlocked
5225 @findex printf_unlocked
5254 @findex significandf
5255 @findex significandl
5326 GCC provides a large number of built-in functions other than the ones
5327 mentioned above. Some of these are for internal use in the processing
5328 of exceptions or variable-length argument lists and will not be
5329 documented here because they may change from time to time; we do not
5330 recommend general use of these functions.
5332 The remaining functions are provided for optimization purposes.
5334 @opindex fno-builtin
5335 GCC includes built-in versions of many of the functions in the standard
5336 C library. The versions prefixed with @code{__builtin_} will always be
5337 treated as having the same meaning as the C library function even if you
5338 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
5339 Many of these functions are only optimized in certain cases; if they are
5340 not optimized in a particular case, a call to the library function will
5345 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
5346 @option{-std=c99}), the functions
5347 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
5348 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
5349 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
5350 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked}, @code{fputs_unlocked},
5351 @code{gammaf}, @code{gammal}, @code{gamma}, @code{gettext},
5352 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
5353 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
5354 @code{mempcpy}, @code{pow10f}, @code{pow10l}, @code{pow10},
5355 @code{printf_unlocked}, @code{rindex}, @code{scalbf}, @code{scalbl},
5356 @code{scalb}, @code{signbit}, @code{signbitf}, @code{signbitl},
5357 @code{significandf}, @code{significandl}, @code{significand},
5358 @code{sincosf}, @code{sincosl}, @code{sincos}, @code{stpcpy},
5359 @code{stpncpy}, @code{strcasecmp}, @code{strdup}, @code{strfmon},
5360 @code{strncasecmp}, @code{strndup}, @code{toascii}, @code{y0f},
5361 @code{y0l}, @code{y0}, @code{y1f}, @code{y1l}, @code{y1}, @code{ynf},
5362 @code{ynl} and @code{yn}
5363 may be handled as built-in functions.
5364 All these functions have corresponding versions
5365 prefixed with @code{__builtin_}, which may be used even in strict C89
5368 The ISO C99 functions
5369 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
5370 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
5371 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
5372 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
5373 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
5374 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
5375 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
5376 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
5377 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
5378 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
5379 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
5380 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
5381 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
5382 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
5383 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
5384 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
5385 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
5386 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
5387 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
5388 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
5389 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
5390 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
5391 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
5392 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
5393 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
5394 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
5395 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
5396 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
5397 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
5398 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
5399 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
5400 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
5401 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
5402 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
5403 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
5404 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
5405 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
5406 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
5407 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
5408 are handled as built-in functions
5409 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5411 There are also built-in versions of the ISO C99 functions
5412 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
5413 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
5414 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
5415 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
5416 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
5417 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
5418 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
5419 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
5420 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
5421 that are recognized in any mode since ISO C90 reserves these names for
5422 the purpose to which ISO C99 puts them. All these functions have
5423 corresponding versions prefixed with @code{__builtin_}.
5425 The ISO C94 functions
5426 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
5427 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
5428 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
5430 are handled as built-in functions
5431 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5433 The ISO C90 functions
5434 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
5435 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
5436 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
5437 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
5438 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
5439 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
5440 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
5441 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
5442 @code{malloc}, @code{memcmp}, @code{memcpy}, @code{memset}, @code{modf},
5443 @code{pow}, @code{printf}, @code{putchar}, @code{puts}, @code{scanf},
5444 @code{sinh}, @code{sin}, @code{snprintf}, @code{sprintf}, @code{sqrt},
5445 @code{sscanf}, @code{strcat}, @code{strchr}, @code{strcmp},
5446 @code{strcpy}, @code{strcspn}, @code{strlen}, @code{strncat},
5447 @code{strncmp}, @code{strncpy}, @code{strpbrk}, @code{strrchr},
5448 @code{strspn}, @code{strstr}, @code{tanh}, @code{tan}, @code{vfprintf},
5449 @code{vprintf} and @code{vsprintf}
5450 are all recognized as built-in functions unless
5451 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
5452 is specified for an individual function). All of these functions have
5453 corresponding versions prefixed with @code{__builtin_}.
5455 GCC provides built-in versions of the ISO C99 floating point comparison
5456 macros that avoid raising exceptions for unordered operands. They have
5457 the same names as the standard macros ( @code{isgreater},
5458 @code{isgreaterequal}, @code{isless}, @code{islessequal},
5459 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
5460 prefixed. We intend for a library implementor to be able to simply
5461 @code{#define} each standard macro to its built-in equivalent.
5463 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
5465 You can use the built-in function @code{__builtin_types_compatible_p} to
5466 determine whether two types are the same.
5468 This built-in function returns 1 if the unqualified versions of the
5469 types @var{type1} and @var{type2} (which are types, not expressions) are
5470 compatible, 0 otherwise. The result of this built-in function can be
5471 used in integer constant expressions.
5473 This built-in function ignores top level qualifiers (e.g., @code{const},
5474 @code{volatile}). For example, @code{int} is equivalent to @code{const
5477 The type @code{int[]} and @code{int[5]} are compatible. On the other
5478 hand, @code{int} and @code{char *} are not compatible, even if the size
5479 of their types, on the particular architecture are the same. Also, the
5480 amount of pointer indirection is taken into account when determining
5481 similarity. Consequently, @code{short *} is not similar to
5482 @code{short **}. Furthermore, two types that are typedefed are
5483 considered compatible if their underlying types are compatible.
5485 An @code{enum} type is not considered to be compatible with another
5486 @code{enum} type even if both are compatible with the same integer
5487 type; this is what the C standard specifies.
5488 For example, @code{enum @{foo, bar@}} is not similar to
5489 @code{enum @{hot, dog@}}.
5491 You would typically use this function in code whose execution varies
5492 depending on the arguments' types. For example:
5498 if (__builtin_types_compatible_p (typeof (x), long double)) \
5499 tmp = foo_long_double (tmp); \
5500 else if (__builtin_types_compatible_p (typeof (x), double)) \
5501 tmp = foo_double (tmp); \
5502 else if (__builtin_types_compatible_p (typeof (x), float)) \
5503 tmp = foo_float (tmp); \
5510 @emph{Note:} This construct is only available for C@.
5514 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
5516 You can use the built-in function @code{__builtin_choose_expr} to
5517 evaluate code depending on the value of a constant expression. This
5518 built-in function returns @var{exp1} if @var{const_exp}, which is a
5519 constant expression that must be able to be determined at compile time,
5520 is nonzero. Otherwise it returns 0.
5522 This built-in function is analogous to the @samp{? :} operator in C,
5523 except that the expression returned has its type unaltered by promotion
5524 rules. Also, the built-in function does not evaluate the expression
5525 that was not chosen. For example, if @var{const_exp} evaluates to true,
5526 @var{exp2} is not evaluated even if it has side-effects.
5528 This built-in function can return an lvalue if the chosen argument is an
5531 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
5532 type. Similarly, if @var{exp2} is returned, its return type is the same
5539 __builtin_choose_expr ( \
5540 __builtin_types_compatible_p (typeof (x), double), \
5542 __builtin_choose_expr ( \
5543 __builtin_types_compatible_p (typeof (x), float), \
5545 /* @r{The void expression results in a compile-time error} \
5546 @r{when assigning the result to something.} */ \
5550 @emph{Note:} This construct is only available for C@. Furthermore, the
5551 unused expression (@var{exp1} or @var{exp2} depending on the value of
5552 @var{const_exp}) may still generate syntax errors. This may change in
5557 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
5558 You can use the built-in function @code{__builtin_constant_p} to
5559 determine if a value is known to be constant at compile-time and hence
5560 that GCC can perform constant-folding on expressions involving that
5561 value. The argument of the function is the value to test. The function
5562 returns the integer 1 if the argument is known to be a compile-time
5563 constant and 0 if it is not known to be a compile-time constant. A
5564 return of 0 does not indicate that the value is @emph{not} a constant,
5565 but merely that GCC cannot prove it is a constant with the specified
5566 value of the @option{-O} option.
5568 You would typically use this function in an embedded application where
5569 memory was a critical resource. If you have some complex calculation,
5570 you may want it to be folded if it involves constants, but need to call
5571 a function if it does not. For example:
5574 #define Scale_Value(X) \
5575 (__builtin_constant_p (X) \
5576 ? ((X) * SCALE + OFFSET) : Scale (X))
5579 You may use this built-in function in either a macro or an inline
5580 function. However, if you use it in an inlined function and pass an
5581 argument of the function as the argument to the built-in, GCC will
5582 never return 1 when you call the inline function with a string constant
5583 or compound literal (@pxref{Compound Literals}) and will not return 1
5584 when you pass a constant numeric value to the inline function unless you
5585 specify the @option{-O} option.
5587 You may also use @code{__builtin_constant_p} in initializers for static
5588 data. For instance, you can write
5591 static const int table[] = @{
5592 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
5598 This is an acceptable initializer even if @var{EXPRESSION} is not a
5599 constant expression. GCC must be more conservative about evaluating the
5600 built-in in this case, because it has no opportunity to perform
5603 Previous versions of GCC did not accept this built-in in data
5604 initializers. The earliest version where it is completely safe is
5608 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
5609 @opindex fprofile-arcs
5610 You may use @code{__builtin_expect} to provide the compiler with
5611 branch prediction information. In general, you should prefer to
5612 use actual profile feedback for this (@option{-fprofile-arcs}), as
5613 programmers are notoriously bad at predicting how their programs
5614 actually perform. However, there are applications in which this
5615 data is hard to collect.
5617 The return value is the value of @var{exp}, which should be an
5618 integral expression. The value of @var{c} must be a compile-time
5619 constant. The semantics of the built-in are that it is expected
5620 that @var{exp} == @var{c}. For example:
5623 if (__builtin_expect (x, 0))
5628 would indicate that we do not expect to call @code{foo}, since
5629 we expect @code{x} to be zero. Since you are limited to integral
5630 expressions for @var{exp}, you should use constructions such as
5633 if (__builtin_expect (ptr != NULL, 1))
5638 when testing pointer or floating-point values.
5641 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
5642 This function is used to minimize cache-miss latency by moving data into
5643 a cache before it is accessed.
5644 You can insert calls to @code{__builtin_prefetch} into code for which
5645 you know addresses of data in memory that is likely to be accessed soon.
5646 If the target supports them, data prefetch instructions will be generated.
5647 If the prefetch is done early enough before the access then the data will
5648 be in the cache by the time it is accessed.
5650 The value of @var{addr} is the address of the memory to prefetch.
5651 There are two optional arguments, @var{rw} and @var{locality}.
5652 The value of @var{rw} is a compile-time constant one or zero; one
5653 means that the prefetch is preparing for a write to the memory address
5654 and zero, the default, means that the prefetch is preparing for a read.
5655 The value @var{locality} must be a compile-time constant integer between
5656 zero and three. A value of zero means that the data has no temporal
5657 locality, so it need not be left in the cache after the access. A value
5658 of three means that the data has a high degree of temporal locality and
5659 should be left in all levels of cache possible. Values of one and two
5660 mean, respectively, a low or moderate degree of temporal locality. The
5664 for (i = 0; i < n; i++)
5667 __builtin_prefetch (&a[i+j], 1, 1);
5668 __builtin_prefetch (&b[i+j], 0, 1);
5673 Data prefetch does not generate faults if @var{addr} is invalid, but
5674 the address expression itself must be valid. For example, a prefetch
5675 of @code{p->next} will not fault if @code{p->next} is not a valid
5676 address, but evaluation will fault if @code{p} is not a valid address.
5678 If the target does not support data prefetch, the address expression
5679 is evaluated if it includes side effects but no other code is generated
5680 and GCC does not issue a warning.
5683 @deftypefn {Built-in Function} double __builtin_huge_val (void)
5684 Returns a positive infinity, if supported by the floating-point format,
5685 else @code{DBL_MAX}. This function is suitable for implementing the
5686 ISO C macro @code{HUGE_VAL}.
5689 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
5690 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
5693 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
5694 Similar to @code{__builtin_huge_val}, except the return
5695 type is @code{long double}.
5698 @deftypefn {Built-in Function} double __builtin_inf (void)
5699 Similar to @code{__builtin_huge_val}, except a warning is generated
5700 if the target floating-point format does not support infinities.
5703 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
5704 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
5707 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
5708 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
5711 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
5712 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
5715 @deftypefn {Built-in Function} float __builtin_inff (void)
5716 Similar to @code{__builtin_inf}, except the return type is @code{float}.
5717 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
5720 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
5721 Similar to @code{__builtin_inf}, except the return
5722 type is @code{long double}.
5725 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
5726 This is an implementation of the ISO C99 function @code{nan}.
5728 Since ISO C99 defines this function in terms of @code{strtod}, which we
5729 do not implement, a description of the parsing is in order. The string
5730 is parsed as by @code{strtol}; that is, the base is recognized by
5731 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
5732 in the significand such that the least significant bit of the number
5733 is at the least significant bit of the significand. The number is
5734 truncated to fit the significand field provided. The significand is
5735 forced to be a quiet NaN@.
5737 This function, if given a string literal, is evaluated early enough
5738 that it is considered a compile-time constant.
5741 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
5742 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
5745 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
5746 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
5749 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
5750 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
5753 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
5754 Similar to @code{__builtin_nan}, except the return type is @code{float}.
5757 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
5758 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
5761 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
5762 Similar to @code{__builtin_nan}, except the significand is forced
5763 to be a signaling NaN@. The @code{nans} function is proposed by
5764 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
5767 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
5768 Similar to @code{__builtin_nans}, except the return type is @code{float}.
5771 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
5772 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
5775 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
5776 Returns one plus the index of the least significant 1-bit of @var{x}, or
5777 if @var{x} is zero, returns zero.
5780 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
5781 Returns the number of leading 0-bits in @var{x}, starting at the most
5782 significant bit position. If @var{x} is 0, the result is undefined.
5785 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
5786 Returns the number of trailing 0-bits in @var{x}, starting at the least
5787 significant bit position. If @var{x} is 0, the result is undefined.
5790 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
5791 Returns the number of 1-bits in @var{x}.
5794 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
5795 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
5799 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
5800 Similar to @code{__builtin_ffs}, except the argument type is
5801 @code{unsigned long}.
5804 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
5805 Similar to @code{__builtin_clz}, except the argument type is
5806 @code{unsigned long}.
5809 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
5810 Similar to @code{__builtin_ctz}, except the argument type is
5811 @code{unsigned long}.
5814 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
5815 Similar to @code{__builtin_popcount}, except the argument type is
5816 @code{unsigned long}.
5819 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
5820 Similar to @code{__builtin_parity}, except the argument type is
5821 @code{unsigned long}.
5824 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
5825 Similar to @code{__builtin_ffs}, except the argument type is
5826 @code{unsigned long long}.
5829 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
5830 Similar to @code{__builtin_clz}, except the argument type is
5831 @code{unsigned long long}.
5834 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
5835 Similar to @code{__builtin_ctz}, except the argument type is
5836 @code{unsigned long long}.
5839 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
5840 Similar to @code{__builtin_popcount}, except the argument type is
5841 @code{unsigned long long}.
5844 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
5845 Similar to @code{__builtin_parity}, except the argument type is
5846 @code{unsigned long long}.
5849 @deftypefn {Built-in Function} double __builtin_powi (double, int)
5850 Returns the first argument raised to the power of the second. Unlike the
5851 @code{pow} function no guarantees about precision and rounding are made.
5854 @deftypefn {Built-in Function} float __builtin_powif (float, int)
5855 Similar to @code{__builtin_powi}, except the argument and return types
5859 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
5860 Similar to @code{__builtin_powi}, except the argument and return types
5861 are @code{long double}.
5865 @node Target Builtins
5866 @section Built-in Functions Specific to Particular Target Machines
5868 On some target machines, GCC supports many built-in functions specific
5869 to those machines. Generally these generate calls to specific machine
5870 instructions, but allow the compiler to schedule those calls.
5873 * Alpha Built-in Functions::
5874 * ARM Built-in Functions::
5875 * Blackfin Built-in Functions::
5876 * FR-V Built-in Functions::
5877 * X86 Built-in Functions::
5878 * MIPS DSP Built-in Functions::
5879 * MIPS Paired-Single Support::
5880 * PowerPC AltiVec Built-in Functions::
5881 * SPARC VIS Built-in Functions::
5884 @node Alpha Built-in Functions
5885 @subsection Alpha Built-in Functions
5887 These built-in functions are available for the Alpha family of
5888 processors, depending on the command-line switches used.
5890 The following built-in functions are always available. They
5891 all generate the machine instruction that is part of the name.
5894 long __builtin_alpha_implver (void)
5895 long __builtin_alpha_rpcc (void)
5896 long __builtin_alpha_amask (long)
5897 long __builtin_alpha_cmpbge (long, long)
5898 long __builtin_alpha_extbl (long, long)
5899 long __builtin_alpha_extwl (long, long)
5900 long __builtin_alpha_extll (long, long)
5901 long __builtin_alpha_extql (long, long)
5902 long __builtin_alpha_extwh (long, long)
5903 long __builtin_alpha_extlh (long, long)
5904 long __builtin_alpha_extqh (long, long)
5905 long __builtin_alpha_insbl (long, long)
5906 long __builtin_alpha_inswl (long, long)
5907 long __builtin_alpha_insll (long, long)
5908 long __builtin_alpha_insql (long, long)
5909 long __builtin_alpha_inswh (long, long)
5910 long __builtin_alpha_inslh (long, long)
5911 long __builtin_alpha_insqh (long, long)
5912 long __builtin_alpha_mskbl (long, long)
5913 long __builtin_alpha_mskwl (long, long)
5914 long __builtin_alpha_mskll (long, long)
5915 long __builtin_alpha_mskql (long, long)
5916 long __builtin_alpha_mskwh (long, long)
5917 long __builtin_alpha_msklh (long, long)
5918 long __builtin_alpha_mskqh (long, long)
5919 long __builtin_alpha_umulh (long, long)
5920 long __builtin_alpha_zap (long, long)
5921 long __builtin_alpha_zapnot (long, long)
5924 The following built-in functions are always with @option{-mmax}
5925 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
5926 later. They all generate the machine instruction that is part
5930 long __builtin_alpha_pklb (long)
5931 long __builtin_alpha_pkwb (long)
5932 long __builtin_alpha_unpkbl (long)
5933 long __builtin_alpha_unpkbw (long)
5934 long __builtin_alpha_minub8 (long, long)
5935 long __builtin_alpha_minsb8 (long, long)
5936 long __builtin_alpha_minuw4 (long, long)
5937 long __builtin_alpha_minsw4 (long, long)
5938 long __builtin_alpha_maxub8 (long, long)
5939 long __builtin_alpha_maxsb8 (long, long)
5940 long __builtin_alpha_maxuw4 (long, long)
5941 long __builtin_alpha_maxsw4 (long, long)
5942 long __builtin_alpha_perr (long, long)
5945 The following built-in functions are always with @option{-mcix}
5946 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
5947 later. They all generate the machine instruction that is part
5951 long __builtin_alpha_cttz (long)
5952 long __builtin_alpha_ctlz (long)
5953 long __builtin_alpha_ctpop (long)
5956 The following builtins are available on systems that use the OSF/1
5957 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
5958 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
5959 @code{rdval} and @code{wrval}.
5962 void *__builtin_thread_pointer (void)
5963 void __builtin_set_thread_pointer (void *)
5966 @node ARM Built-in Functions
5967 @subsection ARM Built-in Functions
5969 These built-in functions are available for the ARM family of
5970 processors, when the @option{-mcpu=iwmmxt} switch is used:
5973 typedef int v2si __attribute__ ((vector_size (8)));
5974 typedef short v4hi __attribute__ ((vector_size (8)));
5975 typedef char v8qi __attribute__ ((vector_size (8)));
5977 int __builtin_arm_getwcx (int)
5978 void __builtin_arm_setwcx (int, int)
5979 int __builtin_arm_textrmsb (v8qi, int)
5980 int __builtin_arm_textrmsh (v4hi, int)
5981 int __builtin_arm_textrmsw (v2si, int)
5982 int __builtin_arm_textrmub (v8qi, int)
5983 int __builtin_arm_textrmuh (v4hi, int)
5984 int __builtin_arm_textrmuw (v2si, int)
5985 v8qi __builtin_arm_tinsrb (v8qi, int)
5986 v4hi __builtin_arm_tinsrh (v4hi, int)
5987 v2si __builtin_arm_tinsrw (v2si, int)
5988 long long __builtin_arm_tmia (long long, int, int)
5989 long long __builtin_arm_tmiabb (long long, int, int)
5990 long long __builtin_arm_tmiabt (long long, int, int)
5991 long long __builtin_arm_tmiaph (long long, int, int)
5992 long long __builtin_arm_tmiatb (long long, int, int)
5993 long long __builtin_arm_tmiatt (long long, int, int)
5994 int __builtin_arm_tmovmskb (v8qi)
5995 int __builtin_arm_tmovmskh (v4hi)
5996 int __builtin_arm_tmovmskw (v2si)
5997 long long __builtin_arm_waccb (v8qi)
5998 long long __builtin_arm_wacch (v4hi)
5999 long long __builtin_arm_waccw (v2si)
6000 v8qi __builtin_arm_waddb (v8qi, v8qi)
6001 v8qi __builtin_arm_waddbss (v8qi, v8qi)
6002 v8qi __builtin_arm_waddbus (v8qi, v8qi)
6003 v4hi __builtin_arm_waddh (v4hi, v4hi)
6004 v4hi __builtin_arm_waddhss (v4hi, v4hi)
6005 v4hi __builtin_arm_waddhus (v4hi, v4hi)
6006 v2si __builtin_arm_waddw (v2si, v2si)
6007 v2si __builtin_arm_waddwss (v2si, v2si)
6008 v2si __builtin_arm_waddwus (v2si, v2si)
6009 v8qi __builtin_arm_walign (v8qi, v8qi, int)
6010 long long __builtin_arm_wand(long long, long long)
6011 long long __builtin_arm_wandn (long long, long long)
6012 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
6013 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
6014 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
6015 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
6016 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
6017 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
6018 v2si __builtin_arm_wcmpeqw (v2si, v2si)
6019 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
6020 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
6021 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
6022 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
6023 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
6024 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
6025 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
6026 long long __builtin_arm_wmacsz (v4hi, v4hi)
6027 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
6028 long long __builtin_arm_wmacuz (v4hi, v4hi)
6029 v4hi __builtin_arm_wmadds (v4hi, v4hi)
6030 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
6031 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
6032 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
6033 v2si __builtin_arm_wmaxsw (v2si, v2si)
6034 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
6035 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
6036 v2si __builtin_arm_wmaxuw (v2si, v2si)
6037 v8qi __builtin_arm_wminsb (v8qi, v8qi)
6038 v4hi __builtin_arm_wminsh (v4hi, v4hi)
6039 v2si __builtin_arm_wminsw (v2si, v2si)
6040 v8qi __builtin_arm_wminub (v8qi, v8qi)
6041 v4hi __builtin_arm_wminuh (v4hi, v4hi)
6042 v2si __builtin_arm_wminuw (v2si, v2si)
6043 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
6044 v4hi __builtin_arm_wmulul (v4hi, v4hi)
6045 v4hi __builtin_arm_wmulum (v4hi, v4hi)
6046 long long __builtin_arm_wor (long long, long long)
6047 v2si __builtin_arm_wpackdss (long long, long long)
6048 v2si __builtin_arm_wpackdus (long long, long long)
6049 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
6050 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
6051 v4hi __builtin_arm_wpackwss (v2si, v2si)
6052 v4hi __builtin_arm_wpackwus (v2si, v2si)
6053 long long __builtin_arm_wrord (long long, long long)
6054 long long __builtin_arm_wrordi (long long, int)
6055 v4hi __builtin_arm_wrorh (v4hi, long long)
6056 v4hi __builtin_arm_wrorhi (v4hi, int)
6057 v2si __builtin_arm_wrorw (v2si, long long)
6058 v2si __builtin_arm_wrorwi (v2si, int)
6059 v2si __builtin_arm_wsadb (v8qi, v8qi)
6060 v2si __builtin_arm_wsadbz (v8qi, v8qi)
6061 v2si __builtin_arm_wsadh (v4hi, v4hi)
6062 v2si __builtin_arm_wsadhz (v4hi, v4hi)
6063 v4hi __builtin_arm_wshufh (v4hi, int)
6064 long long __builtin_arm_wslld (long long, long long)
6065 long long __builtin_arm_wslldi (long long, int)
6066 v4hi __builtin_arm_wsllh (v4hi, long long)
6067 v4hi __builtin_arm_wsllhi (v4hi, int)
6068 v2si __builtin_arm_wsllw (v2si, long long)
6069 v2si __builtin_arm_wsllwi (v2si, int)
6070 long long __builtin_arm_wsrad (long long, long long)
6071 long long __builtin_arm_wsradi (long long, int)
6072 v4hi __builtin_arm_wsrah (v4hi, long long)
6073 v4hi __builtin_arm_wsrahi (v4hi, int)
6074 v2si __builtin_arm_wsraw (v2si, long long)
6075 v2si __builtin_arm_wsrawi (v2si, int)
6076 long long __builtin_arm_wsrld (long long, long long)
6077 long long __builtin_arm_wsrldi (long long, int)
6078 v4hi __builtin_arm_wsrlh (v4hi, long long)
6079 v4hi __builtin_arm_wsrlhi (v4hi, int)
6080 v2si __builtin_arm_wsrlw (v2si, long long)
6081 v2si __builtin_arm_wsrlwi (v2si, int)
6082 v8qi __builtin_arm_wsubb (v8qi, v8qi)
6083 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
6084 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
6085 v4hi __builtin_arm_wsubh (v4hi, v4hi)
6086 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
6087 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
6088 v2si __builtin_arm_wsubw (v2si, v2si)
6089 v2si __builtin_arm_wsubwss (v2si, v2si)
6090 v2si __builtin_arm_wsubwus (v2si, v2si)
6091 v4hi __builtin_arm_wunpckehsb (v8qi)
6092 v2si __builtin_arm_wunpckehsh (v4hi)
6093 long long __builtin_arm_wunpckehsw (v2si)
6094 v4hi __builtin_arm_wunpckehub (v8qi)
6095 v2si __builtin_arm_wunpckehuh (v4hi)
6096 long long __builtin_arm_wunpckehuw (v2si)
6097 v4hi __builtin_arm_wunpckelsb (v8qi)
6098 v2si __builtin_arm_wunpckelsh (v4hi)
6099 long long __builtin_arm_wunpckelsw (v2si)
6100 v4hi __builtin_arm_wunpckelub (v8qi)
6101 v2si __builtin_arm_wunpckeluh (v4hi)
6102 long long __builtin_arm_wunpckeluw (v2si)
6103 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
6104 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
6105 v2si __builtin_arm_wunpckihw (v2si, v2si)
6106 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
6107 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
6108 v2si __builtin_arm_wunpckilw (v2si, v2si)
6109 long long __builtin_arm_wxor (long long, long long)
6110 long long __builtin_arm_wzero ()
6113 @node Blackfin Built-in Functions
6114 @subsection Blackfin Built-in Functions
6116 Currently, there are two Blackfin-specific built-in functions. These are
6117 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
6118 using inline assembly; by using these built-in functions the compiler can
6119 automatically add workarounds for hardware errata involving these
6120 instructions. These functions are named as follows:
6123 void __builtin_bfin_csync (void)
6124 void __builtin_bfin_ssync (void)
6127 @node FR-V Built-in Functions
6128 @subsection FR-V Built-in Functions
6130 GCC provides many FR-V-specific built-in functions. In general,
6131 these functions are intended to be compatible with those described
6132 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
6133 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
6134 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
6135 pointer rather than by value.
6137 Most of the functions are named after specific FR-V instructions.
6138 Such functions are said to be ``directly mapped'' and are summarized
6139 here in tabular form.
6143 * Directly-mapped Integer Functions::
6144 * Directly-mapped Media Functions::
6145 * Raw read/write Functions::
6146 * Other Built-in Functions::
6149 @node Argument Types
6150 @subsubsection Argument Types
6152 The arguments to the built-in functions can be divided into three groups:
6153 register numbers, compile-time constants and run-time values. In order
6154 to make this classification clear at a glance, the arguments and return
6155 values are given the following pseudo types:
6157 @multitable @columnfractions .20 .30 .15 .35
6158 @item Pseudo type @tab Real C type @tab Constant? @tab Description
6159 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
6160 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
6161 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
6162 @item @code{uw2} @tab @code{unsigned long long} @tab No
6163 @tab an unsigned doubleword
6164 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
6165 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
6166 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
6167 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
6170 These pseudo types are not defined by GCC, they are simply a notational
6171 convenience used in this manual.
6173 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
6174 and @code{sw2} are evaluated at run time. They correspond to
6175 register operands in the underlying FR-V instructions.
6177 @code{const} arguments represent immediate operands in the underlying
6178 FR-V instructions. They must be compile-time constants.
6180 @code{acc} arguments are evaluated at compile time and specify the number
6181 of an accumulator register. For example, an @code{acc} argument of 2
6182 will select the ACC2 register.
6184 @code{iacc} arguments are similar to @code{acc} arguments but specify the
6185 number of an IACC register. See @pxref{Other Built-in Functions}
6188 @node Directly-mapped Integer Functions
6189 @subsubsection Directly-mapped Integer Functions
6191 The functions listed below map directly to FR-V I-type instructions.
6193 @multitable @columnfractions .45 .32 .23
6194 @item Function prototype @tab Example usage @tab Assembly output
6195 @item @code{sw1 __ADDSS (sw1, sw1)}
6196 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
6197 @tab @code{ADDSS @var{a},@var{b},@var{c}}
6198 @item @code{sw1 __SCAN (sw1, sw1)}
6199 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
6200 @tab @code{SCAN @var{a},@var{b},@var{c}}
6201 @item @code{sw1 __SCUTSS (sw1)}
6202 @tab @code{@var{b} = __SCUTSS (@var{a})}
6203 @tab @code{SCUTSS @var{a},@var{b}}
6204 @item @code{sw1 __SLASS (sw1, sw1)}
6205 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
6206 @tab @code{SLASS @var{a},@var{b},@var{c}}
6207 @item @code{void __SMASS (sw1, sw1)}
6208 @tab @code{__SMASS (@var{a}, @var{b})}
6209 @tab @code{SMASS @var{a},@var{b}}
6210 @item @code{void __SMSSS (sw1, sw1)}
6211 @tab @code{__SMSSS (@var{a}, @var{b})}
6212 @tab @code{SMSSS @var{a},@var{b}}
6213 @item @code{void __SMU (sw1, sw1)}
6214 @tab @code{__SMU (@var{a}, @var{b})}
6215 @tab @code{SMU @var{a},@var{b}}
6216 @item @code{sw2 __SMUL (sw1, sw1)}
6217 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
6218 @tab @code{SMUL @var{a},@var{b},@var{c}}
6219 @item @code{sw1 __SUBSS (sw1, sw1)}
6220 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
6221 @tab @code{SUBSS @var{a},@var{b},@var{c}}
6222 @item @code{uw2 __UMUL (uw1, uw1)}
6223 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
6224 @tab @code{UMUL @var{a},@var{b},@var{c}}
6227 @node Directly-mapped Media Functions
6228 @subsubsection Directly-mapped Media Functions
6230 The functions listed below map directly to FR-V M-type instructions.
6232 @multitable @columnfractions .45 .32 .23
6233 @item Function prototype @tab Example usage @tab Assembly output
6234 @item @code{uw1 __MABSHS (sw1)}
6235 @tab @code{@var{b} = __MABSHS (@var{a})}
6236 @tab @code{MABSHS @var{a},@var{b}}
6237 @item @code{void __MADDACCS (acc, acc)}
6238 @tab @code{__MADDACCS (@var{b}, @var{a})}
6239 @tab @code{MADDACCS @var{a},@var{b}}
6240 @item @code{sw1 __MADDHSS (sw1, sw1)}
6241 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
6242 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
6243 @item @code{uw1 __MADDHUS (uw1, uw1)}
6244 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
6245 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
6246 @item @code{uw1 __MAND (uw1, uw1)}
6247 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
6248 @tab @code{MAND @var{a},@var{b},@var{c}}
6249 @item @code{void __MASACCS (acc, acc)}
6250 @tab @code{__MASACCS (@var{b}, @var{a})}
6251 @tab @code{MASACCS @var{a},@var{b}}
6252 @item @code{uw1 __MAVEH (uw1, uw1)}
6253 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
6254 @tab @code{MAVEH @var{a},@var{b},@var{c}}
6255 @item @code{uw2 __MBTOH (uw1)}
6256 @tab @code{@var{b} = __MBTOH (@var{a})}
6257 @tab @code{MBTOH @var{a},@var{b}}
6258 @item @code{void __MBTOHE (uw1 *, uw1)}
6259 @tab @code{__MBTOHE (&@var{b}, @var{a})}
6260 @tab @code{MBTOHE @var{a},@var{b}}
6261 @item @code{void __MCLRACC (acc)}
6262 @tab @code{__MCLRACC (@var{a})}
6263 @tab @code{MCLRACC @var{a}}
6264 @item @code{void __MCLRACCA (void)}
6265 @tab @code{__MCLRACCA ()}
6266 @tab @code{MCLRACCA}
6267 @item @code{uw1 __Mcop1 (uw1, uw1)}
6268 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
6269 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
6270 @item @code{uw1 __Mcop2 (uw1, uw1)}
6271 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
6272 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
6273 @item @code{uw1 __MCPLHI (uw2, const)}
6274 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
6275 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
6276 @item @code{uw1 __MCPLI (uw2, const)}
6277 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
6278 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
6279 @item @code{void __MCPXIS (acc, sw1, sw1)}
6280 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
6281 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
6282 @item @code{void __MCPXIU (acc, uw1, uw1)}
6283 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
6284 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
6285 @item @code{void __MCPXRS (acc, sw1, sw1)}
6286 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
6287 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
6288 @item @code{void __MCPXRU (acc, uw1, uw1)}
6289 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
6290 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
6291 @item @code{uw1 __MCUT (acc, uw1)}
6292 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
6293 @tab @code{MCUT @var{a},@var{b},@var{c}}
6294 @item @code{uw1 __MCUTSS (acc, sw1)}
6295 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
6296 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
6297 @item @code{void __MDADDACCS (acc, acc)}
6298 @tab @code{__MDADDACCS (@var{b}, @var{a})}
6299 @tab @code{MDADDACCS @var{a},@var{b}}
6300 @item @code{void __MDASACCS (acc, acc)}
6301 @tab @code{__MDASACCS (@var{b}, @var{a})}
6302 @tab @code{MDASACCS @var{a},@var{b}}
6303 @item @code{uw2 __MDCUTSSI (acc, const)}
6304 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
6305 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
6306 @item @code{uw2 __MDPACKH (uw2, uw2)}
6307 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
6308 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
6309 @item @code{uw2 __MDROTLI (uw2, const)}
6310 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
6311 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
6312 @item @code{void __MDSUBACCS (acc, acc)}
6313 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
6314 @tab @code{MDSUBACCS @var{a},@var{b}}
6315 @item @code{void __MDUNPACKH (uw1 *, uw2)}
6316 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
6317 @tab @code{MDUNPACKH @var{a},@var{b}}
6318 @item @code{uw2 __MEXPDHD (uw1, const)}
6319 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
6320 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
6321 @item @code{uw1 __MEXPDHW (uw1, const)}
6322 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
6323 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
6324 @item @code{uw1 __MHDSETH (uw1, const)}
6325 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
6326 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
6327 @item @code{sw1 __MHDSETS (const)}
6328 @tab @code{@var{b} = __MHDSETS (@var{a})}
6329 @tab @code{MHDSETS #@var{a},@var{b}}
6330 @item @code{uw1 __MHSETHIH (uw1, const)}
6331 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
6332 @tab @code{MHSETHIH #@var{a},@var{b}}
6333 @item @code{sw1 __MHSETHIS (sw1, const)}
6334 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
6335 @tab @code{MHSETHIS #@var{a},@var{b}}
6336 @item @code{uw1 __MHSETLOH (uw1, const)}
6337 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
6338 @tab @code{MHSETLOH #@var{a},@var{b}}
6339 @item @code{sw1 __MHSETLOS (sw1, const)}
6340 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
6341 @tab @code{MHSETLOS #@var{a},@var{b}}
6342 @item @code{uw1 __MHTOB (uw2)}
6343 @tab @code{@var{b} = __MHTOB (@var{a})}
6344 @tab @code{MHTOB @var{a},@var{b}}
6345 @item @code{void __MMACHS (acc, sw1, sw1)}
6346 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
6347 @tab @code{MMACHS @var{a},@var{b},@var{c}}
6348 @item @code{void __MMACHU (acc, uw1, uw1)}
6349 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
6350 @tab @code{MMACHU @var{a},@var{b},@var{c}}
6351 @item @code{void __MMRDHS (acc, sw1, sw1)}
6352 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
6353 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
6354 @item @code{void __MMRDHU (acc, uw1, uw1)}
6355 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
6356 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
6357 @item @code{void __MMULHS (acc, sw1, sw1)}
6358 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
6359 @tab @code{MMULHS @var{a},@var{b},@var{c}}
6360 @item @code{void __MMULHU (acc, uw1, uw1)}
6361 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
6362 @tab @code{MMULHU @var{a},@var{b},@var{c}}
6363 @item @code{void __MMULXHS (acc, sw1, sw1)}
6364 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
6365 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
6366 @item @code{void __MMULXHU (acc, uw1, uw1)}
6367 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
6368 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
6369 @item @code{uw1 __MNOT (uw1)}
6370 @tab @code{@var{b} = __MNOT (@var{a})}
6371 @tab @code{MNOT @var{a},@var{b}}
6372 @item @code{uw1 __MOR (uw1, uw1)}
6373 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
6374 @tab @code{MOR @var{a},@var{b},@var{c}}
6375 @item @code{uw1 __MPACKH (uh, uh)}
6376 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
6377 @tab @code{MPACKH @var{a},@var{b},@var{c}}
6378 @item @code{sw2 __MQADDHSS (sw2, sw2)}
6379 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
6380 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
6381 @item @code{uw2 __MQADDHUS (uw2, uw2)}
6382 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
6383 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
6384 @item @code{void __MQCPXIS (acc, sw2, sw2)}
6385 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
6386 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
6387 @item @code{void __MQCPXIU (acc, uw2, uw2)}
6388 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
6389 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
6390 @item @code{void __MQCPXRS (acc, sw2, sw2)}
6391 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
6392 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
6393 @item @code{void __MQCPXRU (acc, uw2, uw2)}
6394 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
6395 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
6396 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
6397 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
6398 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
6399 @item @code{sw2 __MQLMTHS (sw2, sw2)}
6400 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
6401 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
6402 @item @code{void __MQMACHS (acc, sw2, sw2)}
6403 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
6404 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
6405 @item @code{void __MQMACHU (acc, uw2, uw2)}
6406 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
6407 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
6408 @item @code{void __MQMACXHS (acc, sw2, sw2)}
6409 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
6410 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
6411 @item @code{void __MQMULHS (acc, sw2, sw2)}
6412 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
6413 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
6414 @item @code{void __MQMULHU (acc, uw2, uw2)}
6415 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
6416 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
6417 @item @code{void __MQMULXHS (acc, sw2, sw2)}
6418 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
6419 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
6420 @item @code{void __MQMULXHU (acc, uw2, uw2)}
6421 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
6422 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
6423 @item @code{sw2 __MQSATHS (sw2, sw2)}
6424 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
6425 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
6426 @item @code{uw2 __MQSLLHI (uw2, int)}
6427 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
6428 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
6429 @item @code{sw2 __MQSRAHI (sw2, int)}
6430 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
6431 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
6432 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
6433 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
6434 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
6435 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
6436 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
6437 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
6438 @item @code{void __MQXMACHS (acc, sw2, sw2)}
6439 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
6440 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
6441 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
6442 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
6443 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
6444 @item @code{uw1 __MRDACC (acc)}
6445 @tab @code{@var{b} = __MRDACC (@var{a})}
6446 @tab @code{MRDACC @var{a},@var{b}}
6447 @item @code{uw1 __MRDACCG (acc)}
6448 @tab @code{@var{b} = __MRDACCG (@var{a})}
6449 @tab @code{MRDACCG @var{a},@var{b}}
6450 @item @code{uw1 __MROTLI (uw1, const)}
6451 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
6452 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
6453 @item @code{uw1 __MROTRI (uw1, const)}
6454 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
6455 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
6456 @item @code{sw1 __MSATHS (sw1, sw1)}
6457 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
6458 @tab @code{MSATHS @var{a},@var{b},@var{c}}
6459 @item @code{uw1 __MSATHU (uw1, uw1)}
6460 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
6461 @tab @code{MSATHU @var{a},@var{b},@var{c}}
6462 @item @code{uw1 __MSLLHI (uw1, const)}
6463 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
6464 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
6465 @item @code{sw1 __MSRAHI (sw1, const)}
6466 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
6467 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
6468 @item @code{uw1 __MSRLHI (uw1, const)}
6469 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
6470 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
6471 @item @code{void __MSUBACCS (acc, acc)}
6472 @tab @code{__MSUBACCS (@var{b}, @var{a})}
6473 @tab @code{MSUBACCS @var{a},@var{b}}
6474 @item @code{sw1 __MSUBHSS (sw1, sw1)}
6475 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
6476 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
6477 @item @code{uw1 __MSUBHUS (uw1, uw1)}
6478 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
6479 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
6480 @item @code{void __MTRAP (void)}
6481 @tab @code{__MTRAP ()}
6483 @item @code{uw2 __MUNPACKH (uw1)}
6484 @tab @code{@var{b} = __MUNPACKH (@var{a})}
6485 @tab @code{MUNPACKH @var{a},@var{b}}
6486 @item @code{uw1 __MWCUT (uw2, uw1)}
6487 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
6488 @tab @code{MWCUT @var{a},@var{b},@var{c}}
6489 @item @code{void __MWTACC (acc, uw1)}
6490 @tab @code{__MWTACC (@var{b}, @var{a})}
6491 @tab @code{MWTACC @var{a},@var{b}}
6492 @item @code{void __MWTACCG (acc, uw1)}
6493 @tab @code{__MWTACCG (@var{b}, @var{a})}
6494 @tab @code{MWTACCG @var{a},@var{b}}
6495 @item @code{uw1 __MXOR (uw1, uw1)}
6496 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
6497 @tab @code{MXOR @var{a},@var{b},@var{c}}
6500 @node Raw read/write Functions
6501 @subsubsection Raw read/write Functions
6503 This sections describes built-in functions related to read and write
6504 instructions to access memory. These functions generate
6505 @code{membar} instructions to flush the I/O load and stores where
6506 appropriate, as described in Fujitsu's manual described above.
6510 @item unsigned char __builtin_read8 (void *@var{data})
6511 @item unsigned short __builtin_read16 (void *@var{data})
6512 @item unsigned long __builtin_read32 (void *@var{data})
6513 @item unsigned long long __builtin_read64 (void *@var{data})
6515 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
6516 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
6517 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
6518 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
6521 @node Other Built-in Functions
6522 @subsubsection Other Built-in Functions
6524 This section describes built-in functions that are not named after
6525 a specific FR-V instruction.
6528 @item sw2 __IACCreadll (iacc @var{reg})
6529 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
6530 for future expansion and must be 0.
6532 @item sw1 __IACCreadl (iacc @var{reg})
6533 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
6534 Other values of @var{reg} are rejected as invalid.
6536 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
6537 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
6538 is reserved for future expansion and must be 0.
6540 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
6541 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
6542 is 1. Other values of @var{reg} are rejected as invalid.
6544 @item void __data_prefetch0 (const void *@var{x})
6545 Use the @code{dcpl} instruction to load the contents of address @var{x}
6546 into the data cache.
6548 @item void __data_prefetch (const void *@var{x})
6549 Use the @code{nldub} instruction to load the contents of address @var{x}
6550 into the data cache. The instruction will be issued in slot I1@.
6553 @node X86 Built-in Functions
6554 @subsection X86 Built-in Functions
6556 These built-in functions are available for the i386 and x86-64 family
6557 of computers, depending on the command-line switches used.
6559 Note that, if you specify command-line switches such as @option{-msse},
6560 the compiler could use the extended instruction sets even if the built-ins
6561 are not used explicitly in the program. For this reason, applications
6562 which perform runtime CPU detection must compile separate files for each
6563 supported architecture, using the appropriate flags. In particular,
6564 the file containing the CPU detection code should be compiled without
6567 The following machine modes are available for use with MMX built-in functions
6568 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
6569 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
6570 vector of eight 8-bit integers. Some of the built-in functions operate on
6571 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
6573 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
6574 of two 32-bit floating point values.
6576 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
6577 floating point values. Some instructions use a vector of four 32-bit
6578 integers, these use @code{V4SI}. Finally, some instructions operate on an
6579 entire vector register, interpreting it as a 128-bit integer, these use mode
6582 The following built-in functions are made available by @option{-mmmx}.
6583 All of them generate the machine instruction that is part of the name.
6586 v8qi __builtin_ia32_paddb (v8qi, v8qi)
6587 v4hi __builtin_ia32_paddw (v4hi, v4hi)
6588 v2si __builtin_ia32_paddd (v2si, v2si)
6589 v8qi __builtin_ia32_psubb (v8qi, v8qi)
6590 v4hi __builtin_ia32_psubw (v4hi, v4hi)
6591 v2si __builtin_ia32_psubd (v2si, v2si)
6592 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
6593 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
6594 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
6595 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
6596 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
6597 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
6598 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
6599 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
6600 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
6601 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
6602 di __builtin_ia32_pand (di, di)
6603 di __builtin_ia32_pandn (di,di)
6604 di __builtin_ia32_por (di, di)
6605 di __builtin_ia32_pxor (di, di)
6606 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
6607 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
6608 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
6609 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
6610 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
6611 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
6612 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
6613 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
6614 v2si __builtin_ia32_punpckhdq (v2si, v2si)
6615 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
6616 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
6617 v2si __builtin_ia32_punpckldq (v2si, v2si)
6618 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
6619 v4hi __builtin_ia32_packssdw (v2si, v2si)
6620 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
6623 The following built-in functions are made available either with
6624 @option{-msse}, or with a combination of @option{-m3dnow} and
6625 @option{-march=athlon}. All of them generate the machine
6626 instruction that is part of the name.
6629 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
6630 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
6631 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
6632 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
6633 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
6634 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
6635 v8qi __builtin_ia32_pminub (v8qi, v8qi)
6636 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
6637 int __builtin_ia32_pextrw (v4hi, int)
6638 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
6639 int __builtin_ia32_pmovmskb (v8qi)
6640 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
6641 void __builtin_ia32_movntq (di *, di)
6642 void __builtin_ia32_sfence (void)
6645 The following built-in functions are available when @option{-msse} is used.
6646 All of them generate the machine instruction that is part of the name.
6649 int __builtin_ia32_comieq (v4sf, v4sf)
6650 int __builtin_ia32_comineq (v4sf, v4sf)
6651 int __builtin_ia32_comilt (v4sf, v4sf)
6652 int __builtin_ia32_comile (v4sf, v4sf)
6653 int __builtin_ia32_comigt (v4sf, v4sf)
6654 int __builtin_ia32_comige (v4sf, v4sf)
6655 int __builtin_ia32_ucomieq (v4sf, v4sf)
6656 int __builtin_ia32_ucomineq (v4sf, v4sf)
6657 int __builtin_ia32_ucomilt (v4sf, v4sf)
6658 int __builtin_ia32_ucomile (v4sf, v4sf)
6659 int __builtin_ia32_ucomigt (v4sf, v4sf)
6660 int __builtin_ia32_ucomige (v4sf, v4sf)
6661 v4sf __builtin_ia32_addps (v4sf, v4sf)
6662 v4sf __builtin_ia32_subps (v4sf, v4sf)
6663 v4sf __builtin_ia32_mulps (v4sf, v4sf)
6664 v4sf __builtin_ia32_divps (v4sf, v4sf)
6665 v4sf __builtin_ia32_addss (v4sf, v4sf)
6666 v4sf __builtin_ia32_subss (v4sf, v4sf)
6667 v4sf __builtin_ia32_mulss (v4sf, v4sf)
6668 v4sf __builtin_ia32_divss (v4sf, v4sf)
6669 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
6670 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
6671 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
6672 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
6673 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
6674 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
6675 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
6676 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
6677 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
6678 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
6679 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
6680 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
6681 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
6682 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
6683 v4si __builtin_ia32_cmpless (v4sf, v4sf)
6684 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
6685 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
6686 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
6687 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
6688 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
6689 v4sf __builtin_ia32_maxps (v4sf, v4sf)
6690 v4sf __builtin_ia32_maxss (v4sf, v4sf)
6691 v4sf __builtin_ia32_minps (v4sf, v4sf)
6692 v4sf __builtin_ia32_minss (v4sf, v4sf)
6693 v4sf __builtin_ia32_andps (v4sf, v4sf)
6694 v4sf __builtin_ia32_andnps (v4sf, v4sf)
6695 v4sf __builtin_ia32_orps (v4sf, v4sf)
6696 v4sf __builtin_ia32_xorps (v4sf, v4sf)
6697 v4sf __builtin_ia32_movss (v4sf, v4sf)
6698 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
6699 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
6700 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
6701 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
6702 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
6703 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
6704 v2si __builtin_ia32_cvtps2pi (v4sf)
6705 int __builtin_ia32_cvtss2si (v4sf)
6706 v2si __builtin_ia32_cvttps2pi (v4sf)
6707 int __builtin_ia32_cvttss2si (v4sf)
6708 v4sf __builtin_ia32_rcpps (v4sf)
6709 v4sf __builtin_ia32_rsqrtps (v4sf)
6710 v4sf __builtin_ia32_sqrtps (v4sf)
6711 v4sf __builtin_ia32_rcpss (v4sf)
6712 v4sf __builtin_ia32_rsqrtss (v4sf)
6713 v4sf __builtin_ia32_sqrtss (v4sf)
6714 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
6715 void __builtin_ia32_movntps (float *, v4sf)
6716 int __builtin_ia32_movmskps (v4sf)
6719 The following built-in functions are available when @option{-msse} is used.
6722 @item v4sf __builtin_ia32_loadaps (float *)
6723 Generates the @code{movaps} machine instruction as a load from memory.
6724 @item void __builtin_ia32_storeaps (float *, v4sf)
6725 Generates the @code{movaps} machine instruction as a store to memory.
6726 @item v4sf __builtin_ia32_loadups (float *)
6727 Generates the @code{movups} machine instruction as a load from memory.
6728 @item void __builtin_ia32_storeups (float *, v4sf)
6729 Generates the @code{movups} machine instruction as a store to memory.
6730 @item v4sf __builtin_ia32_loadsss (float *)
6731 Generates the @code{movss} machine instruction as a load from memory.
6732 @item void __builtin_ia32_storess (float *, v4sf)
6733 Generates the @code{movss} machine instruction as a store to memory.
6734 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
6735 Generates the @code{movhps} machine instruction as a load from memory.
6736 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
6737 Generates the @code{movlps} machine instruction as a load from memory
6738 @item void __builtin_ia32_storehps (v4sf, v2si *)
6739 Generates the @code{movhps} machine instruction as a store to memory.
6740 @item void __builtin_ia32_storelps (v4sf, v2si *)
6741 Generates the @code{movlps} machine instruction as a store to memory.
6744 The following built-in functions are available when @option{-msse3} is used.
6745 All of them generate the machine instruction that is part of the name.
6748 v2df __builtin_ia32_addsubpd (v2df, v2df)
6749 v2df __builtin_ia32_addsubps (v2df, v2df)
6750 v2df __builtin_ia32_haddpd (v2df, v2df)
6751 v2df __builtin_ia32_haddps (v2df, v2df)
6752 v2df __builtin_ia32_hsubpd (v2df, v2df)
6753 v2df __builtin_ia32_hsubps (v2df, v2df)
6754 v16qi __builtin_ia32_lddqu (char const *)
6755 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
6756 v2df __builtin_ia32_movddup (v2df)
6757 v4sf __builtin_ia32_movshdup (v4sf)
6758 v4sf __builtin_ia32_movsldup (v4sf)
6759 void __builtin_ia32_mwait (unsigned int, unsigned int)
6762 The following built-in functions are available when @option{-msse3} is used.
6765 @item v2df __builtin_ia32_loadddup (double const *)
6766 Generates the @code{movddup} machine instruction as a load from memory.
6769 The following built-in functions are available when @option{-m3dnow} is used.
6770 All of them generate the machine instruction that is part of the name.
6773 void __builtin_ia32_femms (void)
6774 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
6775 v2si __builtin_ia32_pf2id (v2sf)
6776 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
6777 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
6778 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
6779 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
6780 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
6781 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
6782 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
6783 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
6784 v2sf __builtin_ia32_pfrcp (v2sf)
6785 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
6786 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
6787 v2sf __builtin_ia32_pfrsqrt (v2sf)
6788 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
6789 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
6790 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
6791 v2sf __builtin_ia32_pi2fd (v2si)
6792 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
6795 The following built-in functions are available when both @option{-m3dnow}
6796 and @option{-march=athlon} are used. All of them generate the machine
6797 instruction that is part of the name.
6800 v2si __builtin_ia32_pf2iw (v2sf)
6801 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
6802 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
6803 v2sf __builtin_ia32_pi2fw (v2si)
6804 v2sf __builtin_ia32_pswapdsf (v2sf)
6805 v2si __builtin_ia32_pswapdsi (v2si)
6808 The following built-in functions are available when @option{-msse2}
6809 is used. All of them generate calls to an SSE2 ABI IEEE754 math intrinsic
6810 that is part of the name. Rather than using these directly you may
6811 want them automatically substituted for calls to the regular intrinsics
6812 using the @option{-msselibm}.
6815 double __builtin_sse2_acos (double)
6816 float __builtin_sse2_acosf (float)
6817 double __builtin_sse2_asin (double)
6818 float __builtin_sse2_asinf (float)
6819 double __builtin_sse2_atan (double)
6820 float __builtin_sse2_atanf (float)
6821 double __builtin_sse2_atan2 (double, double)
6822 float __builtin_sse2_atan2f (float, float)
6823 double __builtin_sse2_cos (double)
6824 float __builtin_sse2_cosf (float)
6825 double __builtin_sse2_exp (double)
6826 float __builtin_sse2_expf (float)
6827 double __builtin_sse2_log10 (double)
6828 float __builtin_sse2_log10f (float)
6829 double __builtin_sse2_log (double)
6830 float __builtin_sse2_logf (float)
6831 double __builtin_sse2_sin (double)
6832 float __builtin_sse2_sinf (float)
6833 double __builtin_sse2_tan (double)
6834 float __builtin_sse2_tanf (float)
6837 @node MIPS DSP Built-in Functions
6838 @subsection MIPS DSP Built-in Functions
6840 The MIPS DSP Application-Specific Extension (ASE) includes new
6841 instructions that are designed to improve the performance of DSP and
6842 media applications. It provides instructions that operate on packed
6843 8-bit integer data, Q15 fractional data and Q31 fractional data.
6845 GCC supports MIPS DSP operations using both the generic
6846 vector extensions (@pxref{Vector Extensions}) and a collection of
6847 MIPS-specific built-in functions. Both kinds of support are
6848 enabled by the @option{-mdsp} command-line option.
6850 At present, GCC only provides support for operations on 32-bit
6851 vectors. The vector type associated with 8-bit integer data is
6852 usually called @code{v4i8} and the vector type associated with Q15 is
6853 usually called @code{v2q15}. They can be defined in C as follows:
6856 typedef char v4i8 __attribute__ ((vector_size(4)));
6857 typedef short v2q15 __attribute__ ((vector_size(4)));
6860 @code{v4i8} and @code{v2q15} values are initialized in the same way as
6861 aggregates. For example:
6864 v4i8 a = @{1, 2, 3, 4@};
6866 b = (v4i8) @{5, 6, 7, 8@};
6868 v2q15 c = @{0x0fcb, 0x3a75@};
6870 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
6873 @emph{Note:} The CPU's endianness determines the order in which values
6874 are packed. On little-endian targets, the first value is the least
6875 significant and the last value is the most significant. The opposite
6876 order applies to big-endian targets. For example, the code above will
6877 set the lowest byte of @code{a} to @code{1} on little-endian targets
6878 and @code{4} on big-endian targets.
6880 @emph{Note:} Q15 and Q31 values must be initialized with their integer
6881 representation. As shown in this example, the integer representation
6882 of a Q15 value can be obtained by multiplying the fractional value by
6883 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
6886 The table below lists the @code{v4i8} and @code{v2q15} operations for which
6887 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
6888 and @code{c} and @code{d} are @code{v2q15} values.
6890 @multitable @columnfractions .50 .50
6891 @item C code @tab MIPS instruction
6892 @item @code{a + b} @tab @code{addu.qb}
6893 @item @code{c + d} @tab @code{addq.ph}
6894 @item @code{a - b} @tab @code{subu.qb}
6895 @item @code{c - d} @tab @code{subq.ph}
6898 It is easier to describe the DSP built-in functions if we first define
6899 the following types:
6904 typedef long long a64;
6907 @code{q31} and @code{i32} are actually the same as @code{int}, but we
6908 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
6909 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
6910 @code{long long}, but we use @code{a64} to indicate values that will
6911 be placed in one of the four DSP accumulators (@code{$ac0},
6912 @code{$ac1}, @code{$ac2} or @code{$ac3}).
6914 Also, some built-in functions prefer or require immediate numbers as
6915 parameters, because the corresponding DSP instructions accept both immediate
6916 numbers and register operands, or accept immediate numbers only. The
6917 immediate parameters are listed as follows.
6925 imm_n32_31: -32 to 31.
6926 imm_n512_511: -512 to 511.
6929 The following built-in functions map directly to a particular MIPS DSP
6930 instruction. Please refer to the architecture specification
6931 for details on what each instruction does.
6934 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
6935 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
6936 q31 __builtin_mips_addq_s_w (q31, q31)
6937 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
6938 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
6939 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
6940 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
6941 q31 __builtin_mips_subq_s_w (q31, q31)
6942 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
6943 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
6944 i32 __builtin_mips_addsc (i32, i32)
6945 i32 __builtin_mips_addwc (i32, i32)
6946 i32 __builtin_mips_modsub (i32, i32)
6947 i32 __builtin_mips_raddu_w_qb (v4i8)
6948 v2q15 __builtin_mips_absq_s_ph (v2q15)
6949 q31 __builtin_mips_absq_s_w (q31)
6950 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
6951 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
6952 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
6953 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
6954 q31 __builtin_mips_preceq_w_phl (v2q15)
6955 q31 __builtin_mips_preceq_w_phr (v2q15)
6956 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
6957 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
6958 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
6959 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
6960 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
6961 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
6962 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
6963 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
6964 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
6965 v4i8 __builtin_mips_shll_qb (v4i8, i32)
6966 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
6967 v2q15 __builtin_mips_shll_ph (v2q15, i32)
6968 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
6969 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
6970 q31 __builtin_mips_shll_s_w (q31, imm0_31)
6971 q31 __builtin_mips_shll_s_w (q31, i32)
6972 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
6973 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
6974 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
6975 v2q15 __builtin_mips_shra_ph (v2q15, i32)
6976 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
6977 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
6978 q31 __builtin_mips_shra_r_w (q31, imm0_31)
6979 q31 __builtin_mips_shra_r_w (q31, i32)
6980 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
6981 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
6982 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
6983 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
6984 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
6985 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
6986 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
6987 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
6988 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
6989 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
6990 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
6991 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
6992 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
6993 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
6994 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
6995 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
6996 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
6997 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
6998 i32 __builtin_mips_bitrev (i32)
6999 i32 __builtin_mips_insv (i32, i32)
7000 v4i8 __builtin_mips_repl_qb (imm0_255)
7001 v4i8 __builtin_mips_repl_qb (i32)
7002 v2q15 __builtin_mips_repl_ph (imm_n512_511)
7003 v2q15 __builtin_mips_repl_ph (i32)
7004 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
7005 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
7006 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
7007 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
7008 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
7009 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
7010 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
7011 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
7012 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
7013 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
7014 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
7015 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
7016 i32 __builtin_mips_extr_w (a64, imm0_31)
7017 i32 __builtin_mips_extr_w (a64, i32)
7018 i32 __builtin_mips_extr_r_w (a64, imm0_31)
7019 i32 __builtin_mips_extr_s_h (a64, i32)
7020 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
7021 i32 __builtin_mips_extr_rs_w (a64, i32)
7022 i32 __builtin_mips_extr_s_h (a64, imm0_31)
7023 i32 __builtin_mips_extr_r_w (a64, i32)
7024 i32 __builtin_mips_extp (a64, imm0_31)
7025 i32 __builtin_mips_extp (a64, i32)
7026 i32 __builtin_mips_extpdp (a64, imm0_31)
7027 i32 __builtin_mips_extpdp (a64, i32)
7028 a64 __builtin_mips_shilo (a64, imm_n32_31)
7029 a64 __builtin_mips_shilo (a64, i32)
7030 a64 __builtin_mips_mthlip (a64, i32)
7031 void __builtin_mips_wrdsp (i32, imm0_63)
7032 i32 __builtin_mips_rddsp (imm0_63)
7033 i32 __builtin_mips_lbux (void *, i32)
7034 i32 __builtin_mips_lhx (void *, i32)
7035 i32 __builtin_mips_lwx (void *, i32)
7036 i32 __builtin_mips_bposge32 (void)
7039 @node MIPS Paired-Single Support
7040 @subsection MIPS Paired-Single Support
7042 The MIPS64 architecture includes a number of instructions that
7043 operate on pairs of single-precision floating-point values.
7044 Each pair is packed into a 64-bit floating-point register,
7045 with one element being designated the ``upper half'' and
7046 the other being designated the ``lower half''.
7048 GCC supports paired-single operations using both the generic
7049 vector extensions (@pxref{Vector Extensions}) and a collection of
7050 MIPS-specific built-in functions. Both kinds of support are
7051 enabled by the @option{-mpaired-single} command-line option.
7053 The vector type associated with paired-single values is usually
7054 called @code{v2sf}. It can be defined in C as follows:
7057 typedef float v2sf __attribute__ ((vector_size (8)));
7060 @code{v2sf} values are initialized in the same way as aggregates.
7064 v2sf a = @{1.5, 9.1@};
7067 b = (v2sf) @{e, f@};
7070 @emph{Note:} The CPU's endianness determines which value is stored in
7071 the upper half of a register and which value is stored in the lower half.
7072 On little-endian targets, the first value is the lower one and the second
7073 value is the upper one. The opposite order applies to big-endian targets.
7074 For example, the code above will set the lower half of @code{a} to
7075 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
7078 * Paired-Single Arithmetic::
7079 * Paired-Single Built-in Functions::
7080 * MIPS-3D Built-in Functions::
7083 @node Paired-Single Arithmetic
7084 @subsubsection Paired-Single Arithmetic
7086 The table below lists the @code{v2sf} operations for which hardware
7087 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
7088 values and @code{x} is an integral value.
7090 @multitable @columnfractions .50 .50
7091 @item C code @tab MIPS instruction
7092 @item @code{a + b} @tab @code{add.ps}
7093 @item @code{a - b} @tab @code{sub.ps}
7094 @item @code{-a} @tab @code{neg.ps}
7095 @item @code{a * b} @tab @code{mul.ps}
7096 @item @code{a * b + c} @tab @code{madd.ps}
7097 @item @code{a * b - c} @tab @code{msub.ps}
7098 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
7099 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
7100 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
7103 Note that the multiply-accumulate instructions can be disabled
7104 using the command-line option @code{-mno-fused-madd}.
7106 @node Paired-Single Built-in Functions
7107 @subsubsection Paired-Single Built-in Functions
7109 The following paired-single functions map directly to a particular
7110 MIPS instruction. Please refer to the architecture specification
7111 for details on what each instruction does.
7114 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
7115 Pair lower lower (@code{pll.ps}).
7117 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
7118 Pair upper lower (@code{pul.ps}).
7120 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
7121 Pair lower upper (@code{plu.ps}).
7123 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
7124 Pair upper upper (@code{puu.ps}).
7126 @item v2sf __builtin_mips_cvt_ps_s (float, float)
7127 Convert pair to paired single (@code{cvt.ps.s}).
7129 @item float __builtin_mips_cvt_s_pl (v2sf)
7130 Convert pair lower to single (@code{cvt.s.pl}).
7132 @item float __builtin_mips_cvt_s_pu (v2sf)
7133 Convert pair upper to single (@code{cvt.s.pu}).
7135 @item v2sf __builtin_mips_abs_ps (v2sf)
7136 Absolute value (@code{abs.ps}).
7138 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
7139 Align variable (@code{alnv.ps}).
7141 @emph{Note:} The value of the third parameter must be 0 or 4
7142 modulo 8, otherwise the result will be unpredictable. Please read the
7143 instruction description for details.
7146 The following multi-instruction functions are also available.
7147 In each case, @var{cond} can be any of the 16 floating-point conditions:
7148 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
7149 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
7150 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
7153 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7154 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7155 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
7156 @code{movt.ps}/@code{movf.ps}).
7158 The @code{movt} functions return the value @var{x} computed by:
7161 c.@var{cond}.ps @var{cc},@var{a},@var{b}
7162 mov.ps @var{x},@var{c}
7163 movt.ps @var{x},@var{d},@var{cc}
7166 The @code{movf} functions are similar but use @code{movf.ps} instead
7169 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7170 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7171 Comparison of two paired-single values (@code{c.@var{cond}.ps},
7172 @code{bc1t}/@code{bc1f}).
7174 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
7175 and return either the upper or lower half of the result. For example:
7179 if (__builtin_mips_upper_c_eq_ps (a, b))
7180 upper_halves_are_equal ();
7182 upper_halves_are_unequal ();
7184 if (__builtin_mips_lower_c_eq_ps (a, b))
7185 lower_halves_are_equal ();
7187 lower_halves_are_unequal ();
7191 @node MIPS-3D Built-in Functions
7192 @subsubsection MIPS-3D Built-in Functions
7194 The MIPS-3D Application-Specific Extension (ASE) includes additional
7195 paired-single instructions that are designed to improve the performance
7196 of 3D graphics operations. Support for these instructions is controlled
7197 by the @option{-mips3d} command-line option.
7199 The functions listed below map directly to a particular MIPS-3D
7200 instruction. Please refer to the architecture specification for
7201 more details on what each instruction does.
7204 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
7205 Reduction add (@code{addr.ps}).
7207 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
7208 Reduction multiply (@code{mulr.ps}).
7210 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
7211 Convert paired single to paired word (@code{cvt.pw.ps}).
7213 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
7214 Convert paired word to paired single (@code{cvt.ps.pw}).
7216 @item float __builtin_mips_recip1_s (float)
7217 @itemx double __builtin_mips_recip1_d (double)
7218 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
7219 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
7221 @item float __builtin_mips_recip2_s (float, float)
7222 @itemx double __builtin_mips_recip2_d (double, double)
7223 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
7224 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
7226 @item float __builtin_mips_rsqrt1_s (float)
7227 @itemx double __builtin_mips_rsqrt1_d (double)
7228 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
7229 Reduced precision reciprocal square root (sequence step 1)
7230 (@code{rsqrt1.@var{fmt}}).
7232 @item float __builtin_mips_rsqrt2_s (float, float)
7233 @itemx double __builtin_mips_rsqrt2_d (double, double)
7234 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
7235 Reduced precision reciprocal square root (sequence step 2)
7236 (@code{rsqrt2.@var{fmt}}).
7239 The following multi-instruction functions are also available.
7240 In each case, @var{cond} can be any of the 16 floating-point conditions:
7241 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
7242 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
7243 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
7246 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
7247 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
7248 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
7249 @code{bc1t}/@code{bc1f}).
7251 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
7252 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
7257 if (__builtin_mips_cabs_eq_s (a, b))
7263 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7264 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7265 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
7266 @code{bc1t}/@code{bc1f}).
7268 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
7269 and return either the upper or lower half of the result. For example:
7273 if (__builtin_mips_upper_cabs_eq_ps (a, b))
7274 upper_halves_are_equal ();
7276 upper_halves_are_unequal ();
7278 if (__builtin_mips_lower_cabs_eq_ps (a, b))
7279 lower_halves_are_equal ();
7281 lower_halves_are_unequal ();
7284 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7285 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7286 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
7287 @code{movt.ps}/@code{movf.ps}).
7289 The @code{movt} functions return the value @var{x} computed by:
7292 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
7293 mov.ps @var{x},@var{c}
7294 movt.ps @var{x},@var{d},@var{cc}
7297 The @code{movf} functions are similar but use @code{movf.ps} instead
7300 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7301 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7302 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7303 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7304 Comparison of two paired-single values
7305 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
7306 @code{bc1any2t}/@code{bc1any2f}).
7308 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
7309 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
7310 result is true and the @code{all} forms return true if both results are true.
7315 if (__builtin_mips_any_c_eq_ps (a, b))
7320 if (__builtin_mips_all_c_eq_ps (a, b))
7326 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7327 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7328 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7329 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7330 Comparison of four paired-single values
7331 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
7332 @code{bc1any4t}/@code{bc1any4f}).
7334 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
7335 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
7336 The @code{any} forms return true if any of the four results are true
7337 and the @code{all} forms return true if all four results are true.
7342 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
7347 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
7354 @node PowerPC AltiVec Built-in Functions
7355 @subsection PowerPC AltiVec Built-in Functions
7357 GCC provides an interface for the PowerPC family of processors to access
7358 the AltiVec operations described in Motorola's AltiVec Programming
7359 Interface Manual. The interface is made available by including
7360 @code{<altivec.h>} and using @option{-maltivec} and
7361 @option{-mabi=altivec}. The interface supports the following vector
7365 vector unsigned char
7369 vector unsigned short
7380 GCC's implementation of the high-level language interface available from
7381 C and C++ code differs from Motorola's documentation in several ways.
7386 A vector constant is a list of constant expressions within curly braces.
7389 A vector initializer requires no cast if the vector constant is of the
7390 same type as the variable it is initializing.
7393 If @code{signed} or @code{unsigned} is omitted, the signedness of the
7394 vector type is the default signedness of the base type. The default
7395 varies depending on the operating system, so a portable program should
7396 always specify the signedness.
7399 Compiling with @option{-maltivec} adds keywords @code{__vector},
7400 @code{__pixel}, and @code{__bool}. Macros @option{vector},
7401 @code{pixel}, and @code{bool} are defined in @code{<altivec.h>} and can
7405 GCC allows using a @code{typedef} name as the type specifier for a
7409 For C, overloaded functions are implemented with macros so the following
7413 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
7416 Since @code{vec_add} is a macro, the vector constant in the example
7417 is treated as four separate arguments. Wrap the entire argument in
7418 parentheses for this to work.
7421 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
7422 Internally, GCC uses built-in functions to achieve the functionality in
7423 the aforementioned header file, but they are not supported and are
7424 subject to change without notice.
7426 The following interfaces are supported for the generic and specific
7427 AltiVec operations and the AltiVec predicates. In cases where there
7428 is a direct mapping between generic and specific operations, only the
7429 generic names are shown here, although the specific operations can also
7432 Arguments that are documented as @code{const int} require literal
7433 integral values within the range required for that operation.
7436 vector signed char vec_abs (vector signed char);
7437 vector signed short vec_abs (vector signed short);
7438 vector signed int vec_abs (vector signed int);
7439 vector float vec_abs (vector float);
7441 vector signed char vec_abss (vector signed char);
7442 vector signed short vec_abss (vector signed short);
7443 vector signed int vec_abss (vector signed int);
7445 vector signed char vec_add (vector bool char, vector signed char);
7446 vector signed char vec_add (vector signed char, vector bool char);
7447 vector signed char vec_add (vector signed char, vector signed char);
7448 vector unsigned char vec_add (vector bool char, vector unsigned char);
7449 vector unsigned char vec_add (vector unsigned char, vector bool char);
7450 vector unsigned char vec_add (vector unsigned char,
7451 vector unsigned char);
7452 vector signed short vec_add (vector bool short, vector signed short);
7453 vector signed short vec_add (vector signed short, vector bool short);
7454 vector signed short vec_add (vector signed short, vector signed short);
7455 vector unsigned short vec_add (vector bool short,
7456 vector unsigned short);
7457 vector unsigned short vec_add (vector unsigned short,
7459 vector unsigned short vec_add (vector unsigned short,
7460 vector unsigned short);
7461 vector signed int vec_add (vector bool int, vector signed int);
7462 vector signed int vec_add (vector signed int, vector bool int);
7463 vector signed int vec_add (vector signed int, vector signed int);
7464 vector unsigned int vec_add (vector bool int, vector unsigned int);
7465 vector unsigned int vec_add (vector unsigned int, vector bool int);
7466 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
7467 vector float vec_add (vector float, vector float);
7469 vector float vec_vaddfp (vector float, vector float);
7471 vector signed int vec_vadduwm (vector bool int, vector signed int);
7472 vector signed int vec_vadduwm (vector signed int, vector bool int);
7473 vector signed int vec_vadduwm (vector signed int, vector signed int);
7474 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
7475 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
7476 vector unsigned int vec_vadduwm (vector unsigned int,
7477 vector unsigned int);
7479 vector signed short vec_vadduhm (vector bool short,
7480 vector signed short);
7481 vector signed short vec_vadduhm (vector signed short,
7483 vector signed short vec_vadduhm (vector signed short,
7484 vector signed short);
7485 vector unsigned short vec_vadduhm (vector bool short,
7486 vector unsigned short);
7487 vector unsigned short vec_vadduhm (vector unsigned short,
7489 vector unsigned short vec_vadduhm (vector unsigned short,
7490 vector unsigned short);
7492 vector signed char vec_vaddubm (vector bool char, vector signed char);
7493 vector signed char vec_vaddubm (vector signed char, vector bool char);
7494 vector signed char vec_vaddubm (vector signed char, vector signed char);
7495 vector unsigned char vec_vaddubm (vector bool char,
7496 vector unsigned char);
7497 vector unsigned char vec_vaddubm (vector unsigned char,
7499 vector unsigned char vec_vaddubm (vector unsigned char,
7500 vector unsigned char);
7502 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
7504 vector unsigned char vec_adds (vector bool char, vector unsigned char);
7505 vector unsigned char vec_adds (vector unsigned char, vector bool char);
7506 vector unsigned char vec_adds (vector unsigned char,
7507 vector unsigned char);
7508 vector signed char vec_adds (vector bool char, vector signed char);
7509 vector signed char vec_adds (vector signed char, vector bool char);
7510 vector signed char vec_adds (vector signed char, vector signed char);
7511 vector unsigned short vec_adds (vector bool short,
7512 vector unsigned short);
7513 vector unsigned short vec_adds (vector unsigned short,
7515 vector unsigned short vec_adds (vector unsigned short,
7516 vector unsigned short);
7517 vector signed short vec_adds (vector bool short, vector signed short);
7518 vector signed short vec_adds (vector signed short, vector bool short);
7519 vector signed short vec_adds (vector signed short, vector signed short);
7520 vector unsigned int vec_adds (vector bool int, vector unsigned int);
7521 vector unsigned int vec_adds (vector unsigned int, vector bool int);
7522 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
7523 vector signed int vec_adds (vector bool int, vector signed int);
7524 vector signed int vec_adds (vector signed int, vector bool int);
7525 vector signed int vec_adds (vector signed int, vector signed int);
7527 vector signed int vec_vaddsws (vector bool int, vector signed int);
7528 vector signed int vec_vaddsws (vector signed int, vector bool int);
7529 vector signed int vec_vaddsws (vector signed int, vector signed int);
7531 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
7532 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
7533 vector unsigned int vec_vadduws (vector unsigned int,
7534 vector unsigned int);
7536 vector signed short vec_vaddshs (vector bool short,
7537 vector signed short);
7538 vector signed short vec_vaddshs (vector signed short,
7540 vector signed short vec_vaddshs (vector signed short,
7541 vector signed short);
7543 vector unsigned short vec_vadduhs (vector bool short,
7544 vector unsigned short);
7545 vector unsigned short vec_vadduhs (vector unsigned short,
7547 vector unsigned short vec_vadduhs (vector unsigned short,
7548 vector unsigned short);
7550 vector signed char vec_vaddsbs (vector bool char, vector signed char);
7551 vector signed char vec_vaddsbs (vector signed char, vector bool char);
7552 vector signed char vec_vaddsbs (vector signed char, vector signed char);
7554 vector unsigned char vec_vaddubs (vector bool char,
7555 vector unsigned char);
7556 vector unsigned char vec_vaddubs (vector unsigned char,
7558 vector unsigned char vec_vaddubs (vector unsigned char,
7559 vector unsigned char);
7561 vector float vec_and (vector float, vector float);
7562 vector float vec_and (vector float, vector bool int);
7563 vector float vec_and (vector bool int, vector float);
7564 vector bool int vec_and (vector bool int, vector bool int);
7565 vector signed int vec_and (vector bool int, vector signed int);
7566 vector signed int vec_and (vector signed int, vector bool int);
7567 vector signed int vec_and (vector signed int, vector signed int);
7568 vector unsigned int vec_and (vector bool int, vector unsigned int);
7569 vector unsigned int vec_and (vector unsigned int, vector bool int);
7570 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
7571 vector bool short vec_and (vector bool short, vector bool short);
7572 vector signed short vec_and (vector bool short, vector signed short);
7573 vector signed short vec_and (vector signed short, vector bool short);
7574 vector signed short vec_and (vector signed short, vector signed short);
7575 vector unsigned short vec_and (vector bool short,
7576 vector unsigned short);
7577 vector unsigned short vec_and (vector unsigned short,
7579 vector unsigned short vec_and (vector unsigned short,
7580 vector unsigned short);
7581 vector signed char vec_and (vector bool char, vector signed char);
7582 vector bool char vec_and (vector bool char, vector bool char);
7583 vector signed char vec_and (vector signed char, vector bool char);
7584 vector signed char vec_and (vector signed char, vector signed char);
7585 vector unsigned char vec_and (vector bool char, vector unsigned char);
7586 vector unsigned char vec_and (vector unsigned char, vector bool char);
7587 vector unsigned char vec_and (vector unsigned char,
7588 vector unsigned char);
7590 vector float vec_andc (vector float, vector float);
7591 vector float vec_andc (vector float, vector bool int);
7592 vector float vec_andc (vector bool int, vector float);
7593 vector bool int vec_andc (vector bool int, vector bool int);
7594 vector signed int vec_andc (vector bool int, vector signed int);
7595 vector signed int vec_andc (vector signed int, vector bool int);
7596 vector signed int vec_andc (vector signed int, vector signed int);
7597 vector unsigned int vec_andc (vector bool int, vector unsigned int);
7598 vector unsigned int vec_andc (vector unsigned int, vector bool int);
7599 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
7600 vector bool short vec_andc (vector bool short, vector bool short);
7601 vector signed short vec_andc (vector bool short, vector signed short);
7602 vector signed short vec_andc (vector signed short, vector bool short);
7603 vector signed short vec_andc (vector signed short, vector signed short);
7604 vector unsigned short vec_andc (vector bool short,
7605 vector unsigned short);
7606 vector unsigned short vec_andc (vector unsigned short,
7608 vector unsigned short vec_andc (vector unsigned short,
7609 vector unsigned short);
7610 vector signed char vec_andc (vector bool char, vector signed char);
7611 vector bool char vec_andc (vector bool char, vector bool char);
7612 vector signed char vec_andc (vector signed char, vector bool char);
7613 vector signed char vec_andc (vector signed char, vector signed char);
7614 vector unsigned char vec_andc (vector bool char, vector unsigned char);
7615 vector unsigned char vec_andc (vector unsigned char, vector bool char);
7616 vector unsigned char vec_andc (vector unsigned char,
7617 vector unsigned char);
7619 vector unsigned char vec_avg (vector unsigned char,
7620 vector unsigned char);
7621 vector signed char vec_avg (vector signed char, vector signed char);
7622 vector unsigned short vec_avg (vector unsigned short,
7623 vector unsigned short);
7624 vector signed short vec_avg (vector signed short, vector signed short);
7625 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
7626 vector signed int vec_avg (vector signed int, vector signed int);
7628 vector signed int vec_vavgsw (vector signed int, vector signed int);
7630 vector unsigned int vec_vavguw (vector unsigned int,
7631 vector unsigned int);
7633 vector signed short vec_vavgsh (vector signed short,
7634 vector signed short);
7636 vector unsigned short vec_vavguh (vector unsigned short,
7637 vector unsigned short);
7639 vector signed char vec_vavgsb (vector signed char, vector signed char);
7641 vector unsigned char vec_vavgub (vector unsigned char,
7642 vector unsigned char);
7644 vector float vec_ceil (vector float);
7646 vector signed int vec_cmpb (vector float, vector float);
7648 vector bool char vec_cmpeq (vector signed char, vector signed char);
7649 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
7650 vector bool short vec_cmpeq (vector signed short, vector signed short);
7651 vector bool short vec_cmpeq (vector unsigned short,
7652 vector unsigned short);
7653 vector bool int vec_cmpeq (vector signed int, vector signed int);
7654 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
7655 vector bool int vec_cmpeq (vector float, vector float);
7657 vector bool int vec_vcmpeqfp (vector float, vector float);
7659 vector bool int vec_vcmpequw (vector signed int, vector signed int);
7660 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
7662 vector bool short vec_vcmpequh (vector signed short,
7663 vector signed short);
7664 vector bool short vec_vcmpequh (vector unsigned short,
7665 vector unsigned short);
7667 vector bool char vec_vcmpequb (vector signed char, vector signed char);
7668 vector bool char vec_vcmpequb (vector unsigned char,
7669 vector unsigned char);
7671 vector bool int vec_cmpge (vector float, vector float);
7673 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
7674 vector bool char vec_cmpgt (vector signed char, vector signed char);
7675 vector bool short vec_cmpgt (vector unsigned short,
7676 vector unsigned short);
7677 vector bool short vec_cmpgt (vector signed short, vector signed short);
7678 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
7679 vector bool int vec_cmpgt (vector signed int, vector signed int);
7680 vector bool int vec_cmpgt (vector float, vector float);
7682 vector bool int vec_vcmpgtfp (vector float, vector float);
7684 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
7686 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
7688 vector bool short vec_vcmpgtsh (vector signed short,
7689 vector signed short);
7691 vector bool short vec_vcmpgtuh (vector unsigned short,
7692 vector unsigned short);
7694 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
7696 vector bool char vec_vcmpgtub (vector unsigned char,
7697 vector unsigned char);
7699 vector bool int vec_cmple (vector float, vector float);
7701 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
7702 vector bool char vec_cmplt (vector signed char, vector signed char);
7703 vector bool short vec_cmplt (vector unsigned short,
7704 vector unsigned short);
7705 vector bool short vec_cmplt (vector signed short, vector signed short);
7706 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
7707 vector bool int vec_cmplt (vector signed int, vector signed int);
7708 vector bool int vec_cmplt (vector float, vector float);
7710 vector float vec_ctf (vector unsigned int, const int);
7711 vector float vec_ctf (vector signed int, const int);
7713 vector float vec_vcfsx (vector signed int, const int);
7715 vector float vec_vcfux (vector unsigned int, const int);
7717 vector signed int vec_cts (vector float, const int);
7719 vector unsigned int vec_ctu (vector float, const int);
7721 void vec_dss (const int);
7723 void vec_dssall (void);
7725 void vec_dst (const vector unsigned char *, int, const int);
7726 void vec_dst (const vector signed char *, int, const int);
7727 void vec_dst (const vector bool char *, int, const int);
7728 void vec_dst (const vector unsigned short *, int, const int);
7729 void vec_dst (const vector signed short *, int, const int);
7730 void vec_dst (const vector bool short *, int, const int);
7731 void vec_dst (const vector pixel *, int, const int);
7732 void vec_dst (const vector unsigned int *, int, const int);
7733 void vec_dst (const vector signed int *, int, const int);
7734 void vec_dst (const vector bool int *, int, const int);
7735 void vec_dst (const vector float *, int, const int);
7736 void vec_dst (const unsigned char *, int, const int);
7737 void vec_dst (const signed char *, int, const int);
7738 void vec_dst (const unsigned short *, int, const int);
7739 void vec_dst (const short *, int, const int);
7740 void vec_dst (const unsigned int *, int, const int);
7741 void vec_dst (const int *, int, const int);
7742 void vec_dst (const unsigned long *, int, const int);
7743 void vec_dst (const long *, int, const int);
7744 void vec_dst (const float *, int, const int);
7746 void vec_dstst (const vector unsigned char *, int, const int);
7747 void vec_dstst (const vector signed char *, int, const int);
7748 void vec_dstst (const vector bool char *, int, const int);
7749 void vec_dstst (const vector unsigned short *, int, const int);
7750 void vec_dstst (const vector signed short *, int, const int);
7751 void vec_dstst (const vector bool short *, int, const int);
7752 void vec_dstst (const vector pixel *, int, const int);
7753 void vec_dstst (const vector unsigned int *, int, const int);
7754 void vec_dstst (const vector signed int *, int, const int);
7755 void vec_dstst (const vector bool int *, int, const int);
7756 void vec_dstst (const vector float *, int, const int);
7757 void vec_dstst (const unsigned char *, int, const int);
7758 void vec_dstst (const signed char *, int, const int);
7759 void vec_dstst (const unsigned short *, int, const int);
7760 void vec_dstst (const short *, int, const int);
7761 void vec_dstst (const unsigned int *, int, const int);
7762 void vec_dstst (const int *, int, const int);
7763 void vec_dstst (const unsigned long *, int, const int);
7764 void vec_dstst (const long *, int, const int);
7765 void vec_dstst (const float *, int, const int);
7767 void vec_dststt (const vector unsigned char *, int, const int);
7768 void vec_dststt (const vector signed char *, int, const int);
7769 void vec_dststt (const vector bool char *, int, const int);
7770 void vec_dststt (const vector unsigned short *, int, const int);
7771 void vec_dststt (const vector signed short *, int, const int);
7772 void vec_dststt (const vector bool short *, int, const int);
7773 void vec_dststt (const vector pixel *, int, const int);
7774 void vec_dststt (const vector unsigned int *, int, const int);
7775 void vec_dststt (const vector signed int *, int, const int);
7776 void vec_dststt (const vector bool int *, int, const int);
7777 void vec_dststt (const vector float *, int, const int);
7778 void vec_dststt (const unsigned char *, int, const int);
7779 void vec_dststt (const signed char *, int, const int);
7780 void vec_dststt (const unsigned short *, int, const int);
7781 void vec_dststt (const short *, int, const int);
7782 void vec_dststt (const unsigned int *, int, const int);
7783 void vec_dststt (const int *, int, const int);
7784 void vec_dststt (const unsigned long *, int, const int);
7785 void vec_dststt (const long *, int, const int);
7786 void vec_dststt (const float *, int, const int);
7788 void vec_dstt (const vector unsigned char *, int, const int);
7789 void vec_dstt (const vector signed char *, int, const int);
7790 void vec_dstt (const vector bool char *, int, const int);
7791 void vec_dstt (const vector unsigned short *, int, const int);
7792 void vec_dstt (const vector signed short *, int, const int);
7793 void vec_dstt (const vector bool short *, int, const int);
7794 void vec_dstt (const vector pixel *, int, const int);
7795 void vec_dstt (const vector unsigned int *, int, const int);
7796 void vec_dstt (const vector signed int *, int, const int);
7797 void vec_dstt (const vector bool int *, int, const int);
7798 void vec_dstt (const vector float *, int, const int);
7799 void vec_dstt (const unsigned char *, int, const int);
7800 void vec_dstt (const signed char *, int, const int);
7801 void vec_dstt (const unsigned short *, int, const int);
7802 void vec_dstt (const short *, int, const int);
7803 void vec_dstt (const unsigned int *, int, const int);
7804 void vec_dstt (const int *, int, const int);
7805 void vec_dstt (const unsigned long *, int, const int);
7806 void vec_dstt (const long *, int, const int);
7807 void vec_dstt (const float *, int, const int);
7809 vector float vec_expte (vector float);
7811 vector float vec_floor (vector float);
7813 vector float vec_ld (int, const vector float *);
7814 vector float vec_ld (int, const float *);
7815 vector bool int vec_ld (int, const vector bool int *);
7816 vector signed int vec_ld (int, const vector signed int *);
7817 vector signed int vec_ld (int, const int *);
7818 vector signed int vec_ld (int, const long *);
7819 vector unsigned int vec_ld (int, const vector unsigned int *);
7820 vector unsigned int vec_ld (int, const unsigned int *);
7821 vector unsigned int vec_ld (int, const unsigned long *);
7822 vector bool short vec_ld (int, const vector bool short *);
7823 vector pixel vec_ld (int, const vector pixel *);
7824 vector signed short vec_ld (int, const vector signed short *);
7825 vector signed short vec_ld (int, const short *);
7826 vector unsigned short vec_ld (int, const vector unsigned short *);
7827 vector unsigned short vec_ld (int, const unsigned short *);
7828 vector bool char vec_ld (int, const vector bool char *);
7829 vector signed char vec_ld (int, const vector signed char *);
7830 vector signed char vec_ld (int, const signed char *);
7831 vector unsigned char vec_ld (int, const vector unsigned char *);
7832 vector unsigned char vec_ld (int, const unsigned char *);
7834 vector signed char vec_lde (int, const signed char *);
7835 vector unsigned char vec_lde (int, const unsigned char *);
7836 vector signed short vec_lde (int, const short *);
7837 vector unsigned short vec_lde (int, const unsigned short *);
7838 vector float vec_lde (int, const float *);
7839 vector signed int vec_lde (int, const int *);
7840 vector unsigned int vec_lde (int, const unsigned int *);
7841 vector signed int vec_lde (int, const long *);
7842 vector unsigned int vec_lde (int, const unsigned long *);
7844 vector float vec_lvewx (int, float *);
7845 vector signed int vec_lvewx (int, int *);
7846 vector unsigned int vec_lvewx (int, unsigned int *);
7847 vector signed int vec_lvewx (int, long *);
7848 vector unsigned int vec_lvewx (int, unsigned long *);
7850 vector signed short vec_lvehx (int, short *);
7851 vector unsigned short vec_lvehx (int, unsigned short *);
7853 vector signed char vec_lvebx (int, char *);
7854 vector unsigned char vec_lvebx (int, unsigned char *);
7856 vector float vec_ldl (int, const vector float *);
7857 vector float vec_ldl (int, const float *);
7858 vector bool int vec_ldl (int, const vector bool int *);
7859 vector signed int vec_ldl (int, const vector signed int *);
7860 vector signed int vec_ldl (int, const int *);
7861 vector signed int vec_ldl (int, const long *);
7862 vector unsigned int vec_ldl (int, const vector unsigned int *);
7863 vector unsigned int vec_ldl (int, const unsigned int *);
7864 vector unsigned int vec_ldl (int, const unsigned long *);
7865 vector bool short vec_ldl (int, const vector bool short *);
7866 vector pixel vec_ldl (int, const vector pixel *);
7867 vector signed short vec_ldl (int, const vector signed short *);
7868 vector signed short vec_ldl (int, const short *);
7869 vector unsigned short vec_ldl (int, const vector unsigned short *);
7870 vector unsigned short vec_ldl (int, const unsigned short *);
7871 vector bool char vec_ldl (int, const vector bool char *);
7872 vector signed char vec_ldl (int, const vector signed char *);
7873 vector signed char vec_ldl (int, const signed char *);
7874 vector unsigned char vec_ldl (int, const vector unsigned char *);
7875 vector unsigned char vec_ldl (int, const unsigned char *);
7877 vector float vec_loge (vector float);
7879 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
7880 vector unsigned char vec_lvsl (int, const volatile signed char *);
7881 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
7882 vector unsigned char vec_lvsl (int, const volatile short *);
7883 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
7884 vector unsigned char vec_lvsl (int, const volatile int *);
7885 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
7886 vector unsigned char vec_lvsl (int, const volatile long *);
7887 vector unsigned char vec_lvsl (int, const volatile float *);
7889 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
7890 vector unsigned char vec_lvsr (int, const volatile signed char *);
7891 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
7892 vector unsigned char vec_lvsr (int, const volatile short *);
7893 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
7894 vector unsigned char vec_lvsr (int, const volatile int *);
7895 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
7896 vector unsigned char vec_lvsr (int, const volatile long *);
7897 vector unsigned char vec_lvsr (int, const volatile float *);
7899 vector float vec_madd (vector float, vector float, vector float);
7901 vector signed short vec_madds (vector signed short,
7902 vector signed short,
7903 vector signed short);
7905 vector unsigned char vec_max (vector bool char, vector unsigned char);
7906 vector unsigned char vec_max (vector unsigned char, vector bool char);
7907 vector unsigned char vec_max (vector unsigned char,
7908 vector unsigned char);
7909 vector signed char vec_max (vector bool char, vector signed char);
7910 vector signed char vec_max (vector signed char, vector bool char);
7911 vector signed char vec_max (vector signed char, vector signed char);
7912 vector unsigned short vec_max (vector bool short,
7913 vector unsigned short);
7914 vector unsigned short vec_max (vector unsigned short,
7916 vector unsigned short vec_max (vector unsigned short,
7917 vector unsigned short);
7918 vector signed short vec_max (vector bool short, vector signed short);
7919 vector signed short vec_max (vector signed short, vector bool short);
7920 vector signed short vec_max (vector signed short, vector signed short);
7921 vector unsigned int vec_max (vector bool int, vector unsigned int);
7922 vector unsigned int vec_max (vector unsigned int, vector bool int);
7923 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
7924 vector signed int vec_max (vector bool int, vector signed int);
7925 vector signed int vec_max (vector signed int, vector bool int);
7926 vector signed int vec_max (vector signed int, vector signed int);
7927 vector float vec_max (vector float, vector float);
7929 vector float vec_vmaxfp (vector float, vector float);
7931 vector signed int vec_vmaxsw (vector bool int, vector signed int);
7932 vector signed int vec_vmaxsw (vector signed int, vector bool int);
7933 vector signed int vec_vmaxsw (vector signed int, vector signed int);
7935 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
7936 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
7937 vector unsigned int vec_vmaxuw (vector unsigned int,
7938 vector unsigned int);
7940 vector signed short vec_vmaxsh (vector bool short, vector signed short);
7941 vector signed short vec_vmaxsh (vector signed short, vector bool short);
7942 vector signed short vec_vmaxsh (vector signed short,
7943 vector signed short);
7945 vector unsigned short vec_vmaxuh (vector bool short,
7946 vector unsigned short);
7947 vector unsigned short vec_vmaxuh (vector unsigned short,
7949 vector unsigned short vec_vmaxuh (vector unsigned short,
7950 vector unsigned short);
7952 vector signed char vec_vmaxsb (vector bool char, vector signed char);
7953 vector signed char vec_vmaxsb (vector signed char, vector bool char);
7954 vector signed char vec_vmaxsb (vector signed char, vector signed char);
7956 vector unsigned char vec_vmaxub (vector bool char,
7957 vector unsigned char);
7958 vector unsigned char vec_vmaxub (vector unsigned char,
7960 vector unsigned char vec_vmaxub (vector unsigned char,
7961 vector unsigned char);
7963 vector bool char vec_mergeh (vector bool char, vector bool char);
7964 vector signed char vec_mergeh (vector signed char, vector signed char);
7965 vector unsigned char vec_mergeh (vector unsigned char,
7966 vector unsigned char);
7967 vector bool short vec_mergeh (vector bool short, vector bool short);
7968 vector pixel vec_mergeh (vector pixel, vector pixel);
7969 vector signed short vec_mergeh (vector signed short,
7970 vector signed short);
7971 vector unsigned short vec_mergeh (vector unsigned short,
7972 vector unsigned short);
7973 vector float vec_mergeh (vector float, vector float);
7974 vector bool int vec_mergeh (vector bool int, vector bool int);
7975 vector signed int vec_mergeh (vector signed int, vector signed int);
7976 vector unsigned int vec_mergeh (vector unsigned int,
7977 vector unsigned int);
7979 vector float vec_vmrghw (vector float, vector float);
7980 vector bool int vec_vmrghw (vector bool int, vector bool int);
7981 vector signed int vec_vmrghw (vector signed int, vector signed int);
7982 vector unsigned int vec_vmrghw (vector unsigned int,
7983 vector unsigned int);
7985 vector bool short vec_vmrghh (vector bool short, vector bool short);
7986 vector signed short vec_vmrghh (vector signed short,
7987 vector signed short);
7988 vector unsigned short vec_vmrghh (vector unsigned short,
7989 vector unsigned short);
7990 vector pixel vec_vmrghh (vector pixel, vector pixel);
7992 vector bool char vec_vmrghb (vector bool char, vector bool char);
7993 vector signed char vec_vmrghb (vector signed char, vector signed char);
7994 vector unsigned char vec_vmrghb (vector unsigned char,
7995 vector unsigned char);
7997 vector bool char vec_mergel (vector bool char, vector bool char);
7998 vector signed char vec_mergel (vector signed char, vector signed char);
7999 vector unsigned char vec_mergel (vector unsigned char,
8000 vector unsigned char);
8001 vector bool short vec_mergel (vector bool short, vector bool short);
8002 vector pixel vec_mergel (vector pixel, vector pixel);
8003 vector signed short vec_mergel (vector signed short,
8004 vector signed short);
8005 vector unsigned short vec_mergel (vector unsigned short,
8006 vector unsigned short);
8007 vector float vec_mergel (vector float, vector float);
8008 vector bool int vec_mergel (vector bool int, vector bool int);
8009 vector signed int vec_mergel (vector signed int, vector signed int);
8010 vector unsigned int vec_mergel (vector unsigned int,
8011 vector unsigned int);
8013 vector float vec_vmrglw (vector float, vector float);
8014 vector signed int vec_vmrglw (vector signed int, vector signed int);
8015 vector unsigned int vec_vmrglw (vector unsigned int,
8016 vector unsigned int);
8017 vector bool int vec_vmrglw (vector bool int, vector bool int);
8019 vector bool short vec_vmrglh (vector bool short, vector bool short);
8020 vector signed short vec_vmrglh (vector signed short,
8021 vector signed short);
8022 vector unsigned short vec_vmrglh (vector unsigned short,
8023 vector unsigned short);
8024 vector pixel vec_vmrglh (vector pixel, vector pixel);
8026 vector bool char vec_vmrglb (vector bool char, vector bool char);
8027 vector signed char vec_vmrglb (vector signed char, vector signed char);
8028 vector unsigned char vec_vmrglb (vector unsigned char,
8029 vector unsigned char);
8031 vector unsigned short vec_mfvscr (void);
8033 vector unsigned char vec_min (vector bool char, vector unsigned char);
8034 vector unsigned char vec_min (vector unsigned char, vector bool char);
8035 vector unsigned char vec_min (vector unsigned char,
8036 vector unsigned char);
8037 vector signed char vec_min (vector bool char, vector signed char);
8038 vector signed char vec_min (vector signed char, vector bool char);
8039 vector signed char vec_min (vector signed char, vector signed char);
8040 vector unsigned short vec_min (vector bool short,
8041 vector unsigned short);
8042 vector unsigned short vec_min (vector unsigned short,
8044 vector unsigned short vec_min (vector unsigned short,
8045 vector unsigned short);
8046 vector signed short vec_min (vector bool short, vector signed short);
8047 vector signed short vec_min (vector signed short, vector bool short);
8048 vector signed short vec_min (vector signed short, vector signed short);
8049 vector unsigned int vec_min (vector bool int, vector unsigned int);
8050 vector unsigned int vec_min (vector unsigned int, vector bool int);
8051 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
8052 vector signed int vec_min (vector bool int, vector signed int);
8053 vector signed int vec_min (vector signed int, vector bool int);
8054 vector signed int vec_min (vector signed int, vector signed int);
8055 vector float vec_min (vector float, vector float);
8057 vector float vec_vminfp (vector float, vector float);
8059 vector signed int vec_vminsw (vector bool int, vector signed int);
8060 vector signed int vec_vminsw (vector signed int, vector bool int);
8061 vector signed int vec_vminsw (vector signed int, vector signed int);
8063 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
8064 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
8065 vector unsigned int vec_vminuw (vector unsigned int,
8066 vector unsigned int);
8068 vector signed short vec_vminsh (vector bool short, vector signed short);
8069 vector signed short vec_vminsh (vector signed short, vector bool short);
8070 vector signed short vec_vminsh (vector signed short,
8071 vector signed short);
8073 vector unsigned short vec_vminuh (vector bool short,
8074 vector unsigned short);
8075 vector unsigned short vec_vminuh (vector unsigned short,
8077 vector unsigned short vec_vminuh (vector unsigned short,
8078 vector unsigned short);
8080 vector signed char vec_vminsb (vector bool char, vector signed char);
8081 vector signed char vec_vminsb (vector signed char, vector bool char);
8082 vector signed char vec_vminsb (vector signed char, vector signed char);
8084 vector unsigned char vec_vminub (vector bool char,
8085 vector unsigned char);
8086 vector unsigned char vec_vminub (vector unsigned char,
8088 vector unsigned char vec_vminub (vector unsigned char,
8089 vector unsigned char);
8091 vector signed short vec_mladd (vector signed short,
8092 vector signed short,
8093 vector signed short);
8094 vector signed short vec_mladd (vector signed short,
8095 vector unsigned short,
8096 vector unsigned short);
8097 vector signed short vec_mladd (vector unsigned short,
8098 vector signed short,
8099 vector signed short);
8100 vector unsigned short vec_mladd (vector unsigned short,
8101 vector unsigned short,
8102 vector unsigned short);
8104 vector signed short vec_mradds (vector signed short,
8105 vector signed short,
8106 vector signed short);
8108 vector unsigned int vec_msum (vector unsigned char,
8109 vector unsigned char,
8110 vector unsigned int);
8111 vector signed int vec_msum (vector signed char,
8112 vector unsigned char,
8114 vector unsigned int vec_msum (vector unsigned short,
8115 vector unsigned short,
8116 vector unsigned int);
8117 vector signed int vec_msum (vector signed short,
8118 vector signed short,
8121 vector signed int vec_vmsumshm (vector signed short,
8122 vector signed short,
8125 vector unsigned int vec_vmsumuhm (vector unsigned short,
8126 vector unsigned short,
8127 vector unsigned int);
8129 vector signed int vec_vmsummbm (vector signed char,
8130 vector unsigned char,
8133 vector unsigned int vec_vmsumubm (vector unsigned char,
8134 vector unsigned char,
8135 vector unsigned int);
8137 vector unsigned int vec_msums (vector unsigned short,
8138 vector unsigned short,
8139 vector unsigned int);
8140 vector signed int vec_msums (vector signed short,
8141 vector signed short,
8144 vector signed int vec_vmsumshs (vector signed short,
8145 vector signed short,
8148 vector unsigned int vec_vmsumuhs (vector unsigned short,
8149 vector unsigned short,
8150 vector unsigned int);
8152 void vec_mtvscr (vector signed int);
8153 void vec_mtvscr (vector unsigned int);
8154 void vec_mtvscr (vector bool int);
8155 void vec_mtvscr (vector signed short);
8156 void vec_mtvscr (vector unsigned short);
8157 void vec_mtvscr (vector bool short);
8158 void vec_mtvscr (vector pixel);
8159 void vec_mtvscr (vector signed char);
8160 void vec_mtvscr (vector unsigned char);
8161 void vec_mtvscr (vector bool char);
8163 vector unsigned short vec_mule (vector unsigned char,
8164 vector unsigned char);
8165 vector signed short vec_mule (vector signed char,
8166 vector signed char);
8167 vector unsigned int vec_mule (vector unsigned short,
8168 vector unsigned short);
8169 vector signed int vec_mule (vector signed short, vector signed short);
8171 vector signed int vec_vmulesh (vector signed short,
8172 vector signed short);
8174 vector unsigned int vec_vmuleuh (vector unsigned short,
8175 vector unsigned short);
8177 vector signed short vec_vmulesb (vector signed char,
8178 vector signed char);
8180 vector unsigned short vec_vmuleub (vector unsigned char,
8181 vector unsigned char);
8183 vector unsigned short vec_mulo (vector unsigned char,
8184 vector unsigned char);
8185 vector signed short vec_mulo (vector signed char, vector signed char);
8186 vector unsigned int vec_mulo (vector unsigned short,
8187 vector unsigned short);
8188 vector signed int vec_mulo (vector signed short, vector signed short);
8190 vector signed int vec_vmulosh (vector signed short,
8191 vector signed short);
8193 vector unsigned int vec_vmulouh (vector unsigned short,
8194 vector unsigned short);
8196 vector signed short vec_vmulosb (vector signed char,
8197 vector signed char);
8199 vector unsigned short vec_vmuloub (vector unsigned char,
8200 vector unsigned char);
8202 vector float vec_nmsub (vector float, vector float, vector float);
8204 vector float vec_nor (vector float, vector float);
8205 vector signed int vec_nor (vector signed int, vector signed int);
8206 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
8207 vector bool int vec_nor (vector bool int, vector bool int);
8208 vector signed short vec_nor (vector signed short, vector signed short);
8209 vector unsigned short vec_nor (vector unsigned short,
8210 vector unsigned short);
8211 vector bool short vec_nor (vector bool short, vector bool short);
8212 vector signed char vec_nor (vector signed char, vector signed char);
8213 vector unsigned char vec_nor (vector unsigned char,
8214 vector unsigned char);
8215 vector bool char vec_nor (vector bool char, vector bool char);
8217 vector float vec_or (vector float, vector float);
8218 vector float vec_or (vector float, vector bool int);
8219 vector float vec_or (vector bool int, vector float);
8220 vector bool int vec_or (vector bool int, vector bool int);
8221 vector signed int vec_or (vector bool int, vector signed int);
8222 vector signed int vec_or (vector signed int, vector bool int);
8223 vector signed int vec_or (vector signed int, vector signed int);
8224 vector unsigned int vec_or (vector bool int, vector unsigned int);
8225 vector unsigned int vec_or (vector unsigned int, vector bool int);
8226 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
8227 vector bool short vec_or (vector bool short, vector bool short);
8228 vector signed short vec_or (vector bool short, vector signed short);
8229 vector signed short vec_or (vector signed short, vector bool short);
8230 vector signed short vec_or (vector signed short, vector signed short);
8231 vector unsigned short vec_or (vector bool short, vector unsigned short);
8232 vector unsigned short vec_or (vector unsigned short, vector bool short);
8233 vector unsigned short vec_or (vector unsigned short,
8234 vector unsigned short);
8235 vector signed char vec_or (vector bool char, vector signed char);
8236 vector bool char vec_or (vector bool char, vector bool char);
8237 vector signed char vec_or (vector signed char, vector bool char);
8238 vector signed char vec_or (vector signed char, vector signed char);
8239 vector unsigned char vec_or (vector bool char, vector unsigned char);
8240 vector unsigned char vec_or (vector unsigned char, vector bool char);
8241 vector unsigned char vec_or (vector unsigned char,
8242 vector unsigned char);
8244 vector signed char vec_pack (vector signed short, vector signed short);
8245 vector unsigned char vec_pack (vector unsigned short,
8246 vector unsigned short);
8247 vector bool char vec_pack (vector bool short, vector bool short);
8248 vector signed short vec_pack (vector signed int, vector signed int);
8249 vector unsigned short vec_pack (vector unsigned int,
8250 vector unsigned int);
8251 vector bool short vec_pack (vector bool int, vector bool int);
8253 vector bool short vec_vpkuwum (vector bool int, vector bool int);
8254 vector signed short vec_vpkuwum (vector signed int, vector signed int);
8255 vector unsigned short vec_vpkuwum (vector unsigned int,
8256 vector unsigned int);
8258 vector bool char vec_vpkuhum (vector bool short, vector bool short);
8259 vector signed char vec_vpkuhum (vector signed short,
8260 vector signed short);
8261 vector unsigned char vec_vpkuhum (vector unsigned short,
8262 vector unsigned short);
8264 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
8266 vector unsigned char vec_packs (vector unsigned short,
8267 vector unsigned short);
8268 vector signed char vec_packs (vector signed short, vector signed short);
8269 vector unsigned short vec_packs (vector unsigned int,
8270 vector unsigned int);
8271 vector signed short vec_packs (vector signed int, vector signed int);
8273 vector signed short vec_vpkswss (vector signed int, vector signed int);
8275 vector unsigned short vec_vpkuwus (vector unsigned int,
8276 vector unsigned int);
8278 vector signed char vec_vpkshss (vector signed short,
8279 vector signed short);
8281 vector unsigned char vec_vpkuhus (vector unsigned short,
8282 vector unsigned short);
8284 vector unsigned char vec_packsu (vector unsigned short,
8285 vector unsigned short);
8286 vector unsigned char vec_packsu (vector signed short,
8287 vector signed short);
8288 vector unsigned short vec_packsu (vector unsigned int,
8289 vector unsigned int);
8290 vector unsigned short vec_packsu (vector signed int, vector signed int);
8292 vector unsigned short vec_vpkswus (vector signed int,
8295 vector unsigned char vec_vpkshus (vector signed short,
8296 vector signed short);
8298 vector float vec_perm (vector float,
8300 vector unsigned char);
8301 vector signed int vec_perm (vector signed int,
8303 vector unsigned char);
8304 vector unsigned int vec_perm (vector unsigned int,
8305 vector unsigned int,
8306 vector unsigned char);
8307 vector bool int vec_perm (vector bool int,
8309 vector unsigned char);
8310 vector signed short vec_perm (vector signed short,
8311 vector signed short,
8312 vector unsigned char);
8313 vector unsigned short vec_perm (vector unsigned short,
8314 vector unsigned short,
8315 vector unsigned char);
8316 vector bool short vec_perm (vector bool short,
8318 vector unsigned char);
8319 vector pixel vec_perm (vector pixel,
8321 vector unsigned char);
8322 vector signed char vec_perm (vector signed char,
8324 vector unsigned char);
8325 vector unsigned char vec_perm (vector unsigned char,
8326 vector unsigned char,
8327 vector unsigned char);
8328 vector bool char vec_perm (vector bool char,
8330 vector unsigned char);
8332 vector float vec_re (vector float);
8334 vector signed char vec_rl (vector signed char,
8335 vector unsigned char);
8336 vector unsigned char vec_rl (vector unsigned char,
8337 vector unsigned char);
8338 vector signed short vec_rl (vector signed short, vector unsigned short);
8339 vector unsigned short vec_rl (vector unsigned short,
8340 vector unsigned short);
8341 vector signed int vec_rl (vector signed int, vector unsigned int);
8342 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
8344 vector signed int vec_vrlw (vector signed int, vector unsigned int);
8345 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
8347 vector signed short vec_vrlh (vector signed short,
8348 vector unsigned short);
8349 vector unsigned short vec_vrlh (vector unsigned short,
8350 vector unsigned short);
8352 vector signed char vec_vrlb (vector signed char, vector unsigned char);
8353 vector unsigned char vec_vrlb (vector unsigned char,
8354 vector unsigned char);
8356 vector float vec_round (vector float);
8358 vector float vec_rsqrte (vector float);
8360 vector float vec_sel (vector float, vector float, vector bool int);
8361 vector float vec_sel (vector float, vector float, vector unsigned int);
8362 vector signed int vec_sel (vector signed int,
8365 vector signed int vec_sel (vector signed int,
8367 vector unsigned int);
8368 vector unsigned int vec_sel (vector unsigned int,
8369 vector unsigned int,
8371 vector unsigned int vec_sel (vector unsigned int,
8372 vector unsigned int,
8373 vector unsigned int);
8374 vector bool int vec_sel (vector bool int,
8377 vector bool int vec_sel (vector bool int,
8379 vector unsigned int);
8380 vector signed short vec_sel (vector signed short,
8381 vector signed short,
8383 vector signed short vec_sel (vector signed short,
8384 vector signed short,
8385 vector unsigned short);
8386 vector unsigned short vec_sel (vector unsigned short,
8387 vector unsigned short,
8389 vector unsigned short vec_sel (vector unsigned short,
8390 vector unsigned short,
8391 vector unsigned short);
8392 vector bool short vec_sel (vector bool short,
8395 vector bool short vec_sel (vector bool short,
8397 vector unsigned short);
8398 vector signed char vec_sel (vector signed char,
8401 vector signed char vec_sel (vector signed char,
8403 vector unsigned char);
8404 vector unsigned char vec_sel (vector unsigned char,
8405 vector unsigned char,
8407 vector unsigned char vec_sel (vector unsigned char,
8408 vector unsigned char,
8409 vector unsigned char);
8410 vector bool char vec_sel (vector bool char,
8413 vector bool char vec_sel (vector bool char,
8415 vector unsigned char);
8417 vector signed char vec_sl (vector signed char,
8418 vector unsigned char);
8419 vector unsigned char vec_sl (vector unsigned char,
8420 vector unsigned char);
8421 vector signed short vec_sl (vector signed short, vector unsigned short);
8422 vector unsigned short vec_sl (vector unsigned short,
8423 vector unsigned short);
8424 vector signed int vec_sl (vector signed int, vector unsigned int);
8425 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
8427 vector signed int vec_vslw (vector signed int, vector unsigned int);
8428 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
8430 vector signed short vec_vslh (vector signed short,
8431 vector unsigned short);
8432 vector unsigned short vec_vslh (vector unsigned short,
8433 vector unsigned short);
8435 vector signed char vec_vslb (vector signed char, vector unsigned char);
8436 vector unsigned char vec_vslb (vector unsigned char,
8437 vector unsigned char);
8439 vector float vec_sld (vector float, vector float, const int);
8440 vector signed int vec_sld (vector signed int,
8443 vector unsigned int vec_sld (vector unsigned int,
8444 vector unsigned int,
8446 vector bool int vec_sld (vector bool int,
8449 vector signed short vec_sld (vector signed short,
8450 vector signed short,
8452 vector unsigned short vec_sld (vector unsigned short,
8453 vector unsigned short,
8455 vector bool short vec_sld (vector bool short,
8458 vector pixel vec_sld (vector pixel,
8461 vector signed char vec_sld (vector signed char,
8464 vector unsigned char vec_sld (vector unsigned char,
8465 vector unsigned char,
8467 vector bool char vec_sld (vector bool char,
8471 vector signed int vec_sll (vector signed int,
8472 vector unsigned int);
8473 vector signed int vec_sll (vector signed int,
8474 vector unsigned short);
8475 vector signed int vec_sll (vector signed int,
8476 vector unsigned char);
8477 vector unsigned int vec_sll (vector unsigned int,
8478 vector unsigned int);
8479 vector unsigned int vec_sll (vector unsigned int,
8480 vector unsigned short);
8481 vector unsigned int vec_sll (vector unsigned int,
8482 vector unsigned char);
8483 vector bool int vec_sll (vector bool int,
8484 vector unsigned int);
8485 vector bool int vec_sll (vector bool int,
8486 vector unsigned short);
8487 vector bool int vec_sll (vector bool int,
8488 vector unsigned char);
8489 vector signed short vec_sll (vector signed short,
8490 vector unsigned int);
8491 vector signed short vec_sll (vector signed short,
8492 vector unsigned short);
8493 vector signed short vec_sll (vector signed short,
8494 vector unsigned char);
8495 vector unsigned short vec_sll (vector unsigned short,
8496 vector unsigned int);
8497 vector unsigned short vec_sll (vector unsigned short,
8498 vector unsigned short);
8499 vector unsigned short vec_sll (vector unsigned short,
8500 vector unsigned char);
8501 vector bool short vec_sll (vector bool short, vector unsigned int);
8502 vector bool short vec_sll (vector bool short, vector unsigned short);
8503 vector bool short vec_sll (vector bool short, vector unsigned char);
8504 vector pixel vec_sll (vector pixel, vector unsigned int);
8505 vector pixel vec_sll (vector pixel, vector unsigned short);
8506 vector pixel vec_sll (vector pixel, vector unsigned char);
8507 vector signed char vec_sll (vector signed char, vector unsigned int);
8508 vector signed char vec_sll (vector signed char, vector unsigned short);
8509 vector signed char vec_sll (vector signed char, vector unsigned char);
8510 vector unsigned char vec_sll (vector unsigned char,
8511 vector unsigned int);
8512 vector unsigned char vec_sll (vector unsigned char,
8513 vector unsigned short);
8514 vector unsigned char vec_sll (vector unsigned char,
8515 vector unsigned char);
8516 vector bool char vec_sll (vector bool char, vector unsigned int);
8517 vector bool char vec_sll (vector bool char, vector unsigned short);
8518 vector bool char vec_sll (vector bool char, vector unsigned char);
8520 vector float vec_slo (vector float, vector signed char);
8521 vector float vec_slo (vector float, vector unsigned char);
8522 vector signed int vec_slo (vector signed int, vector signed char);
8523 vector signed int vec_slo (vector signed int, vector unsigned char);
8524 vector unsigned int vec_slo (vector unsigned int, vector signed char);
8525 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
8526 vector signed short vec_slo (vector signed short, vector signed char);
8527 vector signed short vec_slo (vector signed short, vector unsigned char);
8528 vector unsigned short vec_slo (vector unsigned short,
8529 vector signed char);
8530 vector unsigned short vec_slo (vector unsigned short,
8531 vector unsigned char);
8532 vector pixel vec_slo (vector pixel, vector signed char);
8533 vector pixel vec_slo (vector pixel, vector unsigned char);
8534 vector signed char vec_slo (vector signed char, vector signed char);
8535 vector signed char vec_slo (vector signed char, vector unsigned char);
8536 vector unsigned char vec_slo (vector unsigned char, vector signed char);
8537 vector unsigned char vec_slo (vector unsigned char,
8538 vector unsigned char);
8540 vector signed char vec_splat (vector signed char, const int);
8541 vector unsigned char vec_splat (vector unsigned char, const int);
8542 vector bool char vec_splat (vector bool char, const int);
8543 vector signed short vec_splat (vector signed short, const int);
8544 vector unsigned short vec_splat (vector unsigned short, const int);
8545 vector bool short vec_splat (vector bool short, const int);
8546 vector pixel vec_splat (vector pixel, const int);
8547 vector float vec_splat (vector float, const int);
8548 vector signed int vec_splat (vector signed int, const int);
8549 vector unsigned int vec_splat (vector unsigned int, const int);
8550 vector bool int vec_splat (vector bool int, const int);
8552 vector float vec_vspltw (vector float, const int);
8553 vector signed int vec_vspltw (vector signed int, const int);
8554 vector unsigned int vec_vspltw (vector unsigned int, const int);
8555 vector bool int vec_vspltw (vector bool int, const int);
8557 vector bool short vec_vsplth (vector bool short, const int);
8558 vector signed short vec_vsplth (vector signed short, const int);
8559 vector unsigned short vec_vsplth (vector unsigned short, const int);
8560 vector pixel vec_vsplth (vector pixel, const int);
8562 vector signed char vec_vspltb (vector signed char, const int);
8563 vector unsigned char vec_vspltb (vector unsigned char, const int);
8564 vector bool char vec_vspltb (vector bool char, const int);
8566 vector signed char vec_splat_s8 (const int);
8568 vector signed short vec_splat_s16 (const int);
8570 vector signed int vec_splat_s32 (const int);
8572 vector unsigned char vec_splat_u8 (const int);
8574 vector unsigned short vec_splat_u16 (const int);
8576 vector unsigned int vec_splat_u32 (const int);
8578 vector signed char vec_sr (vector signed char, vector unsigned char);
8579 vector unsigned char vec_sr (vector unsigned char,
8580 vector unsigned char);
8581 vector signed short vec_sr (vector signed short,
8582 vector unsigned short);
8583 vector unsigned short vec_sr (vector unsigned short,
8584 vector unsigned short);
8585 vector signed int vec_sr (vector signed int, vector unsigned int);
8586 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
8588 vector signed int vec_vsrw (vector signed int, vector unsigned int);
8589 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
8591 vector signed short vec_vsrh (vector signed short,
8592 vector unsigned short);
8593 vector unsigned short vec_vsrh (vector unsigned short,
8594 vector unsigned short);
8596 vector signed char vec_vsrb (vector signed char, vector unsigned char);
8597 vector unsigned char vec_vsrb (vector unsigned char,
8598 vector unsigned char);
8600 vector signed char vec_sra (vector signed char, vector unsigned char);
8601 vector unsigned char vec_sra (vector unsigned char,
8602 vector unsigned char);
8603 vector signed short vec_sra (vector signed short,
8604 vector unsigned short);
8605 vector unsigned short vec_sra (vector unsigned short,
8606 vector unsigned short);
8607 vector signed int vec_sra (vector signed int, vector unsigned int);
8608 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
8610 vector signed int vec_vsraw (vector signed int, vector unsigned int);
8611 vector unsigned int vec_vsraw (vector unsigned int,
8612 vector unsigned int);
8614 vector signed short vec_vsrah (vector signed short,
8615 vector unsigned short);
8616 vector unsigned short vec_vsrah (vector unsigned short,
8617 vector unsigned short);
8619 vector signed char vec_vsrab (vector signed char, vector unsigned char);
8620 vector unsigned char vec_vsrab (vector unsigned char,
8621 vector unsigned char);
8623 vector signed int vec_srl (vector signed int, vector unsigned int);
8624 vector signed int vec_srl (vector signed int, vector unsigned short);
8625 vector signed int vec_srl (vector signed int, vector unsigned char);
8626 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
8627 vector unsigned int vec_srl (vector unsigned int,
8628 vector unsigned short);
8629 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
8630 vector bool int vec_srl (vector bool int, vector unsigned int);
8631 vector bool int vec_srl (vector bool int, vector unsigned short);
8632 vector bool int vec_srl (vector bool int, vector unsigned char);
8633 vector signed short vec_srl (vector signed short, vector unsigned int);
8634 vector signed short vec_srl (vector signed short,
8635 vector unsigned short);
8636 vector signed short vec_srl (vector signed short, vector unsigned char);
8637 vector unsigned short vec_srl (vector unsigned short,
8638 vector unsigned int);
8639 vector unsigned short vec_srl (vector unsigned short,
8640 vector unsigned short);
8641 vector unsigned short vec_srl (vector unsigned short,
8642 vector unsigned char);
8643 vector bool short vec_srl (vector bool short, vector unsigned int);
8644 vector bool short vec_srl (vector bool short, vector unsigned short);
8645 vector bool short vec_srl (vector bool short, vector unsigned char);
8646 vector pixel vec_srl (vector pixel, vector unsigned int);
8647 vector pixel vec_srl (vector pixel, vector unsigned short);
8648 vector pixel vec_srl (vector pixel, vector unsigned char);
8649 vector signed char vec_srl (vector signed char, vector unsigned int);
8650 vector signed char vec_srl (vector signed char, vector unsigned short);
8651 vector signed char vec_srl (vector signed char, vector unsigned char);
8652 vector unsigned char vec_srl (vector unsigned char,
8653 vector unsigned int);
8654 vector unsigned char vec_srl (vector unsigned char,
8655 vector unsigned short);
8656 vector unsigned char vec_srl (vector unsigned char,
8657 vector unsigned char);
8658 vector bool char vec_srl (vector bool char, vector unsigned int);
8659 vector bool char vec_srl (vector bool char, vector unsigned short);
8660 vector bool char vec_srl (vector bool char, vector unsigned char);
8662 vector float vec_sro (vector float, vector signed char);
8663 vector float vec_sro (vector float, vector unsigned char);
8664 vector signed int vec_sro (vector signed int, vector signed char);
8665 vector signed int vec_sro (vector signed int, vector unsigned char);
8666 vector unsigned int vec_sro (vector unsigned int, vector signed char);
8667 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
8668 vector signed short vec_sro (vector signed short, vector signed char);
8669 vector signed short vec_sro (vector signed short, vector unsigned char);
8670 vector unsigned short vec_sro (vector unsigned short,
8671 vector signed char);
8672 vector unsigned short vec_sro (vector unsigned short,
8673 vector unsigned char);
8674 vector pixel vec_sro (vector pixel, vector signed char);
8675 vector pixel vec_sro (vector pixel, vector unsigned char);
8676 vector signed char vec_sro (vector signed char, vector signed char);
8677 vector signed char vec_sro (vector signed char, vector unsigned char);
8678 vector unsigned char vec_sro (vector unsigned char, vector signed char);
8679 vector unsigned char vec_sro (vector unsigned char,
8680 vector unsigned char);
8682 void vec_st (vector float, int, vector float *);
8683 void vec_st (vector float, int, float *);
8684 void vec_st (vector signed int, int, vector signed int *);
8685 void vec_st (vector signed int, int, int *);
8686 void vec_st (vector unsigned int, int, vector unsigned int *);
8687 void vec_st (vector unsigned int, int, unsigned int *);
8688 void vec_st (vector bool int, int, vector bool int *);
8689 void vec_st (vector bool int, int, unsigned int *);
8690 void vec_st (vector bool int, int, int *);
8691 void vec_st (vector signed short, int, vector signed short *);
8692 void vec_st (vector signed short, int, short *);
8693 void vec_st (vector unsigned short, int, vector unsigned short *);
8694 void vec_st (vector unsigned short, int, unsigned short *);
8695 void vec_st (vector bool short, int, vector bool short *);
8696 void vec_st (vector bool short, int, unsigned short *);
8697 void vec_st (vector pixel, int, vector pixel *);
8698 void vec_st (vector pixel, int, unsigned short *);
8699 void vec_st (vector pixel, int, short *);
8700 void vec_st (vector bool short, int, short *);
8701 void vec_st (vector signed char, int, vector signed char *);
8702 void vec_st (vector signed char, int, signed char *);
8703 void vec_st (vector unsigned char, int, vector unsigned char *);
8704 void vec_st (vector unsigned char, int, unsigned char *);
8705 void vec_st (vector bool char, int, vector bool char *);
8706 void vec_st (vector bool char, int, unsigned char *);
8707 void vec_st (vector bool char, int, signed char *);
8709 void vec_ste (vector signed char, int, signed char *);
8710 void vec_ste (vector unsigned char, int, unsigned char *);
8711 void vec_ste (vector bool char, int, signed char *);
8712 void vec_ste (vector bool char, int, unsigned char *);
8713 void vec_ste (vector signed short, int, short *);
8714 void vec_ste (vector unsigned short, int, unsigned short *);
8715 void vec_ste (vector bool short, int, short *);
8716 void vec_ste (vector bool short, int, unsigned short *);
8717 void vec_ste (vector pixel, int, short *);
8718 void vec_ste (vector pixel, int, unsigned short *);
8719 void vec_ste (vector float, int, float *);
8720 void vec_ste (vector signed int, int, int *);
8721 void vec_ste (vector unsigned int, int, unsigned int *);
8722 void vec_ste (vector bool int, int, int *);
8723 void vec_ste (vector bool int, int, unsigned int *);
8725 void vec_stvewx (vector float, int, float *);
8726 void vec_stvewx (vector signed int, int, int *);
8727 void vec_stvewx (vector unsigned int, int, unsigned int *);
8728 void vec_stvewx (vector bool int, int, int *);
8729 void vec_stvewx (vector bool int, int, unsigned int *);
8731 void vec_stvehx (vector signed short, int, short *);
8732 void vec_stvehx (vector unsigned short, int, unsigned short *);
8733 void vec_stvehx (vector bool short, int, short *);
8734 void vec_stvehx (vector bool short, int, unsigned short *);
8735 void vec_stvehx (vector pixel, int, short *);
8736 void vec_stvehx (vector pixel, int, unsigned short *);
8738 void vec_stvebx (vector signed char, int, signed char *);
8739 void vec_stvebx (vector unsigned char, int, unsigned char *);
8740 void vec_stvebx (vector bool char, int, signed char *);
8741 void vec_stvebx (vector bool char, int, unsigned char *);
8743 void vec_stl (vector float, int, vector float *);
8744 void vec_stl (vector float, int, float *);
8745 void vec_stl (vector signed int, int, vector signed int *);
8746 void vec_stl (vector signed int, int, int *);
8747 void vec_stl (vector unsigned int, int, vector unsigned int *);
8748 void vec_stl (vector unsigned int, int, unsigned int *);
8749 void vec_stl (vector bool int, int, vector bool int *);
8750 void vec_stl (vector bool int, int, unsigned int *);
8751 void vec_stl (vector bool int, int, int *);
8752 void vec_stl (vector signed short, int, vector signed short *);
8753 void vec_stl (vector signed short, int, short *);
8754 void vec_stl (vector unsigned short, int, vector unsigned short *);
8755 void vec_stl (vector unsigned short, int, unsigned short *);
8756 void vec_stl (vector bool short, int, vector bool short *);
8757 void vec_stl (vector bool short, int, unsigned short *);
8758 void vec_stl (vector bool short, int, short *);
8759 void vec_stl (vector pixel, int, vector pixel *);
8760 void vec_stl (vector pixel, int, unsigned short *);
8761 void vec_stl (vector pixel, int, short *);
8762 void vec_stl (vector signed char, int, vector signed char *);
8763 void vec_stl (vector signed char, int, signed char *);
8764 void vec_stl (vector unsigned char, int, vector unsigned char *);
8765 void vec_stl (vector unsigned char, int, unsigned char *);
8766 void vec_stl (vector bool char, int, vector bool char *);
8767 void vec_stl (vector bool char, int, unsigned char *);
8768 void vec_stl (vector bool char, int, signed char *);
8770 vector signed char vec_sub (vector bool char, vector signed char);
8771 vector signed char vec_sub (vector signed char, vector bool char);
8772 vector signed char vec_sub (vector signed char, vector signed char);
8773 vector unsigned char vec_sub (vector bool char, vector unsigned char);
8774 vector unsigned char vec_sub (vector unsigned char, vector bool char);
8775 vector unsigned char vec_sub (vector unsigned char,
8776 vector unsigned char);
8777 vector signed short vec_sub (vector bool short, vector signed short);
8778 vector signed short vec_sub (vector signed short, vector bool short);
8779 vector signed short vec_sub (vector signed short, vector signed short);
8780 vector unsigned short vec_sub (vector bool short,
8781 vector unsigned short);
8782 vector unsigned short vec_sub (vector unsigned short,
8784 vector unsigned short vec_sub (vector unsigned short,
8785 vector unsigned short);
8786 vector signed int vec_sub (vector bool int, vector signed int);
8787 vector signed int vec_sub (vector signed int, vector bool int);
8788 vector signed int vec_sub (vector signed int, vector signed int);
8789 vector unsigned int vec_sub (vector bool int, vector unsigned int);
8790 vector unsigned int vec_sub (vector unsigned int, vector bool int);
8791 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
8792 vector float vec_sub (vector float, vector float);
8794 vector float vec_vsubfp (vector float, vector float);
8796 vector signed int vec_vsubuwm (vector bool int, vector signed int);
8797 vector signed int vec_vsubuwm (vector signed int, vector bool int);
8798 vector signed int vec_vsubuwm (vector signed int, vector signed int);
8799 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
8800 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
8801 vector unsigned int vec_vsubuwm (vector unsigned int,
8802 vector unsigned int);
8804 vector signed short vec_vsubuhm (vector bool short,
8805 vector signed short);
8806 vector signed short vec_vsubuhm (vector signed short,
8808 vector signed short vec_vsubuhm (vector signed short,
8809 vector signed short);
8810 vector unsigned short vec_vsubuhm (vector bool short,
8811 vector unsigned short);
8812 vector unsigned short vec_vsubuhm (vector unsigned short,
8814 vector unsigned short vec_vsubuhm (vector unsigned short,
8815 vector unsigned short);
8817 vector signed char vec_vsububm (vector bool char, vector signed char);
8818 vector signed char vec_vsububm (vector signed char, vector bool char);
8819 vector signed char vec_vsububm (vector signed char, vector signed char);
8820 vector unsigned char vec_vsububm (vector bool char,
8821 vector unsigned char);
8822 vector unsigned char vec_vsububm (vector unsigned char,
8824 vector unsigned char vec_vsububm (vector unsigned char,
8825 vector unsigned char);
8827 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
8829 vector unsigned char vec_subs (vector bool char, vector unsigned char);
8830 vector unsigned char vec_subs (vector unsigned char, vector bool char);
8831 vector unsigned char vec_subs (vector unsigned char,
8832 vector unsigned char);
8833 vector signed char vec_subs (vector bool char, vector signed char);
8834 vector signed char vec_subs (vector signed char, vector bool char);
8835 vector signed char vec_subs (vector signed char, vector signed char);
8836 vector unsigned short vec_subs (vector bool short,
8837 vector unsigned short);
8838 vector unsigned short vec_subs (vector unsigned short,
8840 vector unsigned short vec_subs (vector unsigned short,
8841 vector unsigned short);
8842 vector signed short vec_subs (vector bool short, vector signed short);
8843 vector signed short vec_subs (vector signed short, vector bool short);
8844 vector signed short vec_subs (vector signed short, vector signed short);
8845 vector unsigned int vec_subs (vector bool int, vector unsigned int);
8846 vector unsigned int vec_subs (vector unsigned int, vector bool int);
8847 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
8848 vector signed int vec_subs (vector bool int, vector signed int);
8849 vector signed int vec_subs (vector signed int, vector bool int);
8850 vector signed int vec_subs (vector signed int, vector signed int);
8852 vector signed int vec_vsubsws (vector bool int, vector signed int);
8853 vector signed int vec_vsubsws (vector signed int, vector bool int);
8854 vector signed int vec_vsubsws (vector signed int, vector signed int);
8856 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
8857 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
8858 vector unsigned int vec_vsubuws (vector unsigned int,
8859 vector unsigned int);
8861 vector signed short vec_vsubshs (vector bool short,
8862 vector signed short);
8863 vector signed short vec_vsubshs (vector signed short,
8865 vector signed short vec_vsubshs (vector signed short,
8866 vector signed short);
8868 vector unsigned short vec_vsubuhs (vector bool short,
8869 vector unsigned short);
8870 vector unsigned short vec_vsubuhs (vector unsigned short,
8872 vector unsigned short vec_vsubuhs (vector unsigned short,
8873 vector unsigned short);
8875 vector signed char vec_vsubsbs (vector bool char, vector signed char);
8876 vector signed char vec_vsubsbs (vector signed char, vector bool char);
8877 vector signed char vec_vsubsbs (vector signed char, vector signed char);
8879 vector unsigned char vec_vsububs (vector bool char,
8880 vector unsigned char);
8881 vector unsigned char vec_vsububs (vector unsigned char,
8883 vector unsigned char vec_vsububs (vector unsigned char,
8884 vector unsigned char);
8886 vector unsigned int vec_sum4s (vector unsigned char,
8887 vector unsigned int);
8888 vector signed int vec_sum4s (vector signed char, vector signed int);
8889 vector signed int vec_sum4s (vector signed short, vector signed int);
8891 vector signed int vec_vsum4shs (vector signed short, vector signed int);
8893 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
8895 vector unsigned int vec_vsum4ubs (vector unsigned char,
8896 vector unsigned int);
8898 vector signed int vec_sum2s (vector signed int, vector signed int);
8900 vector signed int vec_sums (vector signed int, vector signed int);
8902 vector float vec_trunc (vector float);
8904 vector signed short vec_unpackh (vector signed char);
8905 vector bool short vec_unpackh (vector bool char);
8906 vector signed int vec_unpackh (vector signed short);
8907 vector bool int vec_unpackh (vector bool short);
8908 vector unsigned int vec_unpackh (vector pixel);
8910 vector bool int vec_vupkhsh (vector bool short);
8911 vector signed int vec_vupkhsh (vector signed short);
8913 vector unsigned int vec_vupkhpx (vector pixel);
8915 vector bool short vec_vupkhsb (vector bool char);
8916 vector signed short vec_vupkhsb (vector signed char);
8918 vector signed short vec_unpackl (vector signed char);
8919 vector bool short vec_unpackl (vector bool char);
8920 vector unsigned int vec_unpackl (vector pixel);
8921 vector signed int vec_unpackl (vector signed short);
8922 vector bool int vec_unpackl (vector bool short);
8924 vector unsigned int vec_vupklpx (vector pixel);
8926 vector bool int vec_vupklsh (vector bool short);
8927 vector signed int vec_vupklsh (vector signed short);
8929 vector bool short vec_vupklsb (vector bool char);
8930 vector signed short vec_vupklsb (vector signed char);
8932 vector float vec_xor (vector float, vector float);
8933 vector float vec_xor (vector float, vector bool int);
8934 vector float vec_xor (vector bool int, vector float);
8935 vector bool int vec_xor (vector bool int, vector bool int);
8936 vector signed int vec_xor (vector bool int, vector signed int);
8937 vector signed int vec_xor (vector signed int, vector bool int);
8938 vector signed int vec_xor (vector signed int, vector signed int);
8939 vector unsigned int vec_xor (vector bool int, vector unsigned int);
8940 vector unsigned int vec_xor (vector unsigned int, vector bool int);
8941 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
8942 vector bool short vec_xor (vector bool short, vector bool short);
8943 vector signed short vec_xor (vector bool short, vector signed short);
8944 vector signed short vec_xor (vector signed short, vector bool short);
8945 vector signed short vec_xor (vector signed short, vector signed short);
8946 vector unsigned short vec_xor (vector bool short,
8947 vector unsigned short);
8948 vector unsigned short vec_xor (vector unsigned short,
8950 vector unsigned short vec_xor (vector unsigned short,
8951 vector unsigned short);
8952 vector signed char vec_xor (vector bool char, vector signed char);
8953 vector bool char vec_xor (vector bool char, vector bool char);
8954 vector signed char vec_xor (vector signed char, vector bool char);
8955 vector signed char vec_xor (vector signed char, vector signed char);
8956 vector unsigned char vec_xor (vector bool char, vector unsigned char);
8957 vector unsigned char vec_xor (vector unsigned char, vector bool char);
8958 vector unsigned char vec_xor (vector unsigned char,
8959 vector unsigned char);
8961 int vec_all_eq (vector signed char, vector bool char);
8962 int vec_all_eq (vector signed char, vector signed char);
8963 int vec_all_eq (vector unsigned char, vector bool char);
8964 int vec_all_eq (vector unsigned char, vector unsigned char);
8965 int vec_all_eq (vector bool char, vector bool char);
8966 int vec_all_eq (vector bool char, vector unsigned char);
8967 int vec_all_eq (vector bool char, vector signed char);
8968 int vec_all_eq (vector signed short, vector bool short);
8969 int vec_all_eq (vector signed short, vector signed short);
8970 int vec_all_eq (vector unsigned short, vector bool short);
8971 int vec_all_eq (vector unsigned short, vector unsigned short);
8972 int vec_all_eq (vector bool short, vector bool short);
8973 int vec_all_eq (vector bool short, vector unsigned short);
8974 int vec_all_eq (vector bool short, vector signed short);
8975 int vec_all_eq (vector pixel, vector pixel);
8976 int vec_all_eq (vector signed int, vector bool int);
8977 int vec_all_eq (vector signed int, vector signed int);
8978 int vec_all_eq (vector unsigned int, vector bool int);
8979 int vec_all_eq (vector unsigned int, vector unsigned int);
8980 int vec_all_eq (vector bool int, vector bool int);
8981 int vec_all_eq (vector bool int, vector unsigned int);
8982 int vec_all_eq (vector bool int, vector signed int);
8983 int vec_all_eq (vector float, vector float);
8985 int vec_all_ge (vector bool char, vector unsigned char);
8986 int vec_all_ge (vector unsigned char, vector bool char);
8987 int vec_all_ge (vector unsigned char, vector unsigned char);
8988 int vec_all_ge (vector bool char, vector signed char);
8989 int vec_all_ge (vector signed char, vector bool char);
8990 int vec_all_ge (vector signed char, vector signed char);
8991 int vec_all_ge (vector bool short, vector unsigned short);
8992 int vec_all_ge (vector unsigned short, vector bool short);
8993 int vec_all_ge (vector unsigned short, vector unsigned short);
8994 int vec_all_ge (vector signed short, vector signed short);
8995 int vec_all_ge (vector bool short, vector signed short);
8996 int vec_all_ge (vector signed short, vector bool short);
8997 int vec_all_ge (vector bool int, vector unsigned int);
8998 int vec_all_ge (vector unsigned int, vector bool int);
8999 int vec_all_ge (vector unsigned int, vector unsigned int);
9000 int vec_all_ge (vector bool int, vector signed int);
9001 int vec_all_ge (vector signed int, vector bool int);
9002 int vec_all_ge (vector signed int, vector signed int);
9003 int vec_all_ge (vector float, vector float);
9005 int vec_all_gt (vector bool char, vector unsigned char);
9006 int vec_all_gt (vector unsigned char, vector bool char);
9007 int vec_all_gt (vector unsigned char, vector unsigned char);
9008 int vec_all_gt (vector bool char, vector signed char);
9009 int vec_all_gt (vector signed char, vector bool char);
9010 int vec_all_gt (vector signed char, vector signed char);
9011 int vec_all_gt (vector bool short, vector unsigned short);
9012 int vec_all_gt (vector unsigned short, vector bool short);
9013 int vec_all_gt (vector unsigned short, vector unsigned short);
9014 int vec_all_gt (vector bool short, vector signed short);
9015 int vec_all_gt (vector signed short, vector bool short);
9016 int vec_all_gt (vector signed short, vector signed short);
9017 int vec_all_gt (vector bool int, vector unsigned int);
9018 int vec_all_gt (vector unsigned int, vector bool int);
9019 int vec_all_gt (vector unsigned int, vector unsigned int);
9020 int vec_all_gt (vector bool int, vector signed int);
9021 int vec_all_gt (vector signed int, vector bool int);
9022 int vec_all_gt (vector signed int, vector signed int);
9023 int vec_all_gt (vector float, vector float);
9025 int vec_all_in (vector float, vector float);
9027 int vec_all_le (vector bool char, vector unsigned char);
9028 int vec_all_le (vector unsigned char, vector bool char);
9029 int vec_all_le (vector unsigned char, vector unsigned char);
9030 int vec_all_le (vector bool char, vector signed char);
9031 int vec_all_le (vector signed char, vector bool char);
9032 int vec_all_le (vector signed char, vector signed char);
9033 int vec_all_le (vector bool short, vector unsigned short);
9034 int vec_all_le (vector unsigned short, vector bool short);
9035 int vec_all_le (vector unsigned short, vector unsigned short);
9036 int vec_all_le (vector bool short, vector signed short);
9037 int vec_all_le (vector signed short, vector bool short);
9038 int vec_all_le (vector signed short, vector signed short);
9039 int vec_all_le (vector bool int, vector unsigned int);
9040 int vec_all_le (vector unsigned int, vector bool int);
9041 int vec_all_le (vector unsigned int, vector unsigned int);
9042 int vec_all_le (vector bool int, vector signed int);
9043 int vec_all_le (vector signed int, vector bool int);
9044 int vec_all_le (vector signed int, vector signed int);
9045 int vec_all_le (vector float, vector float);
9047 int vec_all_lt (vector bool char, vector unsigned char);
9048 int vec_all_lt (vector unsigned char, vector bool char);
9049 int vec_all_lt (vector unsigned char, vector unsigned char);
9050 int vec_all_lt (vector bool char, vector signed char);
9051 int vec_all_lt (vector signed char, vector bool char);
9052 int vec_all_lt (vector signed char, vector signed char);
9053 int vec_all_lt (vector bool short, vector unsigned short);
9054 int vec_all_lt (vector unsigned short, vector bool short);
9055 int vec_all_lt (vector unsigned short, vector unsigned short);
9056 int vec_all_lt (vector bool short, vector signed short);
9057 int vec_all_lt (vector signed short, vector bool short);
9058 int vec_all_lt (vector signed short, vector signed short);
9059 int vec_all_lt (vector bool int, vector unsigned int);
9060 int vec_all_lt (vector unsigned int, vector bool int);
9061 int vec_all_lt (vector unsigned int, vector unsigned int);
9062 int vec_all_lt (vector bool int, vector signed int);
9063 int vec_all_lt (vector signed int, vector bool int);
9064 int vec_all_lt (vector signed int, vector signed int);
9065 int vec_all_lt (vector float, vector float);
9067 int vec_all_nan (vector float);
9069 int vec_all_ne (vector signed char, vector bool char);
9070 int vec_all_ne (vector signed char, vector signed char);
9071 int vec_all_ne (vector unsigned char, vector bool char);
9072 int vec_all_ne (vector unsigned char, vector unsigned char);
9073 int vec_all_ne (vector bool char, vector bool char);
9074 int vec_all_ne (vector bool char, vector unsigned char);
9075 int vec_all_ne (vector bool char, vector signed char);
9076 int vec_all_ne (vector signed short, vector bool short);
9077 int vec_all_ne (vector signed short, vector signed short);
9078 int vec_all_ne (vector unsigned short, vector bool short);
9079 int vec_all_ne (vector unsigned short, vector unsigned short);
9080 int vec_all_ne (vector bool short, vector bool short);
9081 int vec_all_ne (vector bool short, vector unsigned short);
9082 int vec_all_ne (vector bool short, vector signed short);
9083 int vec_all_ne (vector pixel, vector pixel);
9084 int vec_all_ne (vector signed int, vector bool int);
9085 int vec_all_ne (vector signed int, vector signed int);
9086 int vec_all_ne (vector unsigned int, vector bool int);
9087 int vec_all_ne (vector unsigned int, vector unsigned int);
9088 int vec_all_ne (vector bool int, vector bool int);
9089 int vec_all_ne (vector bool int, vector unsigned int);
9090 int vec_all_ne (vector bool int, vector signed int);
9091 int vec_all_ne (vector float, vector float);
9093 int vec_all_nge (vector float, vector float);
9095 int vec_all_ngt (vector float, vector float);
9097 int vec_all_nle (vector float, vector float);
9099 int vec_all_nlt (vector float, vector float);
9101 int vec_all_numeric (vector float);
9103 int vec_any_eq (vector signed char, vector bool char);
9104 int vec_any_eq (vector signed char, vector signed char);
9105 int vec_any_eq (vector unsigned char, vector bool char);
9106 int vec_any_eq (vector unsigned char, vector unsigned char);
9107 int vec_any_eq (vector bool char, vector bool char);
9108 int vec_any_eq (vector bool char, vector unsigned char);
9109 int vec_any_eq (vector bool char, vector signed char);
9110 int vec_any_eq (vector signed short, vector bool short);
9111 int vec_any_eq (vector signed short, vector signed short);
9112 int vec_any_eq (vector unsigned short, vector bool short);
9113 int vec_any_eq (vector unsigned short, vector unsigned short);
9114 int vec_any_eq (vector bool short, vector bool short);
9115 int vec_any_eq (vector bool short, vector unsigned short);
9116 int vec_any_eq (vector bool short, vector signed short);
9117 int vec_any_eq (vector pixel, vector pixel);
9118 int vec_any_eq (vector signed int, vector bool int);
9119 int vec_any_eq (vector signed int, vector signed int);
9120 int vec_any_eq (vector unsigned int, vector bool int);
9121 int vec_any_eq (vector unsigned int, vector unsigned int);
9122 int vec_any_eq (vector bool int, vector bool int);
9123 int vec_any_eq (vector bool int, vector unsigned int);
9124 int vec_any_eq (vector bool int, vector signed int);
9125 int vec_any_eq (vector float, vector float);
9127 int vec_any_ge (vector signed char, vector bool char);
9128 int vec_any_ge (vector unsigned char, vector bool char);
9129 int vec_any_ge (vector unsigned char, vector unsigned char);
9130 int vec_any_ge (vector signed char, vector signed char);
9131 int vec_any_ge (vector bool char, vector unsigned char);
9132 int vec_any_ge (vector bool char, vector signed char);
9133 int vec_any_ge (vector unsigned short, vector bool short);
9134 int vec_any_ge (vector unsigned short, vector unsigned short);
9135 int vec_any_ge (vector signed short, vector signed short);
9136 int vec_any_ge (vector signed short, vector bool short);
9137 int vec_any_ge (vector bool short, vector unsigned short);
9138 int vec_any_ge (vector bool short, vector signed short);
9139 int vec_any_ge (vector signed int, vector bool int);
9140 int vec_any_ge (vector unsigned int, vector bool int);
9141 int vec_any_ge (vector unsigned int, vector unsigned int);
9142 int vec_any_ge (vector signed int, vector signed int);
9143 int vec_any_ge (vector bool int, vector unsigned int);
9144 int vec_any_ge (vector bool int, vector signed int);
9145 int vec_any_ge (vector float, vector float);
9147 int vec_any_gt (vector bool char, vector unsigned char);
9148 int vec_any_gt (vector unsigned char, vector bool char);
9149 int vec_any_gt (vector unsigned char, vector unsigned char);
9150 int vec_any_gt (vector bool char, vector signed char);
9151 int vec_any_gt (vector signed char, vector bool char);
9152 int vec_any_gt (vector signed char, vector signed char);
9153 int vec_any_gt (vector bool short, vector unsigned short);
9154 int vec_any_gt (vector unsigned short, vector bool short);
9155 int vec_any_gt (vector unsigned short, vector unsigned short);
9156 int vec_any_gt (vector bool short, vector signed short);
9157 int vec_any_gt (vector signed short, vector bool short);
9158 int vec_any_gt (vector signed short, vector signed short);
9159 int vec_any_gt (vector bool int, vector unsigned int);
9160 int vec_any_gt (vector unsigned int, vector bool int);
9161 int vec_any_gt (vector unsigned int, vector unsigned int);
9162 int vec_any_gt (vector bool int, vector signed int);
9163 int vec_any_gt (vector signed int, vector bool int);
9164 int vec_any_gt (vector signed int, vector signed int);
9165 int vec_any_gt (vector float, vector float);
9167 int vec_any_le (vector bool char, vector unsigned char);
9168 int vec_any_le (vector unsigned char, vector bool char);
9169 int vec_any_le (vector unsigned char, vector unsigned char);
9170 int vec_any_le (vector bool char, vector signed char);
9171 int vec_any_le (vector signed char, vector bool char);
9172 int vec_any_le (vector signed char, vector signed char);
9173 int vec_any_le (vector bool short, vector unsigned short);
9174 int vec_any_le (vector unsigned short, vector bool short);
9175 int vec_any_le (vector unsigned short, vector unsigned short);
9176 int vec_any_le (vector bool short, vector signed short);
9177 int vec_any_le (vector signed short, vector bool short);
9178 int vec_any_le (vector signed short, vector signed short);
9179 int vec_any_le (vector bool int, vector unsigned int);
9180 int vec_any_le (vector unsigned int, vector bool int);
9181 int vec_any_le (vector unsigned int, vector unsigned int);
9182 int vec_any_le (vector bool int, vector signed int);
9183 int vec_any_le (vector signed int, vector bool int);
9184 int vec_any_le (vector signed int, vector signed int);
9185 int vec_any_le (vector float, vector float);
9187 int vec_any_lt (vector bool char, vector unsigned char);
9188 int vec_any_lt (vector unsigned char, vector bool char);
9189 int vec_any_lt (vector unsigned char, vector unsigned char);
9190 int vec_any_lt (vector bool char, vector signed char);
9191 int vec_any_lt (vector signed char, vector bool char);
9192 int vec_any_lt (vector signed char, vector signed char);
9193 int vec_any_lt (vector bool short, vector unsigned short);
9194 int vec_any_lt (vector unsigned short, vector bool short);
9195 int vec_any_lt (vector unsigned short, vector unsigned short);
9196 int vec_any_lt (vector bool short, vector signed short);
9197 int vec_any_lt (vector signed short, vector bool short);
9198 int vec_any_lt (vector signed short, vector signed short);
9199 int vec_any_lt (vector bool int, vector unsigned int);
9200 int vec_any_lt (vector unsigned int, vector bool int);
9201 int vec_any_lt (vector unsigned int, vector unsigned int);
9202 int vec_any_lt (vector bool int, vector signed int);
9203 int vec_any_lt (vector signed int, vector bool int);
9204 int vec_any_lt (vector signed int, vector signed int);
9205 int vec_any_lt (vector float, vector float);
9207 int vec_any_nan (vector float);
9209 int vec_any_ne (vector signed char, vector bool char);
9210 int vec_any_ne (vector signed char, vector signed char);
9211 int vec_any_ne (vector unsigned char, vector bool char);
9212 int vec_any_ne (vector unsigned char, vector unsigned char);
9213 int vec_any_ne (vector bool char, vector bool char);
9214 int vec_any_ne (vector bool char, vector unsigned char);
9215 int vec_any_ne (vector bool char, vector signed char);
9216 int vec_any_ne (vector signed short, vector bool short);
9217 int vec_any_ne (vector signed short, vector signed short);
9218 int vec_any_ne (vector unsigned short, vector bool short);
9219 int vec_any_ne (vector unsigned short, vector unsigned short);
9220 int vec_any_ne (vector bool short, vector bool short);
9221 int vec_any_ne (vector bool short, vector unsigned short);
9222 int vec_any_ne (vector bool short, vector signed short);
9223 int vec_any_ne (vector pixel, vector pixel);
9224 int vec_any_ne (vector signed int, vector bool int);
9225 int vec_any_ne (vector signed int, vector signed int);
9226 int vec_any_ne (vector unsigned int, vector bool int);
9227 int vec_any_ne (vector unsigned int, vector unsigned int);
9228 int vec_any_ne (vector bool int, vector bool int);
9229 int vec_any_ne (vector bool int, vector unsigned int);
9230 int vec_any_ne (vector bool int, vector signed int);
9231 int vec_any_ne (vector float, vector float);
9233 int vec_any_nge (vector float, vector float);
9235 int vec_any_ngt (vector float, vector float);
9237 int vec_any_nle (vector float, vector float);
9239 int vec_any_nlt (vector float, vector float);
9241 int vec_any_numeric (vector float);
9243 int vec_any_out (vector float, vector float);
9246 @node SPARC VIS Built-in Functions
9247 @subsection SPARC VIS Built-in Functions
9249 GCC supports SIMD operations on the SPARC using both the generic vector
9250 extensions (@pxref{Vector Extensions}) as well as built-in functions for
9251 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
9252 switch, the VIS extension is exposed as the following built-in functions:
9255 typedef int v2si __attribute__ ((vector_size (8)));
9256 typedef short v4hi __attribute__ ((vector_size (8)));
9257 typedef short v2hi __attribute__ ((vector_size (4)));
9258 typedef char v8qi __attribute__ ((vector_size (8)));
9259 typedef char v4qi __attribute__ ((vector_size (4)));
9261 void * __builtin_vis_alignaddr (void *, long);
9262 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
9263 v2si __builtin_vis_faligndatav2si (v2si, v2si);
9264 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
9265 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
9267 v4hi __builtin_vis_fexpand (v4qi);
9269 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
9270 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
9271 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
9272 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
9273 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
9274 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
9275 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
9277 v4qi __builtin_vis_fpack16 (v4hi);
9278 v8qi __builtin_vis_fpack32 (v2si, v2si);
9279 v2hi __builtin_vis_fpackfix (v2si);
9280 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
9282 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
9285 @node Target Format Checks
9286 @section Format Checks Specific to Particular Target Machines
9288 For some target machines, GCC supports additional options to the
9290 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
9293 * Solaris Format Checks::
9296 @node Solaris Format Checks
9297 @subsection Solaris Format Checks
9299 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
9300 check. @code{cmn_err} accepts a subset of the standard @code{printf}
9301 conversions, and the two-argument @code{%b} conversion for displaying
9302 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
9305 @section Pragmas Accepted by GCC
9309 GCC supports several types of pragmas, primarily in order to compile
9310 code originally written for other compilers. Note that in general
9311 we do not recommend the use of pragmas; @xref{Function Attributes},
9312 for further explanation.
9317 * RS/6000 and PowerPC Pragmas::
9320 * Symbol-Renaming Pragmas::
9321 * Structure-Packing Pragmas::
9323 * Diagnostic Pragmas::
9327 @subsection ARM Pragmas
9329 The ARM target defines pragmas for controlling the default addition of
9330 @code{long_call} and @code{short_call} attributes to functions.
9331 @xref{Function Attributes}, for information about the effects of these
9336 @cindex pragma, long_calls
9337 Set all subsequent functions to have the @code{long_call} attribute.
9340 @cindex pragma, no_long_calls
9341 Set all subsequent functions to have the @code{short_call} attribute.
9343 @item long_calls_off
9344 @cindex pragma, long_calls_off
9345 Do not affect the @code{long_call} or @code{short_call} attributes of
9346 subsequent functions.
9350 @subsection M32C Pragmas
9353 @item memregs @var{number}
9354 @cindex pragma, memregs
9355 Overrides the command line option @code{-memregs=} for the current
9356 file. Use with care! This pragma must be before any function in the
9357 file, and mixing different memregs values in different objects may
9358 make them incompatible. This pragma is useful when a
9359 performance-critical function uses a memreg for temporary values,
9360 as it may allow you to reduce the number of memregs used.
9364 @node RS/6000 and PowerPC Pragmas
9365 @subsection RS/6000 and PowerPC Pragmas
9367 The RS/6000 and PowerPC targets define one pragma for controlling
9368 whether or not the @code{longcall} attribute is added to function
9369 declarations by default. This pragma overrides the @option{-mlongcall}
9370 option, but not the @code{longcall} and @code{shortcall} attributes.
9371 @xref{RS/6000 and PowerPC Options}, for more information about when long
9372 calls are and are not necessary.
9376 @cindex pragma, longcall
9377 Apply the @code{longcall} attribute to all subsequent function
9381 Do not apply the @code{longcall} attribute to subsequent function
9385 @c Describe c4x pragmas here.
9386 @c Describe h8300 pragmas here.
9387 @c Describe sh pragmas here.
9388 @c Describe v850 pragmas here.
9390 @node Darwin Pragmas
9391 @subsection Darwin Pragmas
9393 The following pragmas are available for all architectures running the
9394 Darwin operating system. These are useful for compatibility with other
9398 @item mark @var{tokens}@dots{}
9399 @cindex pragma, mark
9400 This pragma is accepted, but has no effect.
9402 @item options align=@var{alignment}
9403 @cindex pragma, options align
9404 This pragma sets the alignment of fields in structures. The values of
9405 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
9406 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
9407 properly; to restore the previous setting, use @code{reset} for the
9410 @item segment @var{tokens}@dots{}
9411 @cindex pragma, segment
9412 This pragma is accepted, but has no effect.
9414 @item unused (@var{var} [, @var{var}]@dots{})
9415 @cindex pragma, unused
9416 This pragma declares variables to be possibly unused. GCC will not
9417 produce warnings for the listed variables. The effect is similar to
9418 that of the @code{unused} attribute, except that this pragma may appear
9419 anywhere within the variables' scopes.
9422 @node Solaris Pragmas
9423 @subsection Solaris Pragmas
9425 The Solaris target supports @code{#pragma redefine_extname}
9426 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
9427 @code{#pragma} directives for compatibility with the system compiler.
9430 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
9431 @cindex pragma, align
9433 Increase the minimum alignment of each @var{variable} to @var{alignment}.
9434 This is the same as GCC's @code{aligned} attribute @pxref{Variable
9435 Attributes}). Macro expansion occurs on the arguments to this pragma
9436 when compiling C and Objective-C. It does not currently occur when
9437 compiling C++, but this is a bug which may be fixed in a future
9440 @item fini (@var{function} [, @var{function}]...)
9441 @cindex pragma, fini
9443 This pragma causes each listed @var{function} to be called after
9444 main, or during shared module unloading, by adding a call to the
9445 @code{.fini} section.
9447 @item init (@var{function} [, @var{function}]...)
9448 @cindex pragma, init
9450 This pragma causes each listed @var{function} to be called during
9451 initialization (before @code{main}) or during shared module loading, by
9452 adding a call to the @code{.init} section.
9456 @node Symbol-Renaming Pragmas
9457 @subsection Symbol-Renaming Pragmas
9459 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
9460 supports two @code{#pragma} directives which change the name used in
9461 assembly for a given declaration. These pragmas are only available on
9462 platforms whose system headers need them. To get this effect on all
9463 platforms supported by GCC, use the asm labels extension (@pxref{Asm
9467 @item redefine_extname @var{oldname} @var{newname}
9468 @cindex pragma, redefine_extname
9470 This pragma gives the C function @var{oldname} the assembly symbol
9471 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
9472 will be defined if this pragma is available (currently only on
9475 @item extern_prefix @var{string}
9476 @cindex pragma, extern_prefix
9478 This pragma causes all subsequent external function and variable
9479 declarations to have @var{string} prepended to their assembly symbols.
9480 This effect may be terminated with another @code{extern_prefix} pragma
9481 whose argument is an empty string. The preprocessor macro
9482 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
9483 available (currently only on Tru64 UNIX)@.
9486 These pragmas and the asm labels extension interact in a complicated
9487 manner. Here are some corner cases you may want to be aware of.
9490 @item Both pragmas silently apply only to declarations with external
9491 linkage. Asm labels do not have this restriction.
9493 @item In C++, both pragmas silently apply only to declarations with
9494 ``C'' linkage. Again, asm labels do not have this restriction.
9496 @item If any of the three ways of changing the assembly name of a
9497 declaration is applied to a declaration whose assembly name has
9498 already been determined (either by a previous use of one of these
9499 features, or because the compiler needed the assembly name in order to
9500 generate code), and the new name is different, a warning issues and
9501 the name does not change.
9503 @item The @var{oldname} used by @code{#pragma redefine_extname} is
9504 always the C-language name.
9506 @item If @code{#pragma extern_prefix} is in effect, and a declaration
9507 occurs with an asm label attached, the prefix is silently ignored for
9510 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
9511 apply to the same declaration, whichever triggered first wins, and a
9512 warning issues if they contradict each other. (We would like to have
9513 @code{#pragma redefine_extname} always win, for consistency with asm
9514 labels, but if @code{#pragma extern_prefix} triggers first we have no
9515 way of knowing that that happened.)
9518 @node Structure-Packing Pragmas
9519 @subsection Structure-Packing Pragmas
9521 For compatibility with Win32, GCC supports a set of @code{#pragma}
9522 directives which change the maximum alignment of members of structures
9523 (other than zero-width bitfields), unions, and classes subsequently
9524 defined. The @var{n} value below always is required to be a small power
9525 of two and specifies the new alignment in bytes.
9528 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
9529 @item @code{#pragma pack()} sets the alignment to the one that was in
9530 effect when compilation started (see also command line option
9531 @option{-fpack-struct[=<n>]} @pxref{Code Gen Options}).
9532 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
9533 setting on an internal stack and then optionally sets the new alignment.
9534 @item @code{#pragma pack(pop)} restores the alignment setting to the one
9535 saved at the top of the internal stack (and removes that stack entry).
9536 Note that @code{#pragma pack([@var{n}])} does not influence this internal
9537 stack; thus it is possible to have @code{#pragma pack(push)} followed by
9538 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
9539 @code{#pragma pack(pop)}.
9543 @subsection Weak Pragmas
9545 For compatibility with SVR4, GCC supports a set of @code{#pragma}
9546 directives for declaring symbols to be weak, and defining weak
9550 @item #pragma weak @var{symbol}
9551 @cindex pragma, weak
9552 This pragma declares @var{symbol} to be weak, as if the declaration
9553 had the attribute of the same name. The pragma may appear before
9554 or after the declaration of @var{symbol}, but must appear before
9555 either its first use or its definition. It is not an error for
9556 @var{symbol} to never be defined at all.
9558 @item #pragma weak @var{symbol1} = @var{symbol2}
9559 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
9560 It is an error if @var{symbol2} is not defined in the current
9564 @node Diagnostic Pragmas
9565 @subsection Diagnostic Pragmas
9567 GCC allows the user to selectively enable or disable certain types of
9568 diagnostics, and change the kind of the diagnostic. For example, a
9569 project's policy might require that all sources compile with
9570 @option{-Werror} but certain files might have exceptions allowing
9571 specific types of warnings. Or, a project might selectively enable
9572 diagnostics and treat them as errors depending on which preprocessor
9576 @item #pragma GCC diagnostic @var{kind} @var{option}
9577 @cindex pragma, diagnostic
9579 Modifies the disposition of a diagnostic. Note that not all
9580 diagnostics are modifyiable; at the moment only warnings (normally
9581 controlled by @samp{-W...}) can be controlled, and not all of them.
9582 Use @option{-fdiagnostics-show-option} to determine which diagnostics
9583 are controllable and which option controls them.
9585 @var{kind} is @samp{error} to treat this diagnostic as an error,
9586 @samp{warning} to treat it like a warning (even if @option{-Werror} is
9587 in effect), or @samp{ignored} if the diagnostic is to be ignored.
9588 @var{option} is a double quoted string which matches the command line
9592 #pragma GCC diagnostic warning "-Wformat"
9593 #pragma GCC diagnostic error "-Walways-true"
9594 #pragma GCC diagnostic ignored "-Walways-true"
9597 Note that these pragmas override any command line options. Also,
9598 while it is syntactically valid to put these pragmas anywhere in your
9599 sources, the only supported location for them is before any data or
9600 functions are defined. Doing otherwise may result in unpredictable
9601 results depending on how the optimizer manages your sources. If the
9602 same option is listed multiple times, the last one specified is the
9603 one that is in effect. This pragma is not intended to be a general
9604 purpose replacement for command line options, but for implementing
9605 strict control over project policies.
9609 @node Unnamed Fields
9610 @section Unnamed struct/union fields within structs/unions
9614 For compatibility with other compilers, GCC allows you to define
9615 a structure or union that contains, as fields, structures and unions
9616 without names. For example:
9629 In this example, the user would be able to access members of the unnamed
9630 union with code like @samp{foo.b}. Note that only unnamed structs and
9631 unions are allowed, you may not have, for example, an unnamed
9634 You must never create such structures that cause ambiguous field definitions.
9635 For example, this structure:
9646 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
9647 Such constructs are not supported and must be avoided. In the future,
9648 such constructs may be detected and treated as compilation errors.
9650 @opindex fms-extensions
9651 Unless @option{-fms-extensions} is used, the unnamed field must be a
9652 structure or union definition without a tag (for example, @samp{struct
9653 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
9654 also be a definition with a tag such as @samp{struct foo @{ int a;
9655 @};}, a reference to a previously defined structure or union such as
9656 @samp{struct foo;}, or a reference to a @code{typedef} name for a
9657 previously defined structure or union type.
9660 @section Thread-Local Storage
9661 @cindex Thread-Local Storage
9662 @cindex @acronym{TLS}
9665 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
9666 are allocated such that there is one instance of the variable per extant
9667 thread. The run-time model GCC uses to implement this originates
9668 in the IA-64 processor-specific ABI, but has since been migrated
9669 to other processors as well. It requires significant support from
9670 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
9671 system libraries (@file{libc.so} and @file{libpthread.so}), so it
9672 is not available everywhere.
9674 At the user level, the extension is visible with a new storage
9675 class keyword: @code{__thread}. For example:
9679 extern __thread struct state s;
9680 static __thread char *p;
9683 The @code{__thread} specifier may be used alone, with the @code{extern}
9684 or @code{static} specifiers, but with no other storage class specifier.
9685 When used with @code{extern} or @code{static}, @code{__thread} must appear
9686 immediately after the other storage class specifier.
9688 The @code{__thread} specifier may be applied to any global, file-scoped
9689 static, function-scoped static, or static data member of a class. It may
9690 not be applied to block-scoped automatic or non-static data member.
9692 When the address-of operator is applied to a thread-local variable, it is
9693 evaluated at run-time and returns the address of the current thread's
9694 instance of that variable. An address so obtained may be used by any
9695 thread. When a thread terminates, any pointers to thread-local variables
9696 in that thread become invalid.
9698 No static initialization may refer to the address of a thread-local variable.
9700 In C++, if an initializer is present for a thread-local variable, it must
9701 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
9704 See @uref{http://people.redhat.com/drepper/tls.pdf,
9705 ELF Handling For Thread-Local Storage} for a detailed explanation of
9706 the four thread-local storage addressing models, and how the run-time
9707 is expected to function.
9710 * C99 Thread-Local Edits::
9711 * C++98 Thread-Local Edits::
9714 @node C99 Thread-Local Edits
9715 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
9717 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
9718 that document the exact semantics of the language extension.
9722 @cite{5.1.2 Execution environments}
9724 Add new text after paragraph 1
9727 Within either execution environment, a @dfn{thread} is a flow of
9728 control within a program. It is implementation defined whether
9729 or not there may be more than one thread associated with a program.
9730 It is implementation defined how threads beyond the first are
9731 created, the name and type of the function called at thread
9732 startup, and how threads may be terminated. However, objects
9733 with thread storage duration shall be initialized before thread
9738 @cite{6.2.4 Storage durations of objects}
9740 Add new text before paragraph 3
9743 An object whose identifier is declared with the storage-class
9744 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
9745 Its lifetime is the entire execution of the thread, and its
9746 stored value is initialized only once, prior to thread startup.
9750 @cite{6.4.1 Keywords}
9752 Add @code{__thread}.
9755 @cite{6.7.1 Storage-class specifiers}
9757 Add @code{__thread} to the list of storage class specifiers in
9760 Change paragraph 2 to
9763 With the exception of @code{__thread}, at most one storage-class
9764 specifier may be given [@dots{}]. The @code{__thread} specifier may
9765 be used alone, or immediately following @code{extern} or
9769 Add new text after paragraph 6
9772 The declaration of an identifier for a variable that has
9773 block scope that specifies @code{__thread} shall also
9774 specify either @code{extern} or @code{static}.
9776 The @code{__thread} specifier shall be used only with
9781 @node C++98 Thread-Local Edits
9782 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
9784 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
9785 that document the exact semantics of the language extension.
9789 @b{[intro.execution]}
9791 New text after paragraph 4
9794 A @dfn{thread} is a flow of control within the abstract machine.
9795 It is implementation defined whether or not there may be more than
9799 New text after paragraph 7
9802 It is unspecified whether additional action must be taken to
9803 ensure when and whether side effects are visible to other threads.
9809 Add @code{__thread}.
9812 @b{[basic.start.main]}
9814 Add after paragraph 5
9817 The thread that begins execution at the @code{main} function is called
9818 the @dfn{main thread}. It is implementation defined how functions
9819 beginning threads other than the main thread are designated or typed.
9820 A function so designated, as well as the @code{main} function, is called
9821 a @dfn{thread startup function}. It is implementation defined what
9822 happens if a thread startup function returns. It is implementation
9823 defined what happens to other threads when any thread calls @code{exit}.
9827 @b{[basic.start.init]}
9829 Add after paragraph 4
9832 The storage for an object of thread storage duration shall be
9833 statically initialized before the first statement of the thread startup
9834 function. An object of thread storage duration shall not require
9835 dynamic initialization.
9839 @b{[basic.start.term]}
9841 Add after paragraph 3
9844 The type of an object with thread storage duration shall not have a
9845 non-trivial destructor, nor shall it be an array type whose elements
9846 (directly or indirectly) have non-trivial destructors.
9852 Add ``thread storage duration'' to the list in paragraph 1.
9857 Thread, static, and automatic storage durations are associated with
9858 objects introduced by declarations [@dots{}].
9861 Add @code{__thread} to the list of specifiers in paragraph 3.
9864 @b{[basic.stc.thread]}
9866 New section before @b{[basic.stc.static]}
9869 The keyword @code{__thread} applied to a non-local object gives the
9870 object thread storage duration.
9872 A local variable or class data member declared both @code{static}
9873 and @code{__thread} gives the variable or member thread storage
9878 @b{[basic.stc.static]}
9883 All objects which have neither thread storage duration, dynamic
9884 storage duration nor are local [@dots{}].
9890 Add @code{__thread} to the list in paragraph 1.
9895 With the exception of @code{__thread}, at most one
9896 @var{storage-class-specifier} shall appear in a given
9897 @var{decl-specifier-seq}. The @code{__thread} specifier may
9898 be used alone, or immediately following the @code{extern} or
9899 @code{static} specifiers. [@dots{}]
9902 Add after paragraph 5
9905 The @code{__thread} specifier can be applied only to the names of objects
9906 and to anonymous unions.
9912 Add after paragraph 6
9915 Non-@code{static} members shall not be @code{__thread}.
9919 @node C++ Extensions
9920 @chapter Extensions to the C++ Language
9921 @cindex extensions, C++ language
9922 @cindex C++ language extensions
9924 The GNU compiler provides these extensions to the C++ language (and you
9925 can also use most of the C language extensions in your C++ programs). If you
9926 want to write code that checks whether these features are available, you can
9927 test for the GNU compiler the same way as for C programs: check for a
9928 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
9929 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
9930 Predefined Macros,cpp,The GNU C Preprocessor}).
9933 * Volatiles:: What constitutes an access to a volatile object.
9934 * Restricted Pointers:: C99 restricted pointers and references.
9935 * Vague Linkage:: Where G++ puts inlines, vtables and such.
9936 * C++ Interface:: You can use a single C++ header file for both
9937 declarations and definitions.
9938 * Template Instantiation:: Methods for ensuring that exactly one copy of
9939 each needed template instantiation is emitted.
9940 * Bound member functions:: You can extract a function pointer to the
9941 method denoted by a @samp{->*} or @samp{.*} expression.
9942 * C++ Attributes:: Variable, function, and type attributes for C++ only.
9943 * Namespace Association:: Strong using-directives for namespace association.
9944 * Java Exceptions:: Tweaking exception handling to work with Java.
9945 * Deprecated Features:: Things will disappear from g++.
9946 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
9950 @section When is a Volatile Object Accessed?
9951 @cindex accessing volatiles
9952 @cindex volatile read
9953 @cindex volatile write
9954 @cindex volatile access
9956 Both the C and C++ standard have the concept of volatile objects. These
9957 are normally accessed by pointers and used for accessing hardware. The
9958 standards encourage compilers to refrain from optimizations
9959 concerning accesses to volatile objects that it might perform on
9960 non-volatile objects. The C standard leaves it implementation defined
9961 as to what constitutes a volatile access. The C++ standard omits to
9962 specify this, except to say that C++ should behave in a similar manner
9963 to C with respect to volatiles, where possible. The minimum either
9964 standard specifies is that at a sequence point all previous accesses to
9965 volatile objects have stabilized and no subsequent accesses have
9966 occurred. Thus an implementation is free to reorder and combine
9967 volatile accesses which occur between sequence points, but cannot do so
9968 for accesses across a sequence point. The use of volatiles does not
9969 allow you to violate the restriction on updating objects multiple times
9970 within a sequence point.
9972 In most expressions, it is intuitively obvious what is a read and what is
9973 a write. For instance
9976 volatile int *dst = @var{somevalue};
9977 volatile int *src = @var{someothervalue};
9982 will cause a read of the volatile object pointed to by @var{src} and stores the
9983 value into the volatile object pointed to by @var{dst}. There is no
9984 guarantee that these reads and writes are atomic, especially for objects
9985 larger than @code{int}.
9987 Less obvious expressions are where something which looks like an access
9988 is used in a void context. An example would be,
9991 volatile int *src = @var{somevalue};
9995 With C, such expressions are rvalues, and as rvalues cause a read of
9996 the object, GCC interprets this as a read of the volatile being pointed
9997 to. The C++ standard specifies that such expressions do not undergo
9998 lvalue to rvalue conversion, and that the type of the dereferenced
9999 object may be incomplete. The C++ standard does not specify explicitly
10000 that it is this lvalue to rvalue conversion which is responsible for
10001 causing an access. However, there is reason to believe that it is,
10002 because otherwise certain simple expressions become undefined. However,
10003 because it would surprise most programmers, G++ treats dereferencing a
10004 pointer to volatile object of complete type in a void context as a read
10005 of the object. When the object has incomplete type, G++ issues a
10010 struct T @{int m;@};
10011 volatile S *ptr1 = @var{somevalue};
10012 volatile T *ptr2 = @var{somevalue};
10017 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
10018 causes a read of the object pointed to. If you wish to force an error on
10019 the first case, you must force a conversion to rvalue with, for instance
10020 a static cast, @code{static_cast<S>(*ptr1)}.
10022 When using a reference to volatile, G++ does not treat equivalent
10023 expressions as accesses to volatiles, but instead issues a warning that
10024 no volatile is accessed. The rationale for this is that otherwise it
10025 becomes difficult to determine where volatile access occur, and not
10026 possible to ignore the return value from functions returning volatile
10027 references. Again, if you wish to force a read, cast the reference to
10030 @node Restricted Pointers
10031 @section Restricting Pointer Aliasing
10032 @cindex restricted pointers
10033 @cindex restricted references
10034 @cindex restricted this pointer
10036 As with the C front end, G++ understands the C99 feature of restricted pointers,
10037 specified with the @code{__restrict__}, or @code{__restrict} type
10038 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
10039 language flag, @code{restrict} is not a keyword in C++.
10041 In addition to allowing restricted pointers, you can specify restricted
10042 references, which indicate that the reference is not aliased in the local
10046 void fn (int *__restrict__ rptr, int &__restrict__ rref)
10053 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
10054 @var{rref} refers to a (different) unaliased integer.
10056 You may also specify whether a member function's @var{this} pointer is
10057 unaliased by using @code{__restrict__} as a member function qualifier.
10060 void T::fn () __restrict__
10067 Within the body of @code{T::fn}, @var{this} will have the effective
10068 definition @code{T *__restrict__ const this}. Notice that the
10069 interpretation of a @code{__restrict__} member function qualifier is
10070 different to that of @code{const} or @code{volatile} qualifier, in that it
10071 is applied to the pointer rather than the object. This is consistent with
10072 other compilers which implement restricted pointers.
10074 As with all outermost parameter qualifiers, @code{__restrict__} is
10075 ignored in function definition matching. This means you only need to
10076 specify @code{__restrict__} in a function definition, rather than
10077 in a function prototype as well.
10079 @node Vague Linkage
10080 @section Vague Linkage
10081 @cindex vague linkage
10083 There are several constructs in C++ which require space in the object
10084 file but are not clearly tied to a single translation unit. We say that
10085 these constructs have ``vague linkage''. Typically such constructs are
10086 emitted wherever they are needed, though sometimes we can be more
10090 @item Inline Functions
10091 Inline functions are typically defined in a header file which can be
10092 included in many different compilations. Hopefully they can usually be
10093 inlined, but sometimes an out-of-line copy is necessary, if the address
10094 of the function is taken or if inlining fails. In general, we emit an
10095 out-of-line copy in all translation units where one is needed. As an
10096 exception, we only emit inline virtual functions with the vtable, since
10097 it will always require a copy.
10099 Local static variables and string constants used in an inline function
10100 are also considered to have vague linkage, since they must be shared
10101 between all inlined and out-of-line instances of the function.
10105 C++ virtual functions are implemented in most compilers using a lookup
10106 table, known as a vtable. The vtable contains pointers to the virtual
10107 functions provided by a class, and each object of the class contains a
10108 pointer to its vtable (or vtables, in some multiple-inheritance
10109 situations). If the class declares any non-inline, non-pure virtual
10110 functions, the first one is chosen as the ``key method'' for the class,
10111 and the vtable is only emitted in the translation unit where the key
10114 @emph{Note:} If the chosen key method is later defined as inline, the
10115 vtable will still be emitted in every translation unit which defines it.
10116 Make sure that any inline virtuals are declared inline in the class
10117 body, even if they are not defined there.
10119 @item type_info objects
10122 C++ requires information about types to be written out in order to
10123 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
10124 For polymorphic classes (classes with virtual functions), the type_info
10125 object is written out along with the vtable so that @samp{dynamic_cast}
10126 can determine the dynamic type of a class object at runtime. For all
10127 other types, we write out the type_info object when it is used: when
10128 applying @samp{typeid} to an expression, throwing an object, or
10129 referring to a type in a catch clause or exception specification.
10131 @item Template Instantiations
10132 Most everything in this section also applies to template instantiations,
10133 but there are other options as well.
10134 @xref{Template Instantiation,,Where's the Template?}.
10138 When used with GNU ld version 2.8 or later on an ELF system such as
10139 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
10140 these constructs will be discarded at link time. This is known as
10143 On targets that don't support COMDAT, but do support weak symbols, GCC
10144 will use them. This way one copy will override all the others, but
10145 the unused copies will still take up space in the executable.
10147 For targets which do not support either COMDAT or weak symbols,
10148 most entities with vague linkage will be emitted as local symbols to
10149 avoid duplicate definition errors from the linker. This will not happen
10150 for local statics in inlines, however, as having multiple copies will
10151 almost certainly break things.
10153 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
10154 another way to control placement of these constructs.
10156 @node C++ Interface
10157 @section #pragma interface and implementation
10159 @cindex interface and implementation headers, C++
10160 @cindex C++ interface and implementation headers
10161 @cindex pragmas, interface and implementation
10163 @code{#pragma interface} and @code{#pragma implementation} provide the
10164 user with a way of explicitly directing the compiler to emit entities
10165 with vague linkage (and debugging information) in a particular
10168 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
10169 most cases, because of COMDAT support and the ``key method'' heuristic
10170 mentioned in @ref{Vague Linkage}. Using them can actually cause your
10171 program to grow due to unnecessary out-of-line copies of inline
10172 functions. Currently (3.4) the only benefit of these
10173 @code{#pragma}s is reduced duplication of debugging information, and
10174 that should be addressed soon on DWARF 2 targets with the use of
10178 @item #pragma interface
10179 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
10180 @kindex #pragma interface
10181 Use this directive in @emph{header files} that define object classes, to save
10182 space in most of the object files that use those classes. Normally,
10183 local copies of certain information (backup copies of inline member
10184 functions, debugging information, and the internal tables that implement
10185 virtual functions) must be kept in each object file that includes class
10186 definitions. You can use this pragma to avoid such duplication. When a
10187 header file containing @samp{#pragma interface} is included in a
10188 compilation, this auxiliary information will not be generated (unless
10189 the main input source file itself uses @samp{#pragma implementation}).
10190 Instead, the object files will contain references to be resolved at link
10193 The second form of this directive is useful for the case where you have
10194 multiple headers with the same name in different directories. If you
10195 use this form, you must specify the same string to @samp{#pragma
10198 @item #pragma implementation
10199 @itemx #pragma implementation "@var{objects}.h"
10200 @kindex #pragma implementation
10201 Use this pragma in a @emph{main input file}, when you want full output from
10202 included header files to be generated (and made globally visible). The
10203 included header file, in turn, should use @samp{#pragma interface}.
10204 Backup copies of inline member functions, debugging information, and the
10205 internal tables used to implement virtual functions are all generated in
10206 implementation files.
10208 @cindex implied @code{#pragma implementation}
10209 @cindex @code{#pragma implementation}, implied
10210 @cindex naming convention, implementation headers
10211 If you use @samp{#pragma implementation} with no argument, it applies to
10212 an include file with the same basename@footnote{A file's @dfn{basename}
10213 was the name stripped of all leading path information and of trailing
10214 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
10215 file. For example, in @file{allclass.cc}, giving just
10216 @samp{#pragma implementation}
10217 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
10219 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
10220 an implementation file whenever you would include it from
10221 @file{allclass.cc} even if you never specified @samp{#pragma
10222 implementation}. This was deemed to be more trouble than it was worth,
10223 however, and disabled.
10225 Use the string argument if you want a single implementation file to
10226 include code from multiple header files. (You must also use
10227 @samp{#include} to include the header file; @samp{#pragma
10228 implementation} only specifies how to use the file---it doesn't actually
10231 There is no way to split up the contents of a single header file into
10232 multiple implementation files.
10235 @cindex inlining and C++ pragmas
10236 @cindex C++ pragmas, effect on inlining
10237 @cindex pragmas in C++, effect on inlining
10238 @samp{#pragma implementation} and @samp{#pragma interface} also have an
10239 effect on function inlining.
10241 If you define a class in a header file marked with @samp{#pragma
10242 interface}, the effect on an inline function defined in that class is
10243 similar to an explicit @code{extern} declaration---the compiler emits
10244 no code at all to define an independent version of the function. Its
10245 definition is used only for inlining with its callers.
10247 @opindex fno-implement-inlines
10248 Conversely, when you include the same header file in a main source file
10249 that declares it as @samp{#pragma implementation}, the compiler emits
10250 code for the function itself; this defines a version of the function
10251 that can be found via pointers (or by callers compiled without
10252 inlining). If all calls to the function can be inlined, you can avoid
10253 emitting the function by compiling with @option{-fno-implement-inlines}.
10254 If any calls were not inlined, you will get linker errors.
10256 @node Template Instantiation
10257 @section Where's the Template?
10258 @cindex template instantiation
10260 C++ templates are the first language feature to require more
10261 intelligence from the environment than one usually finds on a UNIX
10262 system. Somehow the compiler and linker have to make sure that each
10263 template instance occurs exactly once in the executable if it is needed,
10264 and not at all otherwise. There are two basic approaches to this
10265 problem, which are referred to as the Borland model and the Cfront model.
10268 @item Borland model
10269 Borland C++ solved the template instantiation problem by adding the code
10270 equivalent of common blocks to their linker; the compiler emits template
10271 instances in each translation unit that uses them, and the linker
10272 collapses them together. The advantage of this model is that the linker
10273 only has to consider the object files themselves; there is no external
10274 complexity to worry about. This disadvantage is that compilation time
10275 is increased because the template code is being compiled repeatedly.
10276 Code written for this model tends to include definitions of all
10277 templates in the header file, since they must be seen to be
10281 The AT&T C++ translator, Cfront, solved the template instantiation
10282 problem by creating the notion of a template repository, an
10283 automatically maintained place where template instances are stored. A
10284 more modern version of the repository works as follows: As individual
10285 object files are built, the compiler places any template definitions and
10286 instantiations encountered in the repository. At link time, the link
10287 wrapper adds in the objects in the repository and compiles any needed
10288 instances that were not previously emitted. The advantages of this
10289 model are more optimal compilation speed and the ability to use the
10290 system linker; to implement the Borland model a compiler vendor also
10291 needs to replace the linker. The disadvantages are vastly increased
10292 complexity, and thus potential for error; for some code this can be
10293 just as transparent, but in practice it can been very difficult to build
10294 multiple programs in one directory and one program in multiple
10295 directories. Code written for this model tends to separate definitions
10296 of non-inline member templates into a separate file, which should be
10297 compiled separately.
10300 When used with GNU ld version 2.8 or later on an ELF system such as
10301 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
10302 Borland model. On other systems, G++ implements neither automatic
10305 A future version of G++ will support a hybrid model whereby the compiler
10306 will emit any instantiations for which the template definition is
10307 included in the compile, and store template definitions and
10308 instantiation context information into the object file for the rest.
10309 The link wrapper will extract that information as necessary and invoke
10310 the compiler to produce the remaining instantiations. The linker will
10311 then combine duplicate instantiations.
10313 In the mean time, you have the following options for dealing with
10314 template instantiations:
10319 Compile your template-using code with @option{-frepo}. The compiler will
10320 generate files with the extension @samp{.rpo} listing all of the
10321 template instantiations used in the corresponding object files which
10322 could be instantiated there; the link wrapper, @samp{collect2}, will
10323 then update the @samp{.rpo} files to tell the compiler where to place
10324 those instantiations and rebuild any affected object files. The
10325 link-time overhead is negligible after the first pass, as the compiler
10326 will continue to place the instantiations in the same files.
10328 This is your best option for application code written for the Borland
10329 model, as it will just work. Code written for the Cfront model will
10330 need to be modified so that the template definitions are available at
10331 one or more points of instantiation; usually this is as simple as adding
10332 @code{#include <tmethods.cc>} to the end of each template header.
10334 For library code, if you want the library to provide all of the template
10335 instantiations it needs, just try to link all of its object files
10336 together; the link will fail, but cause the instantiations to be
10337 generated as a side effect. Be warned, however, that this may cause
10338 conflicts if multiple libraries try to provide the same instantiations.
10339 For greater control, use explicit instantiation as described in the next
10343 @opindex fno-implicit-templates
10344 Compile your code with @option{-fno-implicit-templates} to disable the
10345 implicit generation of template instances, and explicitly instantiate
10346 all the ones you use. This approach requires more knowledge of exactly
10347 which instances you need than do the others, but it's less
10348 mysterious and allows greater control. You can scatter the explicit
10349 instantiations throughout your program, perhaps putting them in the
10350 translation units where the instances are used or the translation units
10351 that define the templates themselves; you can put all of the explicit
10352 instantiations you need into one big file; or you can create small files
10359 template class Foo<int>;
10360 template ostream& operator <<
10361 (ostream&, const Foo<int>&);
10364 for each of the instances you need, and create a template instantiation
10365 library from those.
10367 If you are using Cfront-model code, you can probably get away with not
10368 using @option{-fno-implicit-templates} when compiling files that don't
10369 @samp{#include} the member template definitions.
10371 If you use one big file to do the instantiations, you may want to
10372 compile it without @option{-fno-implicit-templates} so you get all of the
10373 instances required by your explicit instantiations (but not by any
10374 other files) without having to specify them as well.
10376 G++ has extended the template instantiation syntax given in the ISO
10377 standard to allow forward declaration of explicit instantiations
10378 (with @code{extern}), instantiation of the compiler support data for a
10379 template class (i.e.@: the vtable) without instantiating any of its
10380 members (with @code{inline}), and instantiation of only the static data
10381 members of a template class, without the support data or member
10382 functions (with (@code{static}):
10385 extern template int max (int, int);
10386 inline template class Foo<int>;
10387 static template class Foo<int>;
10391 Do nothing. Pretend G++ does implement automatic instantiation
10392 management. Code written for the Borland model will work fine, but
10393 each translation unit will contain instances of each of the templates it
10394 uses. In a large program, this can lead to an unacceptable amount of code
10398 @node Bound member functions
10399 @section Extracting the function pointer from a bound pointer to member function
10401 @cindex pointer to member function
10402 @cindex bound pointer to member function
10404 In C++, pointer to member functions (PMFs) are implemented using a wide
10405 pointer of sorts to handle all the possible call mechanisms; the PMF
10406 needs to store information about how to adjust the @samp{this} pointer,
10407 and if the function pointed to is virtual, where to find the vtable, and
10408 where in the vtable to look for the member function. If you are using
10409 PMFs in an inner loop, you should really reconsider that decision. If
10410 that is not an option, you can extract the pointer to the function that
10411 would be called for a given object/PMF pair and call it directly inside
10412 the inner loop, to save a bit of time.
10414 Note that you will still be paying the penalty for the call through a
10415 function pointer; on most modern architectures, such a call defeats the
10416 branch prediction features of the CPU@. This is also true of normal
10417 virtual function calls.
10419 The syntax for this extension is
10423 extern int (A::*fp)();
10424 typedef int (*fptr)(A *);
10426 fptr p = (fptr)(a.*fp);
10429 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
10430 no object is needed to obtain the address of the function. They can be
10431 converted to function pointers directly:
10434 fptr p1 = (fptr)(&A::foo);
10437 @opindex Wno-pmf-conversions
10438 You must specify @option{-Wno-pmf-conversions} to use this extension.
10440 @node C++ Attributes
10441 @section C++-Specific Variable, Function, and Type Attributes
10443 Some attributes only make sense for C++ programs.
10446 @item init_priority (@var{priority})
10447 @cindex init_priority attribute
10450 In Standard C++, objects defined at namespace scope are guaranteed to be
10451 initialized in an order in strict accordance with that of their definitions
10452 @emph{in a given translation unit}. No guarantee is made for initializations
10453 across translation units. However, GNU C++ allows users to control the
10454 order of initialization of objects defined at namespace scope with the
10455 @code{init_priority} attribute by specifying a relative @var{priority},
10456 a constant integral expression currently bounded between 101 and 65535
10457 inclusive. Lower numbers indicate a higher priority.
10459 In the following example, @code{A} would normally be created before
10460 @code{B}, but the @code{init_priority} attribute has reversed that order:
10463 Some_Class A __attribute__ ((init_priority (2000)));
10464 Some_Class B __attribute__ ((init_priority (543)));
10468 Note that the particular values of @var{priority} do not matter; only their
10471 @item java_interface
10472 @cindex java_interface attribute
10474 This type attribute informs C++ that the class is a Java interface. It may
10475 only be applied to classes declared within an @code{extern "Java"} block.
10476 Calls to methods declared in this interface will be dispatched using GCJ's
10477 interface table mechanism, instead of regular virtual table dispatch.
10481 See also @xref{Namespace Association}.
10483 @node Namespace Association
10484 @section Namespace Association
10486 @strong{Caution:} The semantics of this extension are not fully
10487 defined. Users should refrain from using this extension as its
10488 semantics may change subtly over time. It is possible that this
10489 extension will be removed in future versions of G++.
10491 A using-directive with @code{__attribute ((strong))} is stronger
10492 than a normal using-directive in two ways:
10496 Templates from the used namespace can be specialized and explicitly
10497 instantiated as though they were members of the using namespace.
10500 The using namespace is considered an associated namespace of all
10501 templates in the used namespace for purposes of argument-dependent
10505 The used namespace must be nested within the using namespace so that
10506 normal unqualified lookup works properly.
10508 This is useful for composing a namespace transparently from
10509 implementation namespaces. For example:
10514 template <class T> struct A @{ @};
10516 using namespace debug __attribute ((__strong__));
10517 template <> struct A<int> @{ @}; // @r{ok to specialize}
10519 template <class T> void f (A<T>);
10524 f (std::A<float>()); // @r{lookup finds} std::f
10529 @node Java Exceptions
10530 @section Java Exceptions
10532 The Java language uses a slightly different exception handling model
10533 from C++. Normally, GNU C++ will automatically detect when you are
10534 writing C++ code that uses Java exceptions, and handle them
10535 appropriately. However, if C++ code only needs to execute destructors
10536 when Java exceptions are thrown through it, GCC will guess incorrectly.
10537 Sample problematic code is:
10540 struct S @{ ~S(); @};
10541 extern void bar(); // @r{is written in Java, and may throw exceptions}
10550 The usual effect of an incorrect guess is a link failure, complaining of
10551 a missing routine called @samp{__gxx_personality_v0}.
10553 You can inform the compiler that Java exceptions are to be used in a
10554 translation unit, irrespective of what it might think, by writing
10555 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
10556 @samp{#pragma} must appear before any functions that throw or catch
10557 exceptions, or run destructors when exceptions are thrown through them.
10559 You cannot mix Java and C++ exceptions in the same translation unit. It
10560 is believed to be safe to throw a C++ exception from one file through
10561 another file compiled for the Java exception model, or vice versa, but
10562 there may be bugs in this area.
10564 @node Deprecated Features
10565 @section Deprecated Features
10567 In the past, the GNU C++ compiler was extended to experiment with new
10568 features, at a time when the C++ language was still evolving. Now that
10569 the C++ standard is complete, some of those features are superseded by
10570 superior alternatives. Using the old features might cause a warning in
10571 some cases that the feature will be dropped in the future. In other
10572 cases, the feature might be gone already.
10574 While the list below is not exhaustive, it documents some of the options
10575 that are now deprecated:
10578 @item -fexternal-templates
10579 @itemx -falt-external-templates
10580 These are two of the many ways for G++ to implement template
10581 instantiation. @xref{Template Instantiation}. The C++ standard clearly
10582 defines how template definitions have to be organized across
10583 implementation units. G++ has an implicit instantiation mechanism that
10584 should work just fine for standard-conforming code.
10586 @item -fstrict-prototype
10587 @itemx -fno-strict-prototype
10588 Previously it was possible to use an empty prototype parameter list to
10589 indicate an unspecified number of parameters (like C), rather than no
10590 parameters, as C++ demands. This feature has been removed, except where
10591 it is required for backwards compatibility @xref{Backwards Compatibility}.
10594 G++ allows a virtual function returning @samp{void *} to be overridden
10595 by one returning a different pointer type. This extension to the
10596 covariant return type rules is now deprecated and will be removed from a
10599 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
10600 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
10601 and will be removed in a future version. Code using these operators
10602 should be modified to use @code{std::min} and @code{std::max} instead.
10604 The named return value extension has been deprecated, and is now
10607 The use of initializer lists with new expressions has been deprecated,
10608 and is now removed from G++.
10610 Floating and complex non-type template parameters have been deprecated,
10611 and are now removed from G++.
10613 The implicit typename extension has been deprecated and is now
10616 The use of default arguments in function pointers, function typedefs and
10617 and other places where they are not permitted by the standard is
10618 deprecated and will be removed from a future version of G++.
10620 G++ allows floating-point literals to appear in integral constant expressions,
10621 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
10622 This extension is deprecated and will be removed from a future version.
10624 G++ allows static data members of const floating-point type to be declared
10625 with an initializer in a class definition. The standard only allows
10626 initializers for static members of const integral types and const
10627 enumeration types so this extension has been deprecated and will be removed
10628 from a future version.
10630 @node Backwards Compatibility
10631 @section Backwards Compatibility
10632 @cindex Backwards Compatibility
10633 @cindex ARM [Annotated C++ Reference Manual]
10635 Now that there is a definitive ISO standard C++, G++ has a specification
10636 to adhere to. The C++ language evolved over time, and features that
10637 used to be acceptable in previous drafts of the standard, such as the ARM
10638 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
10639 compilation of C++ written to such drafts, G++ contains some backwards
10640 compatibilities. @emph{All such backwards compatibility features are
10641 liable to disappear in future versions of G++.} They should be considered
10642 deprecated @xref{Deprecated Features}.
10646 If a variable is declared at for scope, it used to remain in scope until
10647 the end of the scope which contained the for statement (rather than just
10648 within the for scope). G++ retains this, but issues a warning, if such a
10649 variable is accessed outside the for scope.
10651 @item Implicit C language
10652 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
10653 scope to set the language. On such systems, all header files are
10654 implicitly scoped inside a C language scope. Also, an empty prototype
10655 @code{()} will be treated as an unspecified number of arguments, rather
10656 than no arguments, as C++ demands.