1 @c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001,2002,2003,2004,2005
2 @c Free Software Foundation, Inc.
3 @c This is part of the GCC manual.
4 @c For copying conditions, see the file gcc.texi.
7 @chapter Extensions to the C Language Family
8 @cindex extensions, C language
9 @cindex C language extensions
12 GNU C provides several language features not found in ISO standard C@.
13 (The @option{-pedantic} option directs GCC to print a warning message if
14 any of these features is used.) To test for the availability of these
15 features in conditional compilation, check for a predefined macro
16 @code{__GNUC__}, which is always defined under GCC@.
18 These extensions are available in C and Objective-C@. Most of them are
19 also available in C++. @xref{C++ Extensions,,Extensions to the
20 C++ Language}, for extensions that apply @emph{only} to C++.
22 Some features that are in ISO C99 but not C89 or C++ are also, as
23 extensions, accepted by GCC in C89 mode and in C++.
26 * Statement Exprs:: Putting statements and declarations inside expressions.
27 * Local Labels:: Labels local to a block.
28 * Labels as Values:: Getting pointers to labels, and computed gotos.
29 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
30 * Constructing Calls:: Dispatching a call to another function.
31 * Typeof:: @code{typeof}: referring to the type of an expression.
32 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
33 * Long Long:: Double-word integers---@code{long long int}.
34 * Complex:: Data types for complex numbers.
35 * Hex Floats:: Hexadecimal floating-point constants.
36 * Zero Length:: Zero-length arrays.
37 * Variable Length:: Arrays whose length is computed at run time.
38 * Empty Structures:: Structures with no members.
39 * Variadic Macros:: Macros with a variable number of arguments.
40 * Escaped Newlines:: Slightly looser rules for escaped newlines.
41 * Subscripting:: Any array can be subscripted, even if not an lvalue.
42 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
43 * Initializers:: Non-constant initializers.
44 * Compound Literals:: Compound literals give structures, unions
46 * Designated Inits:: Labeling elements of initializers.
47 * Cast to Union:: Casting to union type from any member of the union.
48 * Case Ranges:: `case 1 ... 9' and such.
49 * Mixed Declarations:: Mixing declarations and code.
50 * Function Attributes:: Declaring that functions have no side effects,
51 or that they can never return.
52 * Attribute Syntax:: Formal syntax for attributes.
53 * Function Prototypes:: Prototype declarations and old-style definitions.
54 * C++ Comments:: C++ comments are recognized.
55 * Dollar Signs:: Dollar sign is allowed in identifiers.
56 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
57 * Variable Attributes:: Specifying attributes of variables.
58 * Type Attributes:: Specifying attributes of types.
59 * Alignment:: Inquiring about the alignment of a type or variable.
60 * Inline:: Defining inline functions (as fast as macros).
61 * Extended Asm:: Assembler instructions with C expressions as operands.
62 (With them you can define ``built-in'' functions.)
63 * Constraints:: Constraints for asm operands
64 * Asm Labels:: Specifying the assembler name to use for a C symbol.
65 * Explicit Reg Vars:: Defining variables residing in specified registers.
66 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
67 * Incomplete Enums:: @code{enum foo;}, with details to follow.
68 * Function Names:: Printable strings which are the name of the current
70 * Return Address:: Getting the return or frame address of a function.
71 * Vector Extensions:: Using vector instructions through built-in functions.
72 * Offsetof:: Special syntax for implementing @code{offsetof}.
73 * Atomic Builtins:: Built-in functions for atomic memory access.
74 * Object Size Checking:: Built-in functions for limited buffer overflow
76 * Other Builtins:: Other built-in functions.
77 * Target Builtins:: Built-in functions specific to particular targets.
78 * Target Format Checks:: Format checks specific to particular targets.
79 * Pragmas:: Pragmas accepted by GCC.
80 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
81 * Thread-Local:: Per-thread variables.
85 @section Statements and Declarations in Expressions
86 @cindex statements inside expressions
87 @cindex declarations inside expressions
88 @cindex expressions containing statements
89 @cindex macros, statements in expressions
91 @c the above section title wrapped and causes an underfull hbox.. i
92 @c changed it from "within" to "in". --mew 4feb93
93 A compound statement enclosed in parentheses may appear as an expression
94 in GNU C@. This allows you to use loops, switches, and local variables
97 Recall that a compound statement is a sequence of statements surrounded
98 by braces; in this construct, parentheses go around the braces. For
102 (@{ int y = foo (); int z;
109 is a valid (though slightly more complex than necessary) expression
110 for the absolute value of @code{foo ()}.
112 The last thing in the compound statement should be an expression
113 followed by a semicolon; the value of this subexpression serves as the
114 value of the entire construct. (If you use some other kind of statement
115 last within the braces, the construct has type @code{void}, and thus
116 effectively no value.)
118 This feature is especially useful in making macro definitions ``safe'' (so
119 that they evaluate each operand exactly once). For example, the
120 ``maximum'' function is commonly defined as a macro in standard C as
124 #define max(a,b) ((a) > (b) ? (a) : (b))
128 @cindex side effects, macro argument
129 But this definition computes either @var{a} or @var{b} twice, with bad
130 results if the operand has side effects. In GNU C, if you know the
131 type of the operands (here taken as @code{int}), you can define
132 the macro safely as follows:
135 #define maxint(a,b) \
136 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
139 Embedded statements are not allowed in constant expressions, such as
140 the value of an enumeration constant, the width of a bit-field, or
141 the initial value of a static variable.
143 If you don't know the type of the operand, you can still do this, but you
144 must use @code{typeof} (@pxref{Typeof}).
146 In G++, the result value of a statement expression undergoes array and
147 function pointer decay, and is returned by value to the enclosing
148 expression. For instance, if @code{A} is a class, then
157 will construct a temporary @code{A} object to hold the result of the
158 statement expression, and that will be used to invoke @code{Foo}.
159 Therefore the @code{this} pointer observed by @code{Foo} will not be the
162 Any temporaries created within a statement within a statement expression
163 will be destroyed at the statement's end. This makes statement
164 expressions inside macros slightly different from function calls. In
165 the latter case temporaries introduced during argument evaluation will
166 be destroyed at the end of the statement that includes the function
167 call. In the statement expression case they will be destroyed during
168 the statement expression. For instance,
171 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
172 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
182 will have different places where temporaries are destroyed. For the
183 @code{macro} case, the temporary @code{X} will be destroyed just after
184 the initialization of @code{b}. In the @code{function} case that
185 temporary will be destroyed when the function returns.
187 These considerations mean that it is probably a bad idea to use
188 statement-expressions of this form in header files that are designed to
189 work with C++. (Note that some versions of the GNU C Library contained
190 header files using statement-expression that lead to precisely this
193 Jumping into a statement expression with @code{goto} or using a
194 @code{switch} statement outside the statement expression with a
195 @code{case} or @code{default} label inside the statement expression is
196 not permitted. Jumping into a statement expression with a computed
197 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
198 Jumping out of a statement expression is permitted, but if the
199 statement expression is part of a larger expression then it is
200 unspecified which other subexpressions of that expression have been
201 evaluated except where the language definition requires certain
202 subexpressions to be evaluated before or after the statement
203 expression. In any case, as with a function call the evaluation of a
204 statement expression is not interleaved with the evaluation of other
205 parts of the containing expression. For example,
208 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
212 will call @code{foo} and @code{bar1} and will not call @code{baz} but
213 may or may not call @code{bar2}. If @code{bar2} is called, it will be
214 called after @code{foo} and before @code{bar1}
217 @section Locally Declared Labels
219 @cindex macros, local labels
221 GCC allows you to declare @dfn{local labels} in any nested block
222 scope. A local label is just like an ordinary label, but you can
223 only reference it (with a @code{goto} statement, or by taking its
224 address) within the block in which it was declared.
226 A local label declaration looks like this:
229 __label__ @var{label};
236 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
239 Local label declarations must come at the beginning of the block,
240 before any ordinary declarations or statements.
242 The label declaration defines the label @emph{name}, but does not define
243 the label itself. You must do this in the usual way, with
244 @code{@var{label}:}, within the statements of the statement expression.
246 The local label feature is useful for complex macros. If a macro
247 contains nested loops, a @code{goto} can be useful for breaking out of
248 them. However, an ordinary label whose scope is the whole function
249 cannot be used: if the macro can be expanded several times in one
250 function, the label will be multiply defined in that function. A
251 local label avoids this problem. For example:
254 #define SEARCH(value, array, target) \
257 typeof (target) _SEARCH_target = (target); \
258 typeof (*(array)) *_SEARCH_array = (array); \
261 for (i = 0; i < max; i++) \
262 for (j = 0; j < max; j++) \
263 if (_SEARCH_array[i][j] == _SEARCH_target) \
264 @{ (value) = i; goto found; @} \
270 This could also be written using a statement-expression:
273 #define SEARCH(array, target) \
276 typeof (target) _SEARCH_target = (target); \
277 typeof (*(array)) *_SEARCH_array = (array); \
280 for (i = 0; i < max; i++) \
281 for (j = 0; j < max; j++) \
282 if (_SEARCH_array[i][j] == _SEARCH_target) \
283 @{ value = i; goto found; @} \
290 Local label declarations also make the labels they declare visible to
291 nested functions, if there are any. @xref{Nested Functions}, for details.
293 @node Labels as Values
294 @section Labels as Values
295 @cindex labels as values
296 @cindex computed gotos
297 @cindex goto with computed label
298 @cindex address of a label
300 You can get the address of a label defined in the current function
301 (or a containing function) with the unary operator @samp{&&}. The
302 value has type @code{void *}. This value is a constant and can be used
303 wherever a constant of that type is valid. For example:
311 To use these values, you need to be able to jump to one. This is done
312 with the computed goto statement@footnote{The analogous feature in
313 Fortran is called an assigned goto, but that name seems inappropriate in
314 C, where one can do more than simply store label addresses in label
315 variables.}, @code{goto *@var{exp};}. For example,
322 Any expression of type @code{void *} is allowed.
324 One way of using these constants is in initializing a static array that
325 will serve as a jump table:
328 static void *array[] = @{ &&foo, &&bar, &&hack @};
331 Then you can select a label with indexing, like this:
338 Note that this does not check whether the subscript is in bounds---array
339 indexing in C never does that.
341 Such an array of label values serves a purpose much like that of the
342 @code{switch} statement. The @code{switch} statement is cleaner, so
343 use that rather than an array unless the problem does not fit a
344 @code{switch} statement very well.
346 Another use of label values is in an interpreter for threaded code.
347 The labels within the interpreter function can be stored in the
348 threaded code for super-fast dispatching.
350 You may not use this mechanism to jump to code in a different function.
351 If you do that, totally unpredictable things will happen. The best way to
352 avoid this is to store the label address only in automatic variables and
353 never pass it as an argument.
355 An alternate way to write the above example is
358 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
360 goto *(&&foo + array[i]);
364 This is more friendly to code living in shared libraries, as it reduces
365 the number of dynamic relocations that are needed, and by consequence,
366 allows the data to be read-only.
368 @node Nested Functions
369 @section Nested Functions
370 @cindex nested functions
371 @cindex downward funargs
374 A @dfn{nested function} is a function defined inside another function.
375 (Nested functions are not supported for GNU C++.) The nested function's
376 name is local to the block where it is defined. For example, here we
377 define a nested function named @code{square}, and call it twice:
381 foo (double a, double b)
383 double square (double z) @{ return z * z; @}
385 return square (a) + square (b);
390 The nested function can access all the variables of the containing
391 function that are visible at the point of its definition. This is
392 called @dfn{lexical scoping}. For example, here we show a nested
393 function which uses an inherited variable named @code{offset}:
397 bar (int *array, int offset, int size)
399 int access (int *array, int index)
400 @{ return array[index + offset]; @}
403 for (i = 0; i < size; i++)
404 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
409 Nested function definitions are permitted within functions in the places
410 where variable definitions are allowed; that is, in any block, mixed
411 with the other declarations and statements in the block.
413 It is possible to call the nested function from outside the scope of its
414 name by storing its address or passing the address to another function:
417 hack (int *array, int size)
419 void store (int index, int value)
420 @{ array[index] = value; @}
422 intermediate (store, size);
426 Here, the function @code{intermediate} receives the address of
427 @code{store} as an argument. If @code{intermediate} calls @code{store},
428 the arguments given to @code{store} are used to store into @code{array}.
429 But this technique works only so long as the containing function
430 (@code{hack}, in this example) does not exit.
432 If you try to call the nested function through its address after the
433 containing function has exited, all hell will break loose. If you try
434 to call it after a containing scope level has exited, and if it refers
435 to some of the variables that are no longer in scope, you may be lucky,
436 but it's not wise to take the risk. If, however, the nested function
437 does not refer to anything that has gone out of scope, you should be
440 GCC implements taking the address of a nested function using a technique
441 called @dfn{trampolines}. A paper describing them is available as
444 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
446 A nested function can jump to a label inherited from a containing
447 function, provided the label was explicitly declared in the containing
448 function (@pxref{Local Labels}). Such a jump returns instantly to the
449 containing function, exiting the nested function which did the
450 @code{goto} and any intermediate functions as well. Here is an example:
454 bar (int *array, int offset, int size)
457 int access (int *array, int index)
461 return array[index + offset];
465 for (i = 0; i < size; i++)
466 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
470 /* @r{Control comes here from @code{access}
471 if it detects an error.} */
478 A nested function always has no linkage. Declaring one with
479 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
480 before its definition, use @code{auto} (which is otherwise meaningless
481 for function declarations).
484 bar (int *array, int offset, int size)
487 auto int access (int *, int);
489 int access (int *array, int index)
493 return array[index + offset];
499 @node Constructing Calls
500 @section Constructing Function Calls
501 @cindex constructing calls
502 @cindex forwarding calls
504 Using the built-in functions described below, you can record
505 the arguments a function received, and call another function
506 with the same arguments, without knowing the number or types
509 You can also record the return value of that function call,
510 and later return that value, without knowing what data type
511 the function tried to return (as long as your caller expects
514 However, these built-in functions may interact badly with some
515 sophisticated features or other extensions of the language. It
516 is, therefore, not recommended to use them outside very simple
517 functions acting as mere forwarders for their arguments.
519 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
520 This built-in function returns a pointer to data
521 describing how to perform a call with the same arguments as were passed
522 to the current function.
524 The function saves the arg pointer register, structure value address,
525 and all registers that might be used to pass arguments to a function
526 into a block of memory allocated on the stack. Then it returns the
527 address of that block.
530 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
531 This built-in function invokes @var{function}
532 with a copy of the parameters described by @var{arguments}
535 The value of @var{arguments} should be the value returned by
536 @code{__builtin_apply_args}. The argument @var{size} specifies the size
537 of the stack argument data, in bytes.
539 This function returns a pointer to data describing
540 how to return whatever value was returned by @var{function}. The data
541 is saved in a block of memory allocated on the stack.
543 It is not always simple to compute the proper value for @var{size}. The
544 value is used by @code{__builtin_apply} to compute the amount of data
545 that should be pushed on the stack and copied from the incoming argument
549 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
550 This built-in function returns the value described by @var{result} from
551 the containing function. You should specify, for @var{result}, a value
552 returned by @code{__builtin_apply}.
556 @section Referring to a Type with @code{typeof}
559 @cindex macros, types of arguments
561 Another way to refer to the type of an expression is with @code{typeof}.
562 The syntax of using of this keyword looks like @code{sizeof}, but the
563 construct acts semantically like a type name defined with @code{typedef}.
565 There are two ways of writing the argument to @code{typeof}: with an
566 expression or with a type. Here is an example with an expression:
573 This assumes that @code{x} is an array of pointers to functions;
574 the type described is that of the values of the functions.
576 Here is an example with a typename as the argument:
583 Here the type described is that of pointers to @code{int}.
585 If you are writing a header file that must work when included in ISO C
586 programs, write @code{__typeof__} instead of @code{typeof}.
587 @xref{Alternate Keywords}.
589 A @code{typeof}-construct can be used anywhere a typedef name could be
590 used. For example, you can use it in a declaration, in a cast, or inside
591 of @code{sizeof} or @code{typeof}.
593 @code{typeof} is often useful in conjunction with the
594 statements-within-expressions feature. Here is how the two together can
595 be used to define a safe ``maximum'' macro that operates on any
596 arithmetic type and evaluates each of its arguments exactly once:
600 (@{ typeof (a) _a = (a); \
601 typeof (b) _b = (b); \
602 _a > _b ? _a : _b; @})
605 @cindex underscores in variables in macros
606 @cindex @samp{_} in variables in macros
607 @cindex local variables in macros
608 @cindex variables, local, in macros
609 @cindex macros, local variables in
611 The reason for using names that start with underscores for the local
612 variables is to avoid conflicts with variable names that occur within the
613 expressions that are substituted for @code{a} and @code{b}. Eventually we
614 hope to design a new form of declaration syntax that allows you to declare
615 variables whose scopes start only after their initializers; this will be a
616 more reliable way to prevent such conflicts.
619 Some more examples of the use of @code{typeof}:
623 This declares @code{y} with the type of what @code{x} points to.
630 This declares @code{y} as an array of such values.
637 This declares @code{y} as an array of pointers to characters:
640 typeof (typeof (char *)[4]) y;
644 It is equivalent to the following traditional C declaration:
650 To see the meaning of the declaration using @code{typeof}, and why it
651 might be a useful way to write, rewrite it with these macros:
654 #define pointer(T) typeof(T *)
655 #define array(T, N) typeof(T [N])
659 Now the declaration can be rewritten this way:
662 array (pointer (char), 4) y;
666 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
667 pointers to @code{char}.
670 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
671 a more limited extension which permitted one to write
674 typedef @var{T} = @var{expr};
678 with the effect of declaring @var{T} to have the type of the expression
679 @var{expr}. This extension does not work with GCC 3 (versions between
680 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
681 relies on it should be rewritten to use @code{typeof}:
684 typedef typeof(@var{expr}) @var{T};
688 This will work with all versions of GCC@.
691 @section Conditionals with Omitted Operands
692 @cindex conditional expressions, extensions
693 @cindex omitted middle-operands
694 @cindex middle-operands, omitted
695 @cindex extensions, @code{?:}
696 @cindex @code{?:} extensions
698 The middle operand in a conditional expression may be omitted. Then
699 if the first operand is nonzero, its value is the value of the conditional
702 Therefore, the expression
709 has the value of @code{x} if that is nonzero; otherwise, the value of
712 This example is perfectly equivalent to
718 @cindex side effect in ?:
719 @cindex ?: side effect
721 In this simple case, the ability to omit the middle operand is not
722 especially useful. When it becomes useful is when the first operand does,
723 or may (if it is a macro argument), contain a side effect. Then repeating
724 the operand in the middle would perform the side effect twice. Omitting
725 the middle operand uses the value already computed without the undesirable
726 effects of recomputing it.
729 @section Double-Word Integers
730 @cindex @code{long long} data types
731 @cindex double-word arithmetic
732 @cindex multiprecision arithmetic
733 @cindex @code{LL} integer suffix
734 @cindex @code{ULL} integer suffix
736 ISO C99 supports data types for integers that are at least 64 bits wide,
737 and as an extension GCC supports them in C89 mode and in C++.
738 Simply write @code{long long int} for a signed integer, or
739 @code{unsigned long long int} for an unsigned integer. To make an
740 integer constant of type @code{long long int}, add the suffix @samp{LL}
741 to the integer. To make an integer constant of type @code{unsigned long
742 long int}, add the suffix @samp{ULL} to the integer.
744 You can use these types in arithmetic like any other integer types.
745 Addition, subtraction, and bitwise boolean operations on these types
746 are open-coded on all types of machines. Multiplication is open-coded
747 if the machine supports fullword-to-doubleword a widening multiply
748 instruction. Division and shifts are open-coded only on machines that
749 provide special support. The operations that are not open-coded use
750 special library routines that come with GCC@.
752 There may be pitfalls when you use @code{long long} types for function
753 arguments, unless you declare function prototypes. If a function
754 expects type @code{int} for its argument, and you pass a value of type
755 @code{long long int}, confusion will result because the caller and the
756 subroutine will disagree about the number of bytes for the argument.
757 Likewise, if the function expects @code{long long int} and you pass
758 @code{int}. The best way to avoid such problems is to use prototypes.
761 @section Complex Numbers
762 @cindex complex numbers
763 @cindex @code{_Complex} keyword
764 @cindex @code{__complex__} keyword
766 ISO C99 supports complex floating data types, and as an extension GCC
767 supports them in C89 mode and in C++, and supports complex integer data
768 types which are not part of ISO C99. You can declare complex types
769 using the keyword @code{_Complex}. As an extension, the older GNU
770 keyword @code{__complex__} is also supported.
772 For example, @samp{_Complex double x;} declares @code{x} as a
773 variable whose real part and imaginary part are both of type
774 @code{double}. @samp{_Complex short int y;} declares @code{y} to
775 have real and imaginary parts of type @code{short int}; this is not
776 likely to be useful, but it shows that the set of complex types is
779 To write a constant with a complex data type, use the suffix @samp{i} or
780 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
781 has type @code{_Complex float} and @code{3i} has type
782 @code{_Complex int}. Such a constant always has a pure imaginary
783 value, but you can form any complex value you like by adding one to a
784 real constant. This is a GNU extension; if you have an ISO C99
785 conforming C library (such as GNU libc), and want to construct complex
786 constants of floating type, you should include @code{<complex.h>} and
787 use the macros @code{I} or @code{_Complex_I} instead.
789 @cindex @code{__real__} keyword
790 @cindex @code{__imag__} keyword
791 To extract the real part of a complex-valued expression @var{exp}, write
792 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
793 extract the imaginary part. This is a GNU extension; for values of
794 floating type, you should use the ISO C99 functions @code{crealf},
795 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
796 @code{cimagl}, declared in @code{<complex.h>} and also provided as
797 built-in functions by GCC@.
799 @cindex complex conjugation
800 The operator @samp{~} performs complex conjugation when used on a value
801 with a complex type. This is a GNU extension; for values of
802 floating type, you should use the ISO C99 functions @code{conjf},
803 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
804 provided as built-in functions by GCC@.
806 GCC can allocate complex automatic variables in a noncontiguous
807 fashion; it's even possible for the real part to be in a register while
808 the imaginary part is on the stack (or vice-versa). Only the DWARF2
809 debug info format can represent this, so use of DWARF2 is recommended.
810 If you are using the stabs debug info format, GCC describes a noncontiguous
811 complex variable as if it were two separate variables of noncomplex type.
812 If the variable's actual name is @code{foo}, the two fictitious
813 variables are named @code{foo$real} and @code{foo$imag}. You can
814 examine and set these two fictitious variables with your debugger.
820 ISO C99 supports floating-point numbers written not only in the usual
821 decimal notation, such as @code{1.55e1}, but also numbers such as
822 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
823 supports this in C89 mode (except in some cases when strictly
824 conforming) and in C++. In that format the
825 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
826 mandatory. The exponent is a decimal number that indicates the power of
827 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
834 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
835 is the same as @code{1.55e1}.
837 Unlike for floating-point numbers in the decimal notation the exponent
838 is always required in the hexadecimal notation. Otherwise the compiler
839 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
840 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
841 extension for floating-point constants of type @code{float}.
844 @section Arrays of Length Zero
845 @cindex arrays of length zero
846 @cindex zero-length arrays
847 @cindex length-zero arrays
848 @cindex flexible array members
850 Zero-length arrays are allowed in GNU C@. They are very useful as the
851 last element of a structure which is really a header for a variable-length
860 struct line *thisline = (struct line *)
861 malloc (sizeof (struct line) + this_length);
862 thisline->length = this_length;
865 In ISO C90, you would have to give @code{contents} a length of 1, which
866 means either you waste space or complicate the argument to @code{malloc}.
868 In ISO C99, you would use a @dfn{flexible array member}, which is
869 slightly different in syntax and semantics:
873 Flexible array members are written as @code{contents[]} without
877 Flexible array members have incomplete type, and so the @code{sizeof}
878 operator may not be applied. As a quirk of the original implementation
879 of zero-length arrays, @code{sizeof} evaluates to zero.
882 Flexible array members may only appear as the last member of a
883 @code{struct} that is otherwise non-empty.
886 A structure containing a flexible array member, or a union containing
887 such a structure (possibly recursively), may not be a member of a
888 structure or an element of an array. (However, these uses are
889 permitted by GCC as extensions.)
892 GCC versions before 3.0 allowed zero-length arrays to be statically
893 initialized, as if they were flexible arrays. In addition to those
894 cases that were useful, it also allowed initializations in situations
895 that would corrupt later data. Non-empty initialization of zero-length
896 arrays is now treated like any case where there are more initializer
897 elements than the array holds, in that a suitable warning about "excess
898 elements in array" is given, and the excess elements (all of them, in
899 this case) are ignored.
901 Instead GCC allows static initialization of flexible array members.
902 This is equivalent to defining a new structure containing the original
903 structure followed by an array of sufficient size to contain the data.
904 I.e.@: in the following, @code{f1} is constructed as if it were declared
910 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
913 struct f1 f1; int data[3];
914 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
918 The convenience of this extension is that @code{f1} has the desired
919 type, eliminating the need to consistently refer to @code{f2.f1}.
921 This has symmetry with normal static arrays, in that an array of
922 unknown size is also written with @code{[]}.
924 Of course, this extension only makes sense if the extra data comes at
925 the end of a top-level object, as otherwise we would be overwriting
926 data at subsequent offsets. To avoid undue complication and confusion
927 with initialization of deeply nested arrays, we simply disallow any
928 non-empty initialization except when the structure is the top-level
932 struct foo @{ int x; int y[]; @};
933 struct bar @{ struct foo z; @};
935 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
936 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
937 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
938 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
941 @node Empty Structures
942 @section Structures With No Members
943 @cindex empty structures
944 @cindex zero-size structures
946 GCC permits a C structure to have no members:
953 The structure will have size zero. In C++, empty structures are part
954 of the language. G++ treats empty structures as if they had a single
955 member of type @code{char}.
957 @node Variable Length
958 @section Arrays of Variable Length
959 @cindex variable-length arrays
960 @cindex arrays of variable length
963 Variable-length automatic arrays are allowed in ISO C99, and as an
964 extension GCC accepts them in C89 mode and in C++. (However, GCC's
965 implementation of variable-length arrays does not yet conform in detail
966 to the ISO C99 standard.) These arrays are
967 declared like any other automatic arrays, but with a length that is not
968 a constant expression. The storage is allocated at the point of
969 declaration and deallocated when the brace-level is exited. For
974 concat_fopen (char *s1, char *s2, char *mode)
976 char str[strlen (s1) + strlen (s2) + 1];
979 return fopen (str, mode);
983 @cindex scope of a variable length array
984 @cindex variable-length array scope
985 @cindex deallocating variable length arrays
986 Jumping or breaking out of the scope of the array name deallocates the
987 storage. Jumping into the scope is not allowed; you get an error
990 @cindex @code{alloca} vs variable-length arrays
991 You can use the function @code{alloca} to get an effect much like
992 variable-length arrays. The function @code{alloca} is available in
993 many other C implementations (but not in all). On the other hand,
994 variable-length arrays are more elegant.
996 There are other differences between these two methods. Space allocated
997 with @code{alloca} exists until the containing @emph{function} returns.
998 The space for a variable-length array is deallocated as soon as the array
999 name's scope ends. (If you use both variable-length arrays and
1000 @code{alloca} in the same function, deallocation of a variable-length array
1001 will also deallocate anything more recently allocated with @code{alloca}.)
1003 You can also use variable-length arrays as arguments to functions:
1007 tester (int len, char data[len][len])
1013 The length of an array is computed once when the storage is allocated
1014 and is remembered for the scope of the array in case you access it with
1017 If you want to pass the array first and the length afterward, you can
1018 use a forward declaration in the parameter list---another GNU extension.
1022 tester (int len; char data[len][len], int len)
1028 @cindex parameter forward declaration
1029 The @samp{int len} before the semicolon is a @dfn{parameter forward
1030 declaration}, and it serves the purpose of making the name @code{len}
1031 known when the declaration of @code{data} is parsed.
1033 You can write any number of such parameter forward declarations in the
1034 parameter list. They can be separated by commas or semicolons, but the
1035 last one must end with a semicolon, which is followed by the ``real''
1036 parameter declarations. Each forward declaration must match a ``real''
1037 declaration in parameter name and data type. ISO C99 does not support
1038 parameter forward declarations.
1040 @node Variadic Macros
1041 @section Macros with a Variable Number of Arguments.
1042 @cindex variable number of arguments
1043 @cindex macro with variable arguments
1044 @cindex rest argument (in macro)
1045 @cindex variadic macros
1047 In the ISO C standard of 1999, a macro can be declared to accept a
1048 variable number of arguments much as a function can. The syntax for
1049 defining the macro is similar to that of a function. Here is an
1053 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1056 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1057 such a macro, it represents the zero or more tokens until the closing
1058 parenthesis that ends the invocation, including any commas. This set of
1059 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1060 wherever it appears. See the CPP manual for more information.
1062 GCC has long supported variadic macros, and used a different syntax that
1063 allowed you to give a name to the variable arguments just like any other
1064 argument. Here is an example:
1067 #define debug(format, args...) fprintf (stderr, format, args)
1070 This is in all ways equivalent to the ISO C example above, but arguably
1071 more readable and descriptive.
1073 GNU CPP has two further variadic macro extensions, and permits them to
1074 be used with either of the above forms of macro definition.
1076 In standard C, you are not allowed to leave the variable argument out
1077 entirely; but you are allowed to pass an empty argument. For example,
1078 this invocation is invalid in ISO C, because there is no comma after
1085 GNU CPP permits you to completely omit the variable arguments in this
1086 way. In the above examples, the compiler would complain, though since
1087 the expansion of the macro still has the extra comma after the format
1090 To help solve this problem, CPP behaves specially for variable arguments
1091 used with the token paste operator, @samp{##}. If instead you write
1094 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1097 and if the variable arguments are omitted or empty, the @samp{##}
1098 operator causes the preprocessor to remove the comma before it. If you
1099 do provide some variable arguments in your macro invocation, GNU CPP
1100 does not complain about the paste operation and instead places the
1101 variable arguments after the comma. Just like any other pasted macro
1102 argument, these arguments are not macro expanded.
1104 @node Escaped Newlines
1105 @section Slightly Looser Rules for Escaped Newlines
1106 @cindex escaped newlines
1107 @cindex newlines (escaped)
1109 Recently, the preprocessor has relaxed its treatment of escaped
1110 newlines. Previously, the newline had to immediately follow a
1111 backslash. The current implementation allows whitespace in the form
1112 of spaces, horizontal and vertical tabs, and form feeds between the
1113 backslash and the subsequent newline. The preprocessor issues a
1114 warning, but treats it as a valid escaped newline and combines the two
1115 lines to form a single logical line. This works within comments and
1116 tokens, as well as between tokens. Comments are @emph{not} treated as
1117 whitespace for the purposes of this relaxation, since they have not
1118 yet been replaced with spaces.
1121 @section Non-Lvalue Arrays May Have Subscripts
1122 @cindex subscripting
1123 @cindex arrays, non-lvalue
1125 @cindex subscripting and function values
1126 In ISO C99, arrays that are not lvalues still decay to pointers, and
1127 may be subscripted, although they may not be modified or used after
1128 the next sequence point and the unary @samp{&} operator may not be
1129 applied to them. As an extension, GCC allows such arrays to be
1130 subscripted in C89 mode, though otherwise they do not decay to
1131 pointers outside C99 mode. For example,
1132 this is valid in GNU C though not valid in C89:
1136 struct foo @{int a[4];@};
1142 return f().a[index];
1148 @section Arithmetic on @code{void}- and Function-Pointers
1149 @cindex void pointers, arithmetic
1150 @cindex void, size of pointer to
1151 @cindex function pointers, arithmetic
1152 @cindex function, size of pointer to
1154 In GNU C, addition and subtraction operations are supported on pointers to
1155 @code{void} and on pointers to functions. This is done by treating the
1156 size of a @code{void} or of a function as 1.
1158 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1159 and on function types, and returns 1.
1161 @opindex Wpointer-arith
1162 The option @option{-Wpointer-arith} requests a warning if these extensions
1166 @section Non-Constant Initializers
1167 @cindex initializers, non-constant
1168 @cindex non-constant initializers
1170 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1171 automatic variable are not required to be constant expressions in GNU C@.
1172 Here is an example of an initializer with run-time varying elements:
1175 foo (float f, float g)
1177 float beat_freqs[2] = @{ f-g, f+g @};
1182 @node Compound Literals
1183 @section Compound Literals
1184 @cindex constructor expressions
1185 @cindex initializations in expressions
1186 @cindex structures, constructor expression
1187 @cindex expressions, constructor
1188 @cindex compound literals
1189 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1191 ISO C99 supports compound literals. A compound literal looks like
1192 a cast containing an initializer. Its value is an object of the
1193 type specified in the cast, containing the elements specified in
1194 the initializer; it is an lvalue. As an extension, GCC supports
1195 compound literals in C89 mode and in C++.
1197 Usually, the specified type is a structure. Assume that
1198 @code{struct foo} and @code{structure} are declared as shown:
1201 struct foo @{int a; char b[2];@} structure;
1205 Here is an example of constructing a @code{struct foo} with a compound literal:
1208 structure = ((struct foo) @{x + y, 'a', 0@});
1212 This is equivalent to writing the following:
1216 struct foo temp = @{x + y, 'a', 0@};
1221 You can also construct an array. If all the elements of the compound literal
1222 are (made up of) simple constant expressions, suitable for use in
1223 initializers of objects of static storage duration, then the compound
1224 literal can be coerced to a pointer to its first element and used in
1225 such an initializer, as shown here:
1228 char **foo = (char *[]) @{ "x", "y", "z" @};
1231 Compound literals for scalar types and union types are is
1232 also allowed, but then the compound literal is equivalent
1235 As a GNU extension, GCC allows initialization of objects with static storage
1236 duration by compound literals (which is not possible in ISO C99, because
1237 the initializer is not a constant).
1238 It is handled as if the object was initialized only with the bracket
1239 enclosed list if compound literal's and object types match.
1240 The initializer list of the compound literal must be constant.
1241 If the object being initialized has array type of unknown size, the size is
1242 determined by compound literal size.
1245 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1246 static int y[] = (int []) @{1, 2, 3@};
1247 static int z[] = (int [3]) @{1@};
1251 The above lines are equivalent to the following:
1253 static struct foo x = @{1, 'a', 'b'@};
1254 static int y[] = @{1, 2, 3@};
1255 static int z[] = @{1, 0, 0@};
1258 @node Designated Inits
1259 @section Designated Initializers
1260 @cindex initializers with labeled elements
1261 @cindex labeled elements in initializers
1262 @cindex case labels in initializers
1263 @cindex designated initializers
1265 Standard C89 requires the elements of an initializer to appear in a fixed
1266 order, the same as the order of the elements in the array or structure
1269 In ISO C99 you can give the elements in any order, specifying the array
1270 indices or structure field names they apply to, and GNU C allows this as
1271 an extension in C89 mode as well. This extension is not
1272 implemented in GNU C++.
1274 To specify an array index, write
1275 @samp{[@var{index}] =} before the element value. For example,
1278 int a[6] = @{ [4] = 29, [2] = 15 @};
1285 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1289 The index values must be constant expressions, even if the array being
1290 initialized is automatic.
1292 An alternative syntax for this which has been obsolete since GCC 2.5 but
1293 GCC still accepts is to write @samp{[@var{index}]} before the element
1294 value, with no @samp{=}.
1296 To initialize a range of elements to the same value, write
1297 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1298 extension. For example,
1301 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1305 If the value in it has side-effects, the side-effects will happen only once,
1306 not for each initialized field by the range initializer.
1309 Note that the length of the array is the highest value specified
1312 In a structure initializer, specify the name of a field to initialize
1313 with @samp{.@var{fieldname} =} before the element value. For example,
1314 given the following structure,
1317 struct point @{ int x, y; @};
1321 the following initialization
1324 struct point p = @{ .y = yvalue, .x = xvalue @};
1331 struct point p = @{ xvalue, yvalue @};
1334 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1335 @samp{@var{fieldname}:}, as shown here:
1338 struct point p = @{ y: yvalue, x: xvalue @};
1342 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1343 @dfn{designator}. You can also use a designator (or the obsolete colon
1344 syntax) when initializing a union, to specify which element of the union
1345 should be used. For example,
1348 union foo @{ int i; double d; @};
1350 union foo f = @{ .d = 4 @};
1354 will convert 4 to a @code{double} to store it in the union using
1355 the second element. By contrast, casting 4 to type @code{union foo}
1356 would store it into the union as the integer @code{i}, since it is
1357 an integer. (@xref{Cast to Union}.)
1359 You can combine this technique of naming elements with ordinary C
1360 initialization of successive elements. Each initializer element that
1361 does not have a designator applies to the next consecutive element of the
1362 array or structure. For example,
1365 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1372 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1375 Labeling the elements of an array initializer is especially useful
1376 when the indices are characters or belong to an @code{enum} type.
1381 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1382 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1385 @cindex designator lists
1386 You can also write a series of @samp{.@var{fieldname}} and
1387 @samp{[@var{index}]} designators before an @samp{=} to specify a
1388 nested subobject to initialize; the list is taken relative to the
1389 subobject corresponding to the closest surrounding brace pair. For
1390 example, with the @samp{struct point} declaration above:
1393 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1397 If the same field is initialized multiple times, it will have value from
1398 the last initialization. If any such overridden initialization has
1399 side-effect, it is unspecified whether the side-effect happens or not.
1400 Currently, GCC will discard them and issue a warning.
1403 @section Case Ranges
1405 @cindex ranges in case statements
1407 You can specify a range of consecutive values in a single @code{case} label,
1411 case @var{low} ... @var{high}:
1415 This has the same effect as the proper number of individual @code{case}
1416 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1418 This feature is especially useful for ranges of ASCII character codes:
1424 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1425 it may be parsed wrong when you use it with integer values. For example,
1440 @section Cast to a Union Type
1441 @cindex cast to a union
1442 @cindex union, casting to a
1444 A cast to union type is similar to other casts, except that the type
1445 specified is a union type. You can specify the type either with
1446 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1447 a constructor though, not a cast, and hence does not yield an lvalue like
1448 normal casts. (@xref{Compound Literals}.)
1450 The types that may be cast to the union type are those of the members
1451 of the union. Thus, given the following union and variables:
1454 union foo @{ int i; double d; @};
1460 both @code{x} and @code{y} can be cast to type @code{union foo}.
1462 Using the cast as the right-hand side of an assignment to a variable of
1463 union type is equivalent to storing in a member of the union:
1468 u = (union foo) x @equiv{} u.i = x
1469 u = (union foo) y @equiv{} u.d = y
1472 You can also use the union cast as a function argument:
1475 void hack (union foo);
1477 hack ((union foo) x);
1480 @node Mixed Declarations
1481 @section Mixed Declarations and Code
1482 @cindex mixed declarations and code
1483 @cindex declarations, mixed with code
1484 @cindex code, mixed with declarations
1486 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1487 within compound statements. As an extension, GCC also allows this in
1488 C89 mode. For example, you could do:
1497 Each identifier is visible from where it is declared until the end of
1498 the enclosing block.
1500 @node Function Attributes
1501 @section Declaring Attributes of Functions
1502 @cindex function attributes
1503 @cindex declaring attributes of functions
1504 @cindex functions that never return
1505 @cindex functions that return more than once
1506 @cindex functions that have no side effects
1507 @cindex functions in arbitrary sections
1508 @cindex functions that behave like malloc
1509 @cindex @code{volatile} applied to function
1510 @cindex @code{const} applied to function
1511 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1512 @cindex functions with non-null pointer arguments
1513 @cindex functions that are passed arguments in registers on the 386
1514 @cindex functions that pop the argument stack on the 386
1515 @cindex functions that do not pop the argument stack on the 386
1517 In GNU C, you declare certain things about functions called in your program
1518 which help the compiler optimize function calls and check your code more
1521 The keyword @code{__attribute__} allows you to specify special
1522 attributes when making a declaration. This keyword is followed by an
1523 attribute specification inside double parentheses. The following
1524 attributes are currently defined for functions on all targets:
1525 @code{noreturn}, @code{returns_twice}, @code{noinline}, @code{always_inline},
1526 @code{flatten}, @code{pure}, @code{const}, @code{nothrow}, @code{sentinel},
1527 @code{format}, @code{format_arg}, @code{no_instrument_function},
1528 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1529 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1530 @code{alias}, @code{warn_unused_result}, @code{nonnull}
1531 and @code{externally_visible}. Several other
1532 attributes are defined for functions on particular target systems. Other
1533 attributes, including @code{section} are supported for variables declarations
1534 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1536 You may also specify attributes with @samp{__} preceding and following
1537 each keyword. This allows you to use them in header files without
1538 being concerned about a possible macro of the same name. For example,
1539 you may use @code{__noreturn__} instead of @code{noreturn}.
1541 @xref{Attribute Syntax}, for details of the exact syntax for using
1545 @c Keep this table alphabetized by attribute name. Treat _ as space.
1547 @item alias ("@var{target}")
1548 @cindex @code{alias} attribute
1549 The @code{alias} attribute causes the declaration to be emitted as an
1550 alias for another symbol, which must be specified. For instance,
1553 void __f () @{ /* @r{Do something.} */; @}
1554 void f () __attribute__ ((weak, alias ("__f")));
1557 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
1558 mangled name for the target must be used. It is an error if @samp{__f}
1559 is not defined in the same translation unit.
1561 Not all target machines support this attribute.
1564 @cindex @code{always_inline} function attribute
1565 Generally, functions are not inlined unless optimization is specified.
1566 For functions declared inline, this attribute inlines the function even
1567 if no optimization level was specified.
1569 @cindex @code{flatten} function attribute
1571 Generally, inlining into a function is limited. For a function marked with
1572 this attribute, every call inside this function will be inlined, if possible.
1573 Whether the function itself is considered for inlining depends on its size and
1574 the current inlining parameters. The @code{flatten} attribute only works
1575 reliably in unit-at-a-time mode.
1578 @cindex functions that do pop the argument stack on the 386
1580 On the Intel 386, the @code{cdecl} attribute causes the compiler to
1581 assume that the calling function will pop off the stack space used to
1582 pass arguments. This is
1583 useful to override the effects of the @option{-mrtd} switch.
1586 @cindex @code{const} function attribute
1587 Many functions do not examine any values except their arguments, and
1588 have no effects except the return value. Basically this is just slightly
1589 more strict class than the @code{pure} attribute below, since function is not
1590 allowed to read global memory.
1592 @cindex pointer arguments
1593 Note that a function that has pointer arguments and examines the data
1594 pointed to must @emph{not} be declared @code{const}. Likewise, a
1595 function that calls a non-@code{const} function usually must not be
1596 @code{const}. It does not make sense for a @code{const} function to
1599 The attribute @code{const} is not implemented in GCC versions earlier
1600 than 2.5. An alternative way to declare that a function has no side
1601 effects, which works in the current version and in some older versions,
1605 typedef int intfn ();
1607 extern const intfn square;
1610 This approach does not work in GNU C++ from 2.6.0 on, since the language
1611 specifies that the @samp{const} must be attached to the return value.
1615 @cindex @code{constructor} function attribute
1616 @cindex @code{destructor} function attribute
1617 The @code{constructor} attribute causes the function to be called
1618 automatically before execution enters @code{main ()}. Similarly, the
1619 @code{destructor} attribute causes the function to be called
1620 automatically after @code{main ()} has completed or @code{exit ()} has
1621 been called. Functions with these attributes are useful for
1622 initializing data that will be used implicitly during the execution of
1625 These attributes are not currently implemented for Objective-C@.
1628 @cindex @code{deprecated} attribute.
1629 The @code{deprecated} attribute results in a warning if the function
1630 is used anywhere in the source file. This is useful when identifying
1631 functions that are expected to be removed in a future version of a
1632 program. The warning also includes the location of the declaration
1633 of the deprecated function, to enable users to easily find further
1634 information about why the function is deprecated, or what they should
1635 do instead. Note that the warnings only occurs for uses:
1638 int old_fn () __attribute__ ((deprecated));
1640 int (*fn_ptr)() = old_fn;
1643 results in a warning on line 3 but not line 2.
1645 The @code{deprecated} attribute can also be used for variables and
1646 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
1649 @cindex @code{__declspec(dllexport)}
1650 On Microsoft Windows targets and Symbian OS targets the
1651 @code{dllexport} attribute causes the compiler to provide a global
1652 pointer to a pointer in a DLL, so that it can be referenced with the
1653 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
1654 name is formed by combining @code{_imp__} and the function or variable
1657 You can use @code{__declspec(dllexport)} as a synonym for
1658 @code{__attribute__ ((dllexport))} for compatibility with other
1661 On systems that support the @code{visibility} attribute, this
1662 attribute also implies ``default'' visibility, unless a
1663 @code{visibility} attribute is explicitly specified. You should avoid
1664 the use of @code{dllexport} with ``hidden'' or ``internal''
1665 visibility; in the future GCC may issue an error for those cases.
1667 Currently, the @code{dllexport} attribute is ignored for inlined
1668 functions, unless the @option{-fkeep-inline-functions} flag has been
1669 used. The attribute is also ignored for undefined symbols.
1671 When applied to C++ classes, the attribute marks defined non-inlined
1672 member functions and static data members as exports. Static consts
1673 initialized in-class are not marked unless they are also defined
1676 For Microsoft Windows targets there are alternative methods for
1677 including the symbol in the DLL's export table such as using a
1678 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
1679 the @option{--export-all} linker flag.
1682 @cindex @code{__declspec(dllimport)}
1683 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
1684 attribute causes the compiler to reference a function or variable via
1685 a global pointer to a pointer that is set up by the DLL exporting the
1686 symbol. The attribute implies @code{extern} storage. On Microsoft
1687 Windows targets, the pointer name is formed by combining @code{_imp__}
1688 and the function or variable name.
1690 You can use @code{__declspec(dllimport)} as a synonym for
1691 @code{__attribute__ ((dllimport))} for compatibility with other
1694 Currently, the attribute is ignored for inlined functions. If the
1695 attribute is applied to a symbol @emph{definition}, an error is reported.
1696 If a symbol previously declared @code{dllimport} is later defined, the
1697 attribute is ignored in subsequent references, and a warning is emitted.
1698 The attribute is also overridden by a subsequent declaration as
1701 When applied to C++ classes, the attribute marks non-inlined
1702 member functions and static data members as imports. However, the
1703 attribute is ignored for virtual methods to allow creation of vtables
1706 On the SH Symbian OS target the @code{dllimport} attribute also has
1707 another affect---it can cause the vtable and run-time type information
1708 for a class to be exported. This happens when the class has a
1709 dllimport'ed constructor or a non-inline, non-pure virtual function
1710 and, for either of those two conditions, the class also has a inline
1711 constructor or destructor and has a key function that is defined in
1712 the current translation unit.
1714 For Microsoft Windows based targets the use of the @code{dllimport}
1715 attribute on functions is not necessary, but provides a small
1716 performance benefit by eliminating a thunk in the DLL@. The use of the
1717 @code{dllimport} attribute on imported variables was required on older
1718 versions of the GNU linker, but can now be avoided by passing the
1719 @option{--enable-auto-import} switch to the GNU linker. As with
1720 functions, using the attribute for a variable eliminates a thunk in
1723 One drawback to using this attribute is that a pointer to a function
1724 or variable marked as @code{dllimport} cannot be used as a constant
1725 address. On Microsoft Windows targets, the attribute can be disabled
1726 for functions by setting the @option{-mnop-fun-dllimport} flag.
1729 @cindex eight bit data on the H8/300, H8/300H, and H8S
1730 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1731 variable should be placed into the eight bit data section.
1732 The compiler will generate more efficient code for certain operations
1733 on data in the eight bit data area. Note the eight bit data area is limited to
1736 You must use GAS and GLD from GNU binutils version 2.7 or later for
1737 this attribute to work correctly.
1739 @item exception_handler
1740 @cindex exception handler functions on the Blackfin processor
1741 Use this attribute on the Blackfin to indicate that the specified function
1742 is an exception handler. The compiler will generate function entry and
1743 exit sequences suitable for use in an exception handler when this
1744 attribute is present.
1747 @cindex functions which handle memory bank switching
1748 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
1749 use a calling convention that takes care of switching memory banks when
1750 entering and leaving a function. This calling convention is also the
1751 default when using the @option{-mlong-calls} option.
1753 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
1754 to call and return from a function.
1756 On 68HC11 the compiler will generate a sequence of instructions
1757 to invoke a board-specific routine to switch the memory bank and call the
1758 real function. The board-specific routine simulates a @code{call}.
1759 At the end of a function, it will jump to a board-specific routine
1760 instead of using @code{rts}. The board-specific return routine simulates
1764 @cindex functions that pop the argument stack on the 386
1765 On the Intel 386, the @code{fastcall} attribute causes the compiler to
1766 pass the first argument (if of integral type) in the register ECX and
1767 the second argument (if of integral type) in the register EDX@. Subsequent
1768 and other typed arguments are passed on the stack. The called function will
1769 pop the arguments off the stack. If the number of arguments is variable all
1770 arguments are pushed on the stack.
1772 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
1773 @cindex @code{format} function attribute
1775 The @code{format} attribute specifies that a function takes @code{printf},
1776 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
1777 should be type-checked against a format string. For example, the
1782 my_printf (void *my_object, const char *my_format, ...)
1783 __attribute__ ((format (printf, 2, 3)));
1787 causes the compiler to check the arguments in calls to @code{my_printf}
1788 for consistency with the @code{printf} style format string argument
1791 The parameter @var{archetype} determines how the format string is
1792 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
1793 or @code{strfmon}. (You can also use @code{__printf__},
1794 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
1795 parameter @var{string-index} specifies which argument is the format
1796 string argument (starting from 1), while @var{first-to-check} is the
1797 number of the first argument to check against the format string. For
1798 functions where the arguments are not available to be checked (such as
1799 @code{vprintf}), specify the third parameter as zero. In this case the
1800 compiler only checks the format string for consistency. For
1801 @code{strftime} formats, the third parameter is required to be zero.
1802 Since non-static C++ methods have an implicit @code{this} argument, the
1803 arguments of such methods should be counted from two, not one, when
1804 giving values for @var{string-index} and @var{first-to-check}.
1806 In the example above, the format string (@code{my_format}) is the second
1807 argument of the function @code{my_print}, and the arguments to check
1808 start with the third argument, so the correct parameters for the format
1809 attribute are 2 and 3.
1811 @opindex ffreestanding
1812 @opindex fno-builtin
1813 The @code{format} attribute allows you to identify your own functions
1814 which take format strings as arguments, so that GCC can check the
1815 calls to these functions for errors. The compiler always (unless
1816 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
1817 for the standard library functions @code{printf}, @code{fprintf},
1818 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
1819 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
1820 warnings are requested (using @option{-Wformat}), so there is no need to
1821 modify the header file @file{stdio.h}. In C99 mode, the functions
1822 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
1823 @code{vsscanf} are also checked. Except in strictly conforming C
1824 standard modes, the X/Open function @code{strfmon} is also checked as
1825 are @code{printf_unlocked} and @code{fprintf_unlocked}.
1826 @xref{C Dialect Options,,Options Controlling C Dialect}.
1828 The target may provide additional types of format checks.
1829 @xref{Target Format Checks,,Format Checks Specific to Particular
1832 @item format_arg (@var{string-index})
1833 @cindex @code{format_arg} function attribute
1834 @opindex Wformat-nonliteral
1835 The @code{format_arg} attribute specifies that a function takes a format
1836 string for a @code{printf}, @code{scanf}, @code{strftime} or
1837 @code{strfmon} style function and modifies it (for example, to translate
1838 it into another language), so the result can be passed to a
1839 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
1840 function (with the remaining arguments to the format function the same
1841 as they would have been for the unmodified string). For example, the
1846 my_dgettext (char *my_domain, const char *my_format)
1847 __attribute__ ((format_arg (2)));
1851 causes the compiler to check the arguments in calls to a @code{printf},
1852 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
1853 format string argument is a call to the @code{my_dgettext} function, for
1854 consistency with the format string argument @code{my_format}. If the
1855 @code{format_arg} attribute had not been specified, all the compiler
1856 could tell in such calls to format functions would be that the format
1857 string argument is not constant; this would generate a warning when
1858 @option{-Wformat-nonliteral} is used, but the calls could not be checked
1859 without the attribute.
1861 The parameter @var{string-index} specifies which argument is the format
1862 string argument (starting from one). Since non-static C++ methods have
1863 an implicit @code{this} argument, the arguments of such methods should
1864 be counted from two.
1866 The @code{format-arg} attribute allows you to identify your own
1867 functions which modify format strings, so that GCC can check the
1868 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
1869 type function whose operands are a call to one of your own function.
1870 The compiler always treats @code{gettext}, @code{dgettext}, and
1871 @code{dcgettext} in this manner except when strict ISO C support is
1872 requested by @option{-ansi} or an appropriate @option{-std} option, or
1873 @option{-ffreestanding} or @option{-fno-builtin}
1874 is used. @xref{C Dialect Options,,Options
1875 Controlling C Dialect}.
1877 @item function_vector
1878 @cindex calling functions through the function vector on the H8/300 processors
1879 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1880 function should be called through the function vector. Calling a
1881 function through the function vector will reduce code size, however;
1882 the function vector has a limited size (maximum 128 entries on the H8/300
1883 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
1885 You must use GAS and GLD from GNU binutils version 2.7 or later for
1886 this attribute to work correctly.
1889 @cindex interrupt handler functions
1890 Use this attribute on the ARM, AVR, C4x, CRX, M32C, M32R/D, MS1, and Xstormy16
1891 ports to indicate that the specified function is an interrupt handler.
1892 The compiler will generate function entry and exit sequences suitable
1893 for use in an interrupt handler when this attribute is present.
1895 Note, interrupt handlers for the Blackfin, m68k, H8/300, H8/300H, H8S, and
1896 SH processors can be specified via the @code{interrupt_handler} attribute.
1898 Note, on the AVR, interrupts will be enabled inside the function.
1900 Note, for the ARM, you can specify the kind of interrupt to be handled by
1901 adding an optional parameter to the interrupt attribute like this:
1904 void f () __attribute__ ((interrupt ("IRQ")));
1907 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
1909 @item interrupt_handler
1910 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
1911 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
1912 indicate that the specified function is an interrupt handler. The compiler
1913 will generate function entry and exit sequences suitable for use in an
1914 interrupt handler when this attribute is present.
1917 @cindex User stack pointer in interrupts on the Blackfin
1918 When used together with @code{interrupt_handler}, @code{exception_handler}
1919 or @code{nmi_handler}, code will be generated to load the stack pointer
1920 from the USP register in the function prologue.
1922 @item long_call/short_call
1923 @cindex indirect calls on ARM
1924 This attribute specifies how a particular function is called on
1925 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
1926 command line switch and @code{#pragma long_calls} settings. The
1927 @code{long_call} attribute causes the compiler to always call the
1928 function by first loading its address into a register and then using the
1929 contents of that register. The @code{short_call} attribute always places
1930 the offset to the function from the call site into the @samp{BL}
1931 instruction directly.
1933 @item longcall/shortcall
1934 @cindex functions called via pointer on the RS/6000 and PowerPC
1935 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute causes
1936 the compiler to always call this function via a pointer, just as it would if
1937 the @option{-mlongcall} option had been specified. The @code{shortcall}
1938 attribute causes the compiler not to do this. These attributes override
1939 both the @option{-mlongcall} switch and, on the RS/6000 and PowerPC, the
1940 @code{#pragma longcall} setting.
1942 @xref{RS/6000 and PowerPC Options}, for more information on whether long
1943 calls are necessary.
1946 @cindex indirect calls on MIPS
1947 This attribute specifies how a particular function is called on MIPS@.
1948 The attribute overrides the @option{-mlong-calls} (@pxref{MIPS Options})
1949 command line switch. This attribute causes the compiler to always call
1950 the function by first loading its address into a register, and then using
1951 the contents of that register.
1954 @cindex @code{malloc} attribute
1955 The @code{malloc} attribute is used to tell the compiler that a function
1956 may be treated as if any non-@code{NULL} pointer it returns cannot
1957 alias any other pointer valid when the function returns.
1958 This will often improve optimization.
1959 Standard functions with this property include @code{malloc} and
1960 @code{calloc}. @code{realloc}-like functions have this property as
1961 long as the old pointer is never referred to (including comparing it
1962 to the new pointer) after the function returns a non-@code{NULL}
1965 @item model (@var{model-name})
1966 @cindex function addressability on the M32R/D
1967 @cindex variable addressability on the IA-64
1969 On the M32R/D, use this attribute to set the addressability of an
1970 object, and of the code generated for a function. The identifier
1971 @var{model-name} is one of @code{small}, @code{medium}, or
1972 @code{large}, representing each of the code models.
1974 Small model objects live in the lower 16MB of memory (so that their
1975 addresses can be loaded with the @code{ld24} instruction), and are
1976 callable with the @code{bl} instruction.
1978 Medium model objects may live anywhere in the 32-bit address space (the
1979 compiler will generate @code{seth/add3} instructions to load their addresses),
1980 and are callable with the @code{bl} instruction.
1982 Large model objects may live anywhere in the 32-bit address space (the
1983 compiler will generate @code{seth/add3} instructions to load their addresses),
1984 and may not be reachable with the @code{bl} instruction (the compiler will
1985 generate the much slower @code{seth/add3/jl} instruction sequence).
1987 On IA-64, use this attribute to set the addressability of an object.
1988 At present, the only supported identifier for @var{model-name} is
1989 @code{small}, indicating addressability via ``small'' (22-bit)
1990 addresses (so that their addresses can be loaded with the @code{addl}
1991 instruction). Caveat: such addressing is by definition not position
1992 independent and hence this attribute must not be used for objects
1993 defined by shared libraries.
1996 @cindex function without a prologue/epilogue code
1997 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate that the
1998 specified function does not need prologue/epilogue sequences generated by
1999 the compiler. It is up to the programmer to provide these sequences.
2002 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2003 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2004 use the normal calling convention based on @code{jsr} and @code{rts}.
2005 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2009 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2010 Use this attribute together with @code{interrupt_handler},
2011 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2012 entry code should enable nested interrupts or exceptions.
2015 @cindex NMI handler functions on the Blackfin processor
2016 Use this attribute on the Blackfin to indicate that the specified function
2017 is an NMI handler. The compiler will generate function entry and
2018 exit sequences suitable for use in an NMI handler when this
2019 attribute is present.
2021 @item no_instrument_function
2022 @cindex @code{no_instrument_function} function attribute
2023 @opindex finstrument-functions
2024 If @option{-finstrument-functions} is given, profiling function calls will
2025 be generated at entry and exit of most user-compiled functions.
2026 Functions with this attribute will not be so instrumented.
2029 @cindex @code{noinline} function attribute
2030 This function attribute prevents a function from being considered for
2033 @item nonnull (@var{arg-index}, @dots{})
2034 @cindex @code{nonnull} function attribute
2035 The @code{nonnull} attribute specifies that some function parameters should
2036 be non-null pointers. For instance, the declaration:
2040 my_memcpy (void *dest, const void *src, size_t len)
2041 __attribute__((nonnull (1, 2)));
2045 causes the compiler to check that, in calls to @code{my_memcpy},
2046 arguments @var{dest} and @var{src} are non-null. If the compiler
2047 determines that a null pointer is passed in an argument slot marked
2048 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2049 is issued. The compiler may also choose to make optimizations based
2050 on the knowledge that certain function arguments will not be null.
2052 If no argument index list is given to the @code{nonnull} attribute,
2053 all pointer arguments are marked as non-null. To illustrate, the
2054 following declaration is equivalent to the previous example:
2058 my_memcpy (void *dest, const void *src, size_t len)
2059 __attribute__((nonnull));
2063 @cindex @code{noreturn} function attribute
2064 A few standard library functions, such as @code{abort} and @code{exit},
2065 cannot return. GCC knows this automatically. Some programs define
2066 their own functions that never return. You can declare them
2067 @code{noreturn} to tell the compiler this fact. For example,
2071 void fatal () __attribute__ ((noreturn));
2074 fatal (/* @r{@dots{}} */)
2076 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2082 The @code{noreturn} keyword tells the compiler to assume that
2083 @code{fatal} cannot return. It can then optimize without regard to what
2084 would happen if @code{fatal} ever did return. This makes slightly
2085 better code. More importantly, it helps avoid spurious warnings of
2086 uninitialized variables.
2088 The @code{noreturn} keyword does not affect the exceptional path when that
2089 applies: a @code{noreturn}-marked function may still return to the caller
2090 by throwing an exception or calling @code{longjmp}.
2092 Do not assume that registers saved by the calling function are
2093 restored before calling the @code{noreturn} function.
2095 It does not make sense for a @code{noreturn} function to have a return
2096 type other than @code{void}.
2098 The attribute @code{noreturn} is not implemented in GCC versions
2099 earlier than 2.5. An alternative way to declare that a function does
2100 not return, which works in the current version and in some older
2101 versions, is as follows:
2104 typedef void voidfn ();
2106 volatile voidfn fatal;
2109 This approach does not work in GNU C++.
2112 @cindex @code{nothrow} function attribute
2113 The @code{nothrow} attribute is used to inform the compiler that a
2114 function cannot throw an exception. For example, most functions in
2115 the standard C library can be guaranteed not to throw an exception
2116 with the notable exceptions of @code{qsort} and @code{bsearch} that
2117 take function pointer arguments. The @code{nothrow} attribute is not
2118 implemented in GCC versions earlier than 3.3.
2121 @cindex @code{pure} function attribute
2122 Many functions have no effects except the return value and their
2123 return value depends only on the parameters and/or global variables.
2124 Such a function can be subject
2125 to common subexpression elimination and loop optimization just as an
2126 arithmetic operator would be. These functions should be declared
2127 with the attribute @code{pure}. For example,
2130 int square (int) __attribute__ ((pure));
2134 says that the hypothetical function @code{square} is safe to call
2135 fewer times than the program says.
2137 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2138 Interesting non-pure functions are functions with infinite loops or those
2139 depending on volatile memory or other system resource, that may change between
2140 two consecutive calls (such as @code{feof} in a multithreading environment).
2142 The attribute @code{pure} is not implemented in GCC versions earlier
2145 @item regparm (@var{number})
2146 @cindex @code{regparm} attribute
2147 @cindex functions that are passed arguments in registers on the 386
2148 On the Intel 386, the @code{regparm} attribute causes the compiler to
2149 pass arguments number one to @var{number} if they are of integral type
2150 in registers EAX, EDX, and ECX instead of on the stack. Functions that
2151 take a variable number of arguments will continue to be passed all of their
2152 arguments on the stack.
2154 Beware that on some ELF systems this attribute is unsuitable for
2155 global functions in shared libraries with lazy binding (which is the
2156 default). Lazy binding will send the first call via resolving code in
2157 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2158 per the standard calling conventions. Solaris 8 is affected by this.
2159 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2160 safe since the loaders there save all registers. (Lazy binding can be
2161 disabled with the linker or the loader if desired, to avoid the
2165 @cindex @code{sseregparm} attribute
2166 On the Intel 386 with SSE support, the @code{sseregparm} attribute
2167 causes the compiler to pass up to 8 floating point arguments in
2168 SSE registers instead of on the stack. Functions that take a
2169 variable number of arguments will continue to pass all of their
2170 floating point arguments on the stack.
2173 @cindex @code{returns_twice} attribute
2174 The @code{returns_twice} attribute tells the compiler that a function may
2175 return more than one time. The compiler will ensure that all registers
2176 are dead before calling such a function and will emit a warning about
2177 the variables that may be clobbered after the second return from the
2178 function. Examples of such functions are @code{setjmp} and @code{vfork}.
2179 The @code{longjmp}-like counterpart of such function, if any, might need
2180 to be marked with the @code{noreturn} attribute.
2183 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
2184 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
2185 all registers except the stack pointer should be saved in the prologue
2186 regardless of whether they are used or not.
2188 @item section ("@var{section-name}")
2189 @cindex @code{section} function attribute
2190 Normally, the compiler places the code it generates in the @code{text} section.
2191 Sometimes, however, you need additional sections, or you need certain
2192 particular functions to appear in special sections. The @code{section}
2193 attribute specifies that a function lives in a particular section.
2194 For example, the declaration:
2197 extern void foobar (void) __attribute__ ((section ("bar")));
2201 puts the function @code{foobar} in the @code{bar} section.
2203 Some file formats do not support arbitrary sections so the @code{section}
2204 attribute is not available on all platforms.
2205 If you need to map the entire contents of a module to a particular
2206 section, consider using the facilities of the linker instead.
2209 @cindex @code{sentinel} function attribute
2210 This function attribute ensures that a parameter in a function call is
2211 an explicit @code{NULL}. The attribute is only valid on variadic
2212 functions. By default, the sentinel is located at position zero, the
2213 last parameter of the function call. If an optional integer position
2214 argument P is supplied to the attribute, the sentinel must be located at
2215 position P counting backwards from the end of the argument list.
2218 __attribute__ ((sentinel))
2220 __attribute__ ((sentinel(0)))
2223 The attribute is automatically set with a position of 0 for the built-in
2224 functions @code{execl} and @code{execlp}. The built-in function
2225 @code{execle} has the attribute set with a position of 1.
2227 A valid @code{NULL} in this context is defined as zero with any pointer
2228 type. If your system defines the @code{NULL} macro with an integer type
2229 then you need to add an explicit cast. GCC replaces @code{stddef.h}
2230 with a copy that redefines NULL appropriately.
2232 The warnings for missing or incorrect sentinels are enabled with
2236 See long_call/short_call.
2239 See longcall/shortcall.
2242 @cindex signal handler functions on the AVR processors
2243 Use this attribute on the AVR to indicate that the specified
2244 function is a signal handler. The compiler will generate function
2245 entry and exit sequences suitable for use in a signal handler when this
2246 attribute is present. Interrupts will be disabled inside the function.
2249 Use this attribute on the SH to indicate an @code{interrupt_handler}
2250 function should switch to an alternate stack. It expects a string
2251 argument that names a global variable holding the address of the
2256 void f () __attribute__ ((interrupt_handler,
2257 sp_switch ("alt_stack")));
2261 @cindex functions that pop the argument stack on the 386
2262 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2263 assume that the called function will pop off the stack space used to
2264 pass arguments, unless it takes a variable number of arguments.
2267 @cindex tiny data section on the H8/300H and H8S
2268 Use this attribute on the H8/300H and H8S to indicate that the specified
2269 variable should be placed into the tiny data section.
2270 The compiler will generate more efficient code for loads and stores
2271 on data in the tiny data section. Note the tiny data area is limited to
2272 slightly under 32kbytes of data.
2275 Use this attribute on the SH for an @code{interrupt_handler} to return using
2276 @code{trapa} instead of @code{rte}. This attribute expects an integer
2277 argument specifying the trap number to be used.
2280 @cindex @code{unused} attribute.
2281 This attribute, attached to a function, means that the function is meant
2282 to be possibly unused. GCC will not produce a warning for this
2286 @cindex @code{used} attribute.
2287 This attribute, attached to a function, means that code must be emitted
2288 for the function even if it appears that the function is not referenced.
2289 This is useful, for example, when the function is referenced only in
2292 @item visibility ("@var{visibility_type}")
2293 @cindex @code{visibility} attribute
2294 The @code{visibility} attribute on ELF targets causes the declaration
2295 to be emitted with default, hidden, protected or internal visibility.
2298 void __attribute__ ((visibility ("protected")))
2299 f () @{ /* @r{Do something.} */; @}
2300 int i __attribute__ ((visibility ("hidden")));
2303 See the ELF gABI for complete details, but the short story is:
2306 @c keep this list of visibilities in alphabetical order.
2309 Default visibility is the normal case for ELF@. This value is
2310 available for the visibility attribute to override other options
2311 that may change the assumed visibility of symbols.
2314 Hidden visibility indicates that the symbol will not be placed into
2315 the dynamic symbol table, so no other @dfn{module} (executable or
2316 shared library) can reference it directly.
2319 Internal visibility is like hidden visibility, but with additional
2320 processor specific semantics. Unless otherwise specified by the psABI,
2321 GCC defines internal visibility to mean that the function is @emph{never}
2322 called from another module. Note that hidden symbols, while they cannot
2323 be referenced directly by other modules, can be referenced indirectly via
2324 function pointers. By indicating that a symbol cannot be called from
2325 outside the module, GCC may for instance omit the load of a PIC register
2326 since it is known that the calling function loaded the correct value.
2329 Protected visibility indicates that the symbol will be placed in the
2330 dynamic symbol table, but that references within the defining module
2331 will bind to the local symbol. That is, the symbol cannot be overridden
2336 Not all ELF targets support this attribute.
2338 @item warn_unused_result
2339 @cindex @code{warn_unused_result} attribute
2340 The @code{warn_unused_result} attribute causes a warning to be emitted
2341 if a caller of the function with this attribute does not use its
2342 return value. This is useful for functions where not checking
2343 the result is either a security problem or always a bug, such as
2347 int fn () __attribute__ ((warn_unused_result));
2350 if (fn () < 0) return -1;
2356 results in warning on line 5.
2359 @cindex @code{weak} attribute
2360 The @code{weak} attribute causes the declaration to be emitted as a weak
2361 symbol rather than a global. This is primarily useful in defining
2362 library functions which can be overridden in user code, though it can
2363 also be used with non-function declarations. Weak symbols are supported
2364 for ELF targets, and also for a.out targets when using the GNU assembler
2368 @itemx weakref ("@var{target}")
2369 @cindex @code{weakref} attribute
2370 The @code{weakref} attribute marks a declaration as a weak reference.
2371 Without arguments, it should be accompanied by an @code{alias} attribute
2372 naming the target symbol. Optionally, the @var{target} may be given as
2373 an argument to @code{weakref} itself. In either case, @code{weakref}
2374 implicitly marks the declaration as @code{weak}. Without a
2375 @var{target}, given as an argument to @code{weakref} or to @code{alias},
2376 @code{weakref} is equivalent to @code{weak}.
2379 extern int x() __attribute__ ((weakref ("y")));
2380 /* is equivalent to... */
2381 extern int x() __attribute__ ((weak, weakref, alias ("y")));
2383 extern int x() __attribute__ ((weakref));
2384 extern int x() __attribute__ ((alias ("y")));
2387 A weak reference is an alias that does not by itself require a
2388 definition to be given for the target symbol. If the target symbol is
2389 only referenced through weak references, then the becomes a @code{weak}
2390 undefined symbol. If it is directly referenced, however, then such
2391 strong references prevail, and a definition will be required for the
2392 symbol, not necessarily in the same translation unit.
2394 The effect is equivalent to moving all references to the alias to a
2395 separate translation unit, renaming the alias to the aliased symbol,
2396 declaring it as weak, compiling the two separate translation units and
2397 performing a reloadable link on them.
2399 @item externally_visible
2400 @cindex @code{externally_visible} attribute.
2401 This attribute, attached to a global variable or function nullify
2402 effect of @option{-fwhole-program} command line option, so the object
2403 remain visible outside the current compilation unit
2407 You can specify multiple attributes in a declaration by separating them
2408 by commas within the double parentheses or by immediately following an
2409 attribute declaration with another attribute declaration.
2411 @cindex @code{#pragma}, reason for not using
2412 @cindex pragma, reason for not using
2413 Some people object to the @code{__attribute__} feature, suggesting that
2414 ISO C's @code{#pragma} should be used instead. At the time
2415 @code{__attribute__} was designed, there were two reasons for not doing
2420 It is impossible to generate @code{#pragma} commands from a macro.
2423 There is no telling what the same @code{#pragma} might mean in another
2427 These two reasons applied to almost any application that might have been
2428 proposed for @code{#pragma}. It was basically a mistake to use
2429 @code{#pragma} for @emph{anything}.
2431 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2432 to be generated from macros. In addition, a @code{#pragma GCC}
2433 namespace is now in use for GCC-specific pragmas. However, it has been
2434 found convenient to use @code{__attribute__} to achieve a natural
2435 attachment of attributes to their corresponding declarations, whereas
2436 @code{#pragma GCC} is of use for constructs that do not naturally form
2437 part of the grammar. @xref{Other Directives,,Miscellaneous
2438 Preprocessing Directives, cpp, The GNU C Preprocessor}.
2440 @node Attribute Syntax
2441 @section Attribute Syntax
2442 @cindex attribute syntax
2444 This section describes the syntax with which @code{__attribute__} may be
2445 used, and the constructs to which attribute specifiers bind, for the C
2446 language. Some details may vary for C++ and Objective-C@. Because of
2447 infelicities in the grammar for attributes, some forms described here
2448 may not be successfully parsed in all cases.
2450 There are some problems with the semantics of attributes in C++. For
2451 example, there are no manglings for attributes, although they may affect
2452 code generation, so problems may arise when attributed types are used in
2453 conjunction with templates or overloading. Similarly, @code{typeid}
2454 does not distinguish between types with different attributes. Support
2455 for attributes in C++ may be restricted in future to attributes on
2456 declarations only, but not on nested declarators.
2458 @xref{Function Attributes}, for details of the semantics of attributes
2459 applying to functions. @xref{Variable Attributes}, for details of the
2460 semantics of attributes applying to variables. @xref{Type Attributes},
2461 for details of the semantics of attributes applying to structure, union
2462 and enumerated types.
2464 An @dfn{attribute specifier} is of the form
2465 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2466 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2467 each attribute is one of the following:
2471 Empty. Empty attributes are ignored.
2474 A word (which may be an identifier such as @code{unused}, or a reserved
2475 word such as @code{const}).
2478 A word, followed by, in parentheses, parameters for the attribute.
2479 These parameters take one of the following forms:
2483 An identifier. For example, @code{mode} attributes use this form.
2486 An identifier followed by a comma and a non-empty comma-separated list
2487 of expressions. For example, @code{format} attributes use this form.
2490 A possibly empty comma-separated list of expressions. For example,
2491 @code{format_arg} attributes use this form with the list being a single
2492 integer constant expression, and @code{alias} attributes use this form
2493 with the list being a single string constant.
2497 An @dfn{attribute specifier list} is a sequence of one or more attribute
2498 specifiers, not separated by any other tokens.
2500 In GNU C, an attribute specifier list may appear after the colon following a
2501 label, other than a @code{case} or @code{default} label. The only
2502 attribute it makes sense to use after a label is @code{unused}. This
2503 feature is intended for code generated by programs which contains labels
2504 that may be unused but which is compiled with @option{-Wall}. It would
2505 not normally be appropriate to use in it human-written code, though it
2506 could be useful in cases where the code that jumps to the label is
2507 contained within an @code{#ifdef} conditional. GNU C++ does not permit
2508 such placement of attribute lists, as it is permissible for a
2509 declaration, which could begin with an attribute list, to be labelled in
2510 C++. Declarations cannot be labelled in C90 or C99, so the ambiguity
2511 does not arise there.
2513 An attribute specifier list may appear as part of a @code{struct},
2514 @code{union} or @code{enum} specifier. It may go either immediately
2515 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2516 the closing brace. It is ignored if the content of the structure, union
2517 or enumerated type is not defined in the specifier in which the
2518 attribute specifier list is used---that is, in usages such as
2519 @code{struct __attribute__((foo)) bar} with no following opening brace.
2520 Where attribute specifiers follow the closing brace, they are considered
2521 to relate to the structure, union or enumerated type defined, not to any
2522 enclosing declaration the type specifier appears in, and the type
2523 defined is not complete until after the attribute specifiers.
2524 @c Otherwise, there would be the following problems: a shift/reduce
2525 @c conflict between attributes binding the struct/union/enum and
2526 @c binding to the list of specifiers/qualifiers; and "aligned"
2527 @c attributes could use sizeof for the structure, but the size could be
2528 @c changed later by "packed" attributes.
2530 Otherwise, an attribute specifier appears as part of a declaration,
2531 counting declarations of unnamed parameters and type names, and relates
2532 to that declaration (which may be nested in another declaration, for
2533 example in the case of a parameter declaration), or to a particular declarator
2534 within a declaration. Where an
2535 attribute specifier is applied to a parameter declared as a function or
2536 an array, it should apply to the function or array rather than the
2537 pointer to which the parameter is implicitly converted, but this is not
2538 yet correctly implemented.
2540 Any list of specifiers and qualifiers at the start of a declaration may
2541 contain attribute specifiers, whether or not such a list may in that
2542 context contain storage class specifiers. (Some attributes, however,
2543 are essentially in the nature of storage class specifiers, and only make
2544 sense where storage class specifiers may be used; for example,
2545 @code{section}.) There is one necessary limitation to this syntax: the
2546 first old-style parameter declaration in a function definition cannot
2547 begin with an attribute specifier, because such an attribute applies to
2548 the function instead by syntax described below (which, however, is not
2549 yet implemented in this case). In some other cases, attribute
2550 specifiers are permitted by this grammar but not yet supported by the
2551 compiler. All attribute specifiers in this place relate to the
2552 declaration as a whole. In the obsolescent usage where a type of
2553 @code{int} is implied by the absence of type specifiers, such a list of
2554 specifiers and qualifiers may be an attribute specifier list with no
2555 other specifiers or qualifiers.
2557 At present, the first parameter in a function prototype must have some
2558 type specifier which is not an attribute specifier; this resolves an
2559 ambiguity in the interpretation of @code{void f(int
2560 (__attribute__((foo)) x))}, but is subject to change. At present, if
2561 the parentheses of a function declarator contain only attributes then
2562 those attributes are ignored, rather than yielding an error or warning
2563 or implying a single parameter of type int, but this is subject to
2566 An attribute specifier list may appear immediately before a declarator
2567 (other than the first) in a comma-separated list of declarators in a
2568 declaration of more than one identifier using a single list of
2569 specifiers and qualifiers. Such attribute specifiers apply
2570 only to the identifier before whose declarator they appear. For
2574 __attribute__((noreturn)) void d0 (void),
2575 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2580 the @code{noreturn} attribute applies to all the functions
2581 declared; the @code{format} attribute only applies to @code{d1}.
2583 An attribute specifier list may appear immediately before the comma,
2584 @code{=} or semicolon terminating the declaration of an identifier other
2585 than a function definition. At present, such attribute specifiers apply
2586 to the declared object or function, but in future they may attach to the
2587 outermost adjacent declarator. In simple cases there is no difference,
2588 but, for example, in
2591 void (****f)(void) __attribute__((noreturn));
2595 at present the @code{noreturn} attribute applies to @code{f}, which
2596 causes a warning since @code{f} is not a function, but in future it may
2597 apply to the function @code{****f}. The precise semantics of what
2598 attributes in such cases will apply to are not yet specified. Where an
2599 assembler name for an object or function is specified (@pxref{Asm
2600 Labels}), at present the attribute must follow the @code{asm}
2601 specification; in future, attributes before the @code{asm} specification
2602 may apply to the adjacent declarator, and those after it to the declared
2605 An attribute specifier list may, in future, be permitted to appear after
2606 the declarator in a function definition (before any old-style parameter
2607 declarations or the function body).
2609 Attribute specifiers may be mixed with type qualifiers appearing inside
2610 the @code{[]} of a parameter array declarator, in the C99 construct by
2611 which such qualifiers are applied to the pointer to which the array is
2612 implicitly converted. Such attribute specifiers apply to the pointer,
2613 not to the array, but at present this is not implemented and they are
2616 An attribute specifier list may appear at the start of a nested
2617 declarator. At present, there are some limitations in this usage: the
2618 attributes correctly apply to the declarator, but for most individual
2619 attributes the semantics this implies are not implemented.
2620 When attribute specifiers follow the @code{*} of a pointer
2621 declarator, they may be mixed with any type qualifiers present.
2622 The following describes the formal semantics of this syntax. It will make the
2623 most sense if you are familiar with the formal specification of
2624 declarators in the ISO C standard.
2626 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2627 D1}, where @code{T} contains declaration specifiers that specify a type
2628 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2629 contains an identifier @var{ident}. The type specified for @var{ident}
2630 for derived declarators whose type does not include an attribute
2631 specifier is as in the ISO C standard.
2633 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2634 and the declaration @code{T D} specifies the type
2635 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2636 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2637 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2639 If @code{D1} has the form @code{*
2640 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2641 declaration @code{T D} specifies the type
2642 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2643 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2644 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2650 void (__attribute__((noreturn)) ****f) (void);
2654 specifies the type ``pointer to pointer to pointer to pointer to
2655 non-returning function returning @code{void}''. As another example,
2658 char *__attribute__((aligned(8))) *f;
2662 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2663 Note again that this does not work with most attributes; for example,
2664 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2665 is not yet supported.
2667 For compatibility with existing code written for compiler versions that
2668 did not implement attributes on nested declarators, some laxity is
2669 allowed in the placing of attributes. If an attribute that only applies
2670 to types is applied to a declaration, it will be treated as applying to
2671 the type of that declaration. If an attribute that only applies to
2672 declarations is applied to the type of a declaration, it will be treated
2673 as applying to that declaration; and, for compatibility with code
2674 placing the attributes immediately before the identifier declared, such
2675 an attribute applied to a function return type will be treated as
2676 applying to the function type, and such an attribute applied to an array
2677 element type will be treated as applying to the array type. If an
2678 attribute that only applies to function types is applied to a
2679 pointer-to-function type, it will be treated as applying to the pointer
2680 target type; if such an attribute is applied to a function return type
2681 that is not a pointer-to-function type, it will be treated as applying
2682 to the function type.
2684 @node Function Prototypes
2685 @section Prototypes and Old-Style Function Definitions
2686 @cindex function prototype declarations
2687 @cindex old-style function definitions
2688 @cindex promotion of formal parameters
2690 GNU C extends ISO C to allow a function prototype to override a later
2691 old-style non-prototype definition. Consider the following example:
2694 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2701 /* @r{Prototype function declaration.} */
2702 int isroot P((uid_t));
2704 /* @r{Old-style function definition.} */
2706 isroot (x) /* @r{??? lossage here ???} */
2713 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2714 not allow this example, because subword arguments in old-style
2715 non-prototype definitions are promoted. Therefore in this example the
2716 function definition's argument is really an @code{int}, which does not
2717 match the prototype argument type of @code{short}.
2719 This restriction of ISO C makes it hard to write code that is portable
2720 to traditional C compilers, because the programmer does not know
2721 whether the @code{uid_t} type is @code{short}, @code{int}, or
2722 @code{long}. Therefore, in cases like these GNU C allows a prototype
2723 to override a later old-style definition. More precisely, in GNU C, a
2724 function prototype argument type overrides the argument type specified
2725 by a later old-style definition if the former type is the same as the
2726 latter type before promotion. Thus in GNU C the above example is
2727 equivalent to the following:
2740 GNU C++ does not support old-style function definitions, so this
2741 extension is irrelevant.
2744 @section C++ Style Comments
2746 @cindex C++ comments
2747 @cindex comments, C++ style
2749 In GNU C, you may use C++ style comments, which start with @samp{//} and
2750 continue until the end of the line. Many other C implementations allow
2751 such comments, and they are included in the 1999 C standard. However,
2752 C++ style comments are not recognized if you specify an @option{-std}
2753 option specifying a version of ISO C before C99, or @option{-ansi}
2754 (equivalent to @option{-std=c89}).
2757 @section Dollar Signs in Identifier Names
2759 @cindex dollar signs in identifier names
2760 @cindex identifier names, dollar signs in
2762 In GNU C, you may normally use dollar signs in identifier names.
2763 This is because many traditional C implementations allow such identifiers.
2764 However, dollar signs in identifiers are not supported on a few target
2765 machines, typically because the target assembler does not allow them.
2767 @node Character Escapes
2768 @section The Character @key{ESC} in Constants
2770 You can use the sequence @samp{\e} in a string or character constant to
2771 stand for the ASCII character @key{ESC}.
2774 @section Inquiring on Alignment of Types or Variables
2776 @cindex type alignment
2777 @cindex variable alignment
2779 The keyword @code{__alignof__} allows you to inquire about how an object
2780 is aligned, or the minimum alignment usually required by a type. Its
2781 syntax is just like @code{sizeof}.
2783 For example, if the target machine requires a @code{double} value to be
2784 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2785 This is true on many RISC machines. On more traditional machine
2786 designs, @code{__alignof__ (double)} is 4 or even 2.
2788 Some machines never actually require alignment; they allow reference to any
2789 data type even at an odd address. For these machines, @code{__alignof__}
2790 reports the @emph{recommended} alignment of a type.
2792 If the operand of @code{__alignof__} is an lvalue rather than a type,
2793 its value is the required alignment for its type, taking into account
2794 any minimum alignment specified with GCC's @code{__attribute__}
2795 extension (@pxref{Variable Attributes}). For example, after this
2799 struct foo @{ int x; char y; @} foo1;
2803 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
2804 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
2806 It is an error to ask for the alignment of an incomplete type.
2808 @node Variable Attributes
2809 @section Specifying Attributes of Variables
2810 @cindex attribute of variables
2811 @cindex variable attributes
2813 The keyword @code{__attribute__} allows you to specify special
2814 attributes of variables or structure fields. This keyword is followed
2815 by an attribute specification inside double parentheses. Some
2816 attributes are currently defined generically for variables.
2817 Other attributes are defined for variables on particular target
2818 systems. Other attributes are available for functions
2819 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
2820 Other front ends might define more attributes
2821 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2823 You may also specify attributes with @samp{__} preceding and following
2824 each keyword. This allows you to use them in header files without
2825 being concerned about a possible macro of the same name. For example,
2826 you may use @code{__aligned__} instead of @code{aligned}.
2828 @xref{Attribute Syntax}, for details of the exact syntax for using
2832 @cindex @code{aligned} attribute
2833 @item aligned (@var{alignment})
2834 This attribute specifies a minimum alignment for the variable or
2835 structure field, measured in bytes. For example, the declaration:
2838 int x __attribute__ ((aligned (16))) = 0;
2842 causes the compiler to allocate the global variable @code{x} on a
2843 16-byte boundary. On a 68040, this could be used in conjunction with
2844 an @code{asm} expression to access the @code{move16} instruction which
2845 requires 16-byte aligned operands.
2847 You can also specify the alignment of structure fields. For example, to
2848 create a double-word aligned @code{int} pair, you could write:
2851 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2855 This is an alternative to creating a union with a @code{double} member
2856 that forces the union to be double-word aligned.
2858 As in the preceding examples, you can explicitly specify the alignment
2859 (in bytes) that you wish the compiler to use for a given variable or
2860 structure field. Alternatively, you can leave out the alignment factor
2861 and just ask the compiler to align a variable or field to the maximum
2862 useful alignment for the target machine you are compiling for. For
2863 example, you could write:
2866 short array[3] __attribute__ ((aligned));
2869 Whenever you leave out the alignment factor in an @code{aligned} attribute
2870 specification, the compiler automatically sets the alignment for the declared
2871 variable or field to the largest alignment which is ever used for any data
2872 type on the target machine you are compiling for. Doing this can often make
2873 copy operations more efficient, because the compiler can use whatever
2874 instructions copy the biggest chunks of memory when performing copies to
2875 or from the variables or fields that you have aligned this way.
2877 The @code{aligned} attribute can only increase the alignment; but you
2878 can decrease it by specifying @code{packed} as well. See below.
2880 Note that the effectiveness of @code{aligned} attributes may be limited
2881 by inherent limitations in your linker. On many systems, the linker is
2882 only able to arrange for variables to be aligned up to a certain maximum
2883 alignment. (For some linkers, the maximum supported alignment may
2884 be very very small.) If your linker is only able to align variables
2885 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2886 in an @code{__attribute__} will still only provide you with 8 byte
2887 alignment. See your linker documentation for further information.
2889 @item cleanup (@var{cleanup_function})
2890 @cindex @code{cleanup} attribute
2891 The @code{cleanup} attribute runs a function when the variable goes
2892 out of scope. This attribute can only be applied to auto function
2893 scope variables; it may not be applied to parameters or variables
2894 with static storage duration. The function must take one parameter,
2895 a pointer to a type compatible with the variable. The return value
2896 of the function (if any) is ignored.
2898 If @option{-fexceptions} is enabled, then @var{cleanup_function}
2899 will be run during the stack unwinding that happens during the
2900 processing of the exception. Note that the @code{cleanup} attribute
2901 does not allow the exception to be caught, only to perform an action.
2902 It is undefined what happens if @var{cleanup_function} does not
2907 @cindex @code{common} attribute
2908 @cindex @code{nocommon} attribute
2911 The @code{common} attribute requests GCC to place a variable in
2912 ``common'' storage. The @code{nocommon} attribute requests the
2913 opposite---to allocate space for it directly.
2915 These attributes override the default chosen by the
2916 @option{-fno-common} and @option{-fcommon} flags respectively.
2919 @cindex @code{deprecated} attribute
2920 The @code{deprecated} attribute results in a warning if the variable
2921 is used anywhere in the source file. This is useful when identifying
2922 variables that are expected to be removed in a future version of a
2923 program. The warning also includes the location of the declaration
2924 of the deprecated variable, to enable users to easily find further
2925 information about why the variable is deprecated, or what they should
2926 do instead. Note that the warning only occurs for uses:
2929 extern int old_var __attribute__ ((deprecated));
2931 int new_fn () @{ return old_var; @}
2934 results in a warning on line 3 but not line 2.
2936 The @code{deprecated} attribute can also be used for functions and
2937 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
2939 @item mode (@var{mode})
2940 @cindex @code{mode} attribute
2941 This attribute specifies the data type for the declaration---whichever
2942 type corresponds to the mode @var{mode}. This in effect lets you
2943 request an integer or floating point type according to its width.
2945 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2946 indicate the mode corresponding to a one-byte integer, @samp{word} or
2947 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2948 or @samp{__pointer__} for the mode used to represent pointers.
2951 @cindex @code{packed} attribute
2952 The @code{packed} attribute specifies that a variable or structure field
2953 should have the smallest possible alignment---one byte for a variable,
2954 and one bit for a field, unless you specify a larger value with the
2955 @code{aligned} attribute.
2957 Here is a structure in which the field @code{x} is packed, so that it
2958 immediately follows @code{a}:
2964 int x[2] __attribute__ ((packed));
2968 @item section ("@var{section-name}")
2969 @cindex @code{section} variable attribute
2970 Normally, the compiler places the objects it generates in sections like
2971 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
2972 or you need certain particular variables to appear in special sections,
2973 for example to map to special hardware. The @code{section}
2974 attribute specifies that a variable (or function) lives in a particular
2975 section. For example, this small program uses several specific section names:
2978 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
2979 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
2980 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
2981 int init_data __attribute__ ((section ("INITDATA"))) = 0;
2985 /* @r{Initialize stack pointer} */
2986 init_sp (stack + sizeof (stack));
2988 /* @r{Initialize initialized data} */
2989 memcpy (&init_data, &data, &edata - &data);
2991 /* @r{Turn on the serial ports} */
2998 Use the @code{section} attribute with an @emph{initialized} definition
2999 of a @emph{global} variable, as shown in the example. GCC issues
3000 a warning and otherwise ignores the @code{section} attribute in
3001 uninitialized variable declarations.
3003 You may only use the @code{section} attribute with a fully initialized
3004 global definition because of the way linkers work. The linker requires
3005 each object be defined once, with the exception that uninitialized
3006 variables tentatively go in the @code{common} (or @code{bss}) section
3007 and can be multiply ``defined''. You can force a variable to be
3008 initialized with the @option{-fno-common} flag or the @code{nocommon}
3011 Some file formats do not support arbitrary sections so the @code{section}
3012 attribute is not available on all platforms.
3013 If you need to map the entire contents of a module to a particular
3014 section, consider using the facilities of the linker instead.
3017 @cindex @code{shared} variable attribute
3018 On Microsoft Windows, in addition to putting variable definitions in a named
3019 section, the section can also be shared among all running copies of an
3020 executable or DLL@. For example, this small program defines shared data
3021 by putting it in a named section @code{shared} and marking the section
3025 int foo __attribute__((section ("shared"), shared)) = 0;
3030 /* @r{Read and write foo. All running
3031 copies see the same value.} */
3037 You may only use the @code{shared} attribute along with @code{section}
3038 attribute with a fully initialized global definition because of the way
3039 linkers work. See @code{section} attribute for more information.
3041 The @code{shared} attribute is only available on Microsoft Windows@.
3043 @item tls_model ("@var{tls_model}")
3044 @cindex @code{tls_model} attribute
3045 The @code{tls_model} attribute sets thread-local storage model
3046 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
3047 overriding @option{-ftls-model=} command line switch on a per-variable
3049 The @var{tls_model} argument should be one of @code{global-dynamic},
3050 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
3052 Not all targets support this attribute.
3055 This attribute, attached to a variable, means that the variable is meant
3056 to be possibly unused. GCC will not produce a warning for this
3059 @item vector_size (@var{bytes})
3060 This attribute specifies the vector size for the variable, measured in
3061 bytes. For example, the declaration:
3064 int foo __attribute__ ((vector_size (16)));
3068 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3069 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3070 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3072 This attribute is only applicable to integral and float scalars,
3073 although arrays, pointers, and function return values are allowed in
3074 conjunction with this construct.
3076 Aggregates with this attribute are invalid, even if they are of the same
3077 size as a corresponding scalar. For example, the declaration:
3080 struct S @{ int a; @};
3081 struct S __attribute__ ((vector_size (16))) foo;
3085 is invalid even if the size of the structure is the same as the size of
3089 The @code{selectany} attribute causes an initialized global variable to
3090 have link-once semantics. When multiple definitions of the variable are
3091 encountered by the linker, the first is selected and the remainder are
3092 discarded. Following usage by the Microsoft compiler, the linker is told
3093 @emph{not} to warn about size or content differences of the multiple
3096 Although the primary usage of this attribute is for POD types, the
3097 attribute can also be applied to global C++ objects that are initialized
3098 by a constructor. In this case, the static initialization and destruction
3099 code for the object is emitted in each translation defining the object,
3100 but the calls to the constructor and destructor are protected by a
3101 link-once guard variable.
3103 The @code{selectany} attribute is only available on Microsoft Windows
3104 targets. You can use @code{__declspec (selectany)} as a synonym for
3105 @code{__attribute__ ((selectany))} for compatibility with other
3109 The @code{weak} attribute is described in @xref{Function Attributes}.
3112 The @code{dllimport} attribute is described in @xref{Function Attributes}.
3115 The @code{dllexport} attribute is described in @xref{Function Attributes}.
3119 @subsection M32R/D Variable Attributes
3121 One attribute is currently defined for the M32R/D@.
3124 @item model (@var{model-name})
3125 @cindex variable addressability on the M32R/D
3126 Use this attribute on the M32R/D to set the addressability of an object.
3127 The identifier @var{model-name} is one of @code{small}, @code{medium},
3128 or @code{large}, representing each of the code models.
3130 Small model objects live in the lower 16MB of memory (so that their
3131 addresses can be loaded with the @code{ld24} instruction).
3133 Medium and large model objects may live anywhere in the 32-bit address space
3134 (the compiler will generate @code{seth/add3} instructions to load their
3138 @subsection i386 Variable Attributes
3140 Two attributes are currently defined for i386 configurations:
3141 @code{ms_struct} and @code{gcc_struct}
3146 @cindex @code{ms_struct} attribute
3147 @cindex @code{gcc_struct} attribute
3149 If @code{packed} is used on a structure, or if bit-fields are used
3150 it may be that the Microsoft ABI packs them differently
3151 than GCC would normally pack them. Particularly when moving packed
3152 data between functions compiled with GCC and the native Microsoft compiler
3153 (either via function call or as data in a file), it may be necessary to access
3156 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3157 compilers to match the native Microsoft compiler.
3160 @subsection Xstormy16 Variable Attributes
3162 One attribute is currently defined for xstormy16 configurations:
3167 @cindex @code{below100} attribute
3169 If a variable has the @code{below100} attribute (@code{BELOW100} is
3170 allowed also), GCC will place the variable in the first 0x100 bytes of
3171 memory and use special opcodes to access it. Such variables will be
3172 placed in either the @code{.bss_below100} section or the
3173 @code{.data_below100} section.
3177 @node Type Attributes
3178 @section Specifying Attributes of Types
3179 @cindex attribute of types
3180 @cindex type attributes
3182 The keyword @code{__attribute__} allows you to specify special
3183 attributes of @code{struct} and @code{union} types when you define such
3184 types. This keyword is followed by an attribute specification inside
3185 double parentheses. Six attributes are currently defined for types:
3186 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
3187 @code{deprecated} and @code{may_alias}. Other attributes are defined for
3188 functions (@pxref{Function Attributes}) and for variables
3189 (@pxref{Variable Attributes}).
3191 You may also specify any one of these attributes with @samp{__}
3192 preceding and following its keyword. This allows you to use these
3193 attributes in header files without being concerned about a possible
3194 macro of the same name. For example, you may use @code{__aligned__}
3195 instead of @code{aligned}.
3197 You may specify the @code{aligned} and @code{transparent_union}
3198 attributes either in a @code{typedef} declaration or just past the
3199 closing curly brace of a complete enum, struct or union type
3200 @emph{definition} and the @code{packed} attribute only past the closing
3201 brace of a definition.
3203 You may also specify attributes between the enum, struct or union
3204 tag and the name of the type rather than after the closing brace.
3206 @xref{Attribute Syntax}, for details of the exact syntax for using
3210 @cindex @code{aligned} attribute
3211 @item aligned (@var{alignment})
3212 This attribute specifies a minimum alignment (in bytes) for variables
3213 of the specified type. For example, the declarations:
3216 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3217 typedef int more_aligned_int __attribute__ ((aligned (8)));
3221 force the compiler to insure (as far as it can) that each variable whose
3222 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3223 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
3224 variables of type @code{struct S} aligned to 8-byte boundaries allows
3225 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3226 store) instructions when copying one variable of type @code{struct S} to
3227 another, thus improving run-time efficiency.
3229 Note that the alignment of any given @code{struct} or @code{union} type
3230 is required by the ISO C standard to be at least a perfect multiple of
3231 the lowest common multiple of the alignments of all of the members of
3232 the @code{struct} or @code{union} in question. This means that you @emph{can}
3233 effectively adjust the alignment of a @code{struct} or @code{union}
3234 type by attaching an @code{aligned} attribute to any one of the members
3235 of such a type, but the notation illustrated in the example above is a
3236 more obvious, intuitive, and readable way to request the compiler to
3237 adjust the alignment of an entire @code{struct} or @code{union} type.
3239 As in the preceding example, you can explicitly specify the alignment
3240 (in bytes) that you wish the compiler to use for a given @code{struct}
3241 or @code{union} type. Alternatively, you can leave out the alignment factor
3242 and just ask the compiler to align a type to the maximum
3243 useful alignment for the target machine you are compiling for. For
3244 example, you could write:
3247 struct S @{ short f[3]; @} __attribute__ ((aligned));
3250 Whenever you leave out the alignment factor in an @code{aligned}
3251 attribute specification, the compiler automatically sets the alignment
3252 for the type to the largest alignment which is ever used for any data
3253 type on the target machine you are compiling for. Doing this can often
3254 make copy operations more efficient, because the compiler can use
3255 whatever instructions copy the biggest chunks of memory when performing
3256 copies to or from the variables which have types that you have aligned
3259 In the example above, if the size of each @code{short} is 2 bytes, then
3260 the size of the entire @code{struct S} type is 6 bytes. The smallest
3261 power of two which is greater than or equal to that is 8, so the
3262 compiler sets the alignment for the entire @code{struct S} type to 8
3265 Note that although you can ask the compiler to select a time-efficient
3266 alignment for a given type and then declare only individual stand-alone
3267 objects of that type, the compiler's ability to select a time-efficient
3268 alignment is primarily useful only when you plan to create arrays of
3269 variables having the relevant (efficiently aligned) type. If you
3270 declare or use arrays of variables of an efficiently-aligned type, then
3271 it is likely that your program will also be doing pointer arithmetic (or
3272 subscripting, which amounts to the same thing) on pointers to the
3273 relevant type, and the code that the compiler generates for these
3274 pointer arithmetic operations will often be more efficient for
3275 efficiently-aligned types than for other types.
3277 The @code{aligned} attribute can only increase the alignment; but you
3278 can decrease it by specifying @code{packed} as well. See below.
3280 Note that the effectiveness of @code{aligned} attributes may be limited
3281 by inherent limitations in your linker. On many systems, the linker is
3282 only able to arrange for variables to be aligned up to a certain maximum
3283 alignment. (For some linkers, the maximum supported alignment may
3284 be very very small.) If your linker is only able to align variables
3285 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3286 in an @code{__attribute__} will still only provide you with 8 byte
3287 alignment. See your linker documentation for further information.
3290 This attribute, attached to @code{struct} or @code{union} type
3291 definition, specifies that each member (other than zero-width bitfields)
3292 of the structure or union is placed to minimize the memory required. When
3293 attached to an @code{enum} definition, it indicates that the smallest
3294 integral type should be used.
3296 @opindex fshort-enums
3297 Specifying this attribute for @code{struct} and @code{union} types is
3298 equivalent to specifying the @code{packed} attribute on each of the
3299 structure or union members. Specifying the @option{-fshort-enums}
3300 flag on the line is equivalent to specifying the @code{packed}
3301 attribute on all @code{enum} definitions.
3303 In the following example @code{struct my_packed_struct}'s members are
3304 packed closely together, but the internal layout of its @code{s} member
3305 is not packed---to do that, @code{struct my_unpacked_struct} would need to
3309 struct my_unpacked_struct
3315 struct __attribute__ ((__packed__)) my_packed_struct
3319 struct my_unpacked_struct s;
3323 You may only specify this attribute on the definition of a @code{enum},
3324 @code{struct} or @code{union}, not on a @code{typedef} which does not
3325 also define the enumerated type, structure or union.
3327 @item transparent_union
3328 This attribute, attached to a @code{union} type definition, indicates
3329 that any function parameter having that union type causes calls to that
3330 function to be treated in a special way.
3332 First, the argument corresponding to a transparent union type can be of
3333 any type in the union; no cast is required. Also, if the union contains
3334 a pointer type, the corresponding argument can be a null pointer
3335 constant or a void pointer expression; and if the union contains a void
3336 pointer type, the corresponding argument can be any pointer expression.
3337 If the union member type is a pointer, qualifiers like @code{const} on
3338 the referenced type must be respected, just as with normal pointer
3341 Second, the argument is passed to the function using the calling
3342 conventions of the first member of the transparent union, not the calling
3343 conventions of the union itself. All members of the union must have the
3344 same machine representation; this is necessary for this argument passing
3347 Transparent unions are designed for library functions that have multiple
3348 interfaces for compatibility reasons. For example, suppose the
3349 @code{wait} function must accept either a value of type @code{int *} to
3350 comply with Posix, or a value of type @code{union wait *} to comply with
3351 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3352 @code{wait} would accept both kinds of arguments, but it would also
3353 accept any other pointer type and this would make argument type checking
3354 less useful. Instead, @code{<sys/wait.h>} might define the interface
3362 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3364 pid_t wait (wait_status_ptr_t);
3367 This interface allows either @code{int *} or @code{union wait *}
3368 arguments to be passed, using the @code{int *} calling convention.
3369 The program can call @code{wait} with arguments of either type:
3372 int w1 () @{ int w; return wait (&w); @}
3373 int w2 () @{ union wait w; return wait (&w); @}
3376 With this interface, @code{wait}'s implementation might look like this:
3379 pid_t wait (wait_status_ptr_t p)
3381 return waitpid (-1, p.__ip, 0);
3386 When attached to a type (including a @code{union} or a @code{struct}),
3387 this attribute means that variables of that type are meant to appear
3388 possibly unused. GCC will not produce a warning for any variables of
3389 that type, even if the variable appears to do nothing. This is often
3390 the case with lock or thread classes, which are usually defined and then
3391 not referenced, but contain constructors and destructors that have
3392 nontrivial bookkeeping functions.
3395 The @code{deprecated} attribute results in a warning if the type
3396 is used anywhere in the source file. This is useful when identifying
3397 types that are expected to be removed in a future version of a program.
3398 If possible, the warning also includes the location of the declaration
3399 of the deprecated type, to enable users to easily find further
3400 information about why the type is deprecated, or what they should do
3401 instead. Note that the warnings only occur for uses and then only
3402 if the type is being applied to an identifier that itself is not being
3403 declared as deprecated.
3406 typedef int T1 __attribute__ ((deprecated));
3410 typedef T1 T3 __attribute__ ((deprecated));
3411 T3 z __attribute__ ((deprecated));
3414 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3415 warning is issued for line 4 because T2 is not explicitly
3416 deprecated. Line 5 has no warning because T3 is explicitly
3417 deprecated. Similarly for line 6.
3419 The @code{deprecated} attribute can also be used for functions and
3420 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3423 Accesses to objects with types with this attribute are not subjected to
3424 type-based alias analysis, but are instead assumed to be able to alias
3425 any other type of objects, just like the @code{char} type. See
3426 @option{-fstrict-aliasing} for more information on aliasing issues.
3431 typedef short __attribute__((__may_alias__)) short_a;
3437 short_a *b = (short_a *) &a;
3441 if (a == 0x12345678)
3448 If you replaced @code{short_a} with @code{short} in the variable
3449 declaration, the above program would abort when compiled with
3450 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3451 above in recent GCC versions.
3453 @subsection ARM Type Attributes
3455 On those ARM targets that support @code{dllimport} (such as Symbian
3456 OS), you can use the @code{notshared} attribute to indicate that the
3457 virtual table and other similar data for a class should not be
3458 exported from a DLL@. For example:
3461 class __declspec(notshared) C @{
3463 __declspec(dllimport) C();
3467 __declspec(dllexport)
3471 In this code, @code{C::C} is exported from the current DLL, but the
3472 virtual table for @code{C} is not exported. (You can use
3473 @code{__attribute__} instead of @code{__declspec} if you prefer, but
3474 most Symbian OS code uses @code{__declspec}.)
3476 @subsection i386 Type Attributes
3478 Two attributes are currently defined for i386 configurations:
3479 @code{ms_struct} and @code{gcc_struct}
3483 @cindex @code{ms_struct}
3484 @cindex @code{gcc_struct}
3486 If @code{packed} is used on a structure, or if bit-fields are used
3487 it may be that the Microsoft ABI packs them differently
3488 than GCC would normally pack them. Particularly when moving packed
3489 data between functions compiled with GCC and the native Microsoft compiler
3490 (either via function call or as data in a file), it may be necessary to access
3493 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3494 compilers to match the native Microsoft compiler.
3497 To specify multiple attributes, separate them by commas within the
3498 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3502 @section An Inline Function is As Fast As a Macro
3503 @cindex inline functions
3504 @cindex integrating function code
3506 @cindex macros, inline alternative
3508 By declaring a function @code{inline}, you can direct GCC to
3509 integrate that function's code into the code for its callers. This
3510 makes execution faster by eliminating the function-call overhead; in
3511 addition, if any of the actual argument values are constant, their known
3512 values may permit simplifications at compile time so that not all of the
3513 inline function's code needs to be included. The effect on code size is
3514 less predictable; object code may be larger or smaller with function
3515 inlining, depending on the particular case. Inlining of functions is an
3516 optimization and it really ``works'' only in optimizing compilation. If
3517 you don't use @option{-O}, no function is really inline.
3519 Inline functions are included in the ISO C99 standard, but there are
3520 currently substantial differences between what GCC implements and what
3521 the ISO C99 standard requires.
3523 To declare a function inline, use the @code{inline} keyword in its
3524 declaration, like this:
3534 (If you are writing a header file to be included in ISO C programs, write
3535 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3536 You can also make all ``simple enough'' functions inline with the option
3537 @option{-finline-functions}.
3540 Note that certain usages in a function definition can make it unsuitable
3541 for inline substitution. Among these usages are: use of varargs, use of
3542 alloca, use of variable sized data types (@pxref{Variable Length}),
3543 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3544 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3545 will warn when a function marked @code{inline} could not be substituted,
3546 and will give the reason for the failure.
3548 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3549 does not affect the linkage of the function.
3551 @cindex automatic @code{inline} for C++ member fns
3552 @cindex @code{inline} automatic for C++ member fns
3553 @cindex member fns, automatically @code{inline}
3554 @cindex C++ member fns, automatically @code{inline}
3555 @opindex fno-default-inline
3556 GCC automatically inlines member functions defined within the class
3557 body of C++ programs even if they are not explicitly declared
3558 @code{inline}. (You can override this with @option{-fno-default-inline};
3559 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3561 @cindex inline functions, omission of
3562 @opindex fkeep-inline-functions
3563 When a function is both inline and @code{static}, if all calls to the
3564 function are integrated into the caller, and the function's address is
3565 never used, then the function's own assembler code is never referenced.
3566 In this case, GCC does not actually output assembler code for the
3567 function, unless you specify the option @option{-fkeep-inline-functions}.
3568 Some calls cannot be integrated for various reasons (in particular,
3569 calls that precede the function's definition cannot be integrated, and
3570 neither can recursive calls within the definition). If there is a
3571 nonintegrated call, then the function is compiled to assembler code as
3572 usual. The function must also be compiled as usual if the program
3573 refers to its address, because that can't be inlined.
3575 @cindex non-static inline function
3576 When an inline function is not @code{static}, then the compiler must assume
3577 that there may be calls from other source files; since a global symbol can
3578 be defined only once in any program, the function must not be defined in
3579 the other source files, so the calls therein cannot be integrated.
3580 Therefore, a non-@code{static} inline function is always compiled on its
3581 own in the usual fashion.
3583 If you specify both @code{inline} and @code{extern} in the function
3584 definition, then the definition is used only for inlining. In no case
3585 is the function compiled on its own, not even if you refer to its
3586 address explicitly. Such an address becomes an external reference, as
3587 if you had only declared the function, and had not defined it.
3589 This combination of @code{inline} and @code{extern} has almost the
3590 effect of a macro. The way to use it is to put a function definition in
3591 a header file with these keywords, and put another copy of the
3592 definition (lacking @code{inline} and @code{extern}) in a library file.
3593 The definition in the header file will cause most calls to the function
3594 to be inlined. If any uses of the function remain, they will refer to
3595 the single copy in the library.
3597 Since GCC eventually will implement ISO C99 semantics for
3598 inline functions, it is best to use @code{static inline} only
3599 to guarantee compatibility. (The
3600 existing semantics will remain available when @option{-std=gnu89} is
3601 specified, but eventually the default will be @option{-std=gnu99} and
3602 that will implement the C99 semantics, though it does not do so yet.)
3604 GCC does not inline any functions when not optimizing unless you specify
3605 the @samp{always_inline} attribute for the function, like this:
3608 /* @r{Prototype.} */
3609 inline void foo (const char) __attribute__((always_inline));
3613 @section Assembler Instructions with C Expression Operands
3614 @cindex extended @code{asm}
3615 @cindex @code{asm} expressions
3616 @cindex assembler instructions
3619 In an assembler instruction using @code{asm}, you can specify the
3620 operands of the instruction using C expressions. This means you need not
3621 guess which registers or memory locations will contain the data you want
3624 You must specify an assembler instruction template much like what
3625 appears in a machine description, plus an operand constraint string for
3628 For example, here is how to use the 68881's @code{fsinx} instruction:
3631 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3635 Here @code{angle} is the C expression for the input operand while
3636 @code{result} is that of the output operand. Each has @samp{"f"} as its
3637 operand constraint, saying that a floating point register is required.
3638 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3639 output operands' constraints must use @samp{=}. The constraints use the
3640 same language used in the machine description (@pxref{Constraints}).
3642 Each operand is described by an operand-constraint string followed by
3643 the C expression in parentheses. A colon separates the assembler
3644 template from the first output operand and another separates the last
3645 output operand from the first input, if any. Commas separate the
3646 operands within each group. The total number of operands is currently
3647 limited to 30; this limitation may be lifted in some future version of
3650 If there are no output operands but there are input operands, you must
3651 place two consecutive colons surrounding the place where the output
3654 As of GCC version 3.1, it is also possible to specify input and output
3655 operands using symbolic names which can be referenced within the
3656 assembler code. These names are specified inside square brackets
3657 preceding the constraint string, and can be referenced inside the
3658 assembler code using @code{%[@var{name}]} instead of a percentage sign
3659 followed by the operand number. Using named operands the above example
3663 asm ("fsinx %[angle],%[output]"
3664 : [output] "=f" (result)
3665 : [angle] "f" (angle));
3669 Note that the symbolic operand names have no relation whatsoever to
3670 other C identifiers. You may use any name you like, even those of
3671 existing C symbols, but you must ensure that no two operands within the same
3672 assembler construct use the same symbolic name.
3674 Output operand expressions must be lvalues; the compiler can check this.
3675 The input operands need not be lvalues. The compiler cannot check
3676 whether the operands have data types that are reasonable for the
3677 instruction being executed. It does not parse the assembler instruction
3678 template and does not know what it means or even whether it is valid
3679 assembler input. The extended @code{asm} feature is most often used for
3680 machine instructions the compiler itself does not know exist. If
3681 the output expression cannot be directly addressed (for example, it is a
3682 bit-field), your constraint must allow a register. In that case, GCC
3683 will use the register as the output of the @code{asm}, and then store
3684 that register into the output.
3686 The ordinary output operands must be write-only; GCC will assume that
3687 the values in these operands before the instruction are dead and need
3688 not be generated. Extended asm supports input-output or read-write
3689 operands. Use the constraint character @samp{+} to indicate such an
3690 operand and list it with the output operands. You should only use
3691 read-write operands when the constraints for the operand (or the
3692 operand in which only some of the bits are to be changed) allow a
3695 You may, as an alternative, logically split its function into two
3696 separate operands, one input operand and one write-only output
3697 operand. The connection between them is expressed by constraints
3698 which say they need to be in the same location when the instruction
3699 executes. You can use the same C expression for both operands, or
3700 different expressions. For example, here we write the (fictitious)
3701 @samp{combine} instruction with @code{bar} as its read-only source
3702 operand and @code{foo} as its read-write destination:
3705 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3709 The constraint @samp{"0"} for operand 1 says that it must occupy the
3710 same location as operand 0. A number in constraint is allowed only in
3711 an input operand and it must refer to an output operand.
3713 Only a number in the constraint can guarantee that one operand will be in
3714 the same place as another. The mere fact that @code{foo} is the value
3715 of both operands is not enough to guarantee that they will be in the
3716 same place in the generated assembler code. The following would not
3720 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3723 Various optimizations or reloading could cause operands 0 and 1 to be in
3724 different registers; GCC knows no reason not to do so. For example, the
3725 compiler might find a copy of the value of @code{foo} in one register and
3726 use it for operand 1, but generate the output operand 0 in a different
3727 register (copying it afterward to @code{foo}'s own address). Of course,
3728 since the register for operand 1 is not even mentioned in the assembler
3729 code, the result will not work, but GCC can't tell that.
3731 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3732 the operand number for a matching constraint. For example:
3735 asm ("cmoveq %1,%2,%[result]"
3736 : [result] "=r"(result)
3737 : "r" (test), "r"(new), "[result]"(old));
3740 Sometimes you need to make an @code{asm} operand be a specific register,
3741 but there's no matching constraint letter for that register @emph{by
3742 itself}. To force the operand into that register, use a local variable
3743 for the operand and specify the register in the variable declaration.
3744 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
3745 register constraint letter that matches the register:
3748 register int *p1 asm ("r0") = @dots{};
3749 register int *p2 asm ("r1") = @dots{};
3750 register int *result asm ("r0");
3751 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
3754 @anchor{Example of asm with clobbered asm reg}
3755 In the above example, beware that a register that is call-clobbered by
3756 the target ABI will be overwritten by any function call in the
3757 assignment, including library calls for arithmetic operators.
3758 Assuming it is a call-clobbered register, this may happen to @code{r0}
3759 above by the assignment to @code{p2}. If you have to use such a
3760 register, use temporary variables for expressions between the register
3765 register int *p1 asm ("r0") = @dots{};
3766 register int *p2 asm ("r1") = t1;
3767 register int *result asm ("r0");
3768 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
3771 Some instructions clobber specific hard registers. To describe this,
3772 write a third colon after the input operands, followed by the names of
3773 the clobbered hard registers (given as strings). Here is a realistic
3774 example for the VAX:
3777 asm volatile ("movc3 %0,%1,%2"
3778 : /* @r{no outputs} */
3779 : "g" (from), "g" (to), "g" (count)
3780 : "r0", "r1", "r2", "r3", "r4", "r5");
3783 You may not write a clobber description in a way that overlaps with an
3784 input or output operand. For example, you may not have an operand
3785 describing a register class with one member if you mention that register
3786 in the clobber list. Variables declared to live in specific registers
3787 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
3788 have no part mentioned in the clobber description.
3789 There is no way for you to specify that an input
3790 operand is modified without also specifying it as an output
3791 operand. Note that if all the output operands you specify are for this
3792 purpose (and hence unused), you will then also need to specify
3793 @code{volatile} for the @code{asm} construct, as described below, to
3794 prevent GCC from deleting the @code{asm} statement as unused.
3796 If you refer to a particular hardware register from the assembler code,
3797 you will probably have to list the register after the third colon to
3798 tell the compiler the register's value is modified. In some assemblers,
3799 the register names begin with @samp{%}; to produce one @samp{%} in the
3800 assembler code, you must write @samp{%%} in the input.
3802 If your assembler instruction can alter the condition code register, add
3803 @samp{cc} to the list of clobbered registers. GCC on some machines
3804 represents the condition codes as a specific hardware register;
3805 @samp{cc} serves to name this register. On other machines, the
3806 condition code is handled differently, and specifying @samp{cc} has no
3807 effect. But it is valid no matter what the machine.
3809 If your assembler instructions access memory in an unpredictable
3810 fashion, add @samp{memory} to the list of clobbered registers. This
3811 will cause GCC to not keep memory values cached in registers across the
3812 assembler instruction and not optimize stores or loads to that memory.
3813 You will also want to add the @code{volatile} keyword if the memory
3814 affected is not listed in the inputs or outputs of the @code{asm}, as
3815 the @samp{memory} clobber does not count as a side-effect of the
3816 @code{asm}. If you know how large the accessed memory is, you can add
3817 it as input or output but if this is not known, you should add
3818 @samp{memory}. As an example, if you access ten bytes of a string, you
3819 can use a memory input like:
3822 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
3825 Note that in the following example the memory input is necessary,
3826 otherwise GCC might optimize the store to @code{x} away:
3833 asm ("magic stuff accessing an 'int' pointed to by '%1'"
3834 "=&d" (r) : "a" (y), "m" (*y));
3839 You can put multiple assembler instructions together in a single
3840 @code{asm} template, separated by the characters normally used in assembly
3841 code for the system. A combination that works in most places is a newline
3842 to break the line, plus a tab character to move to the instruction field
3843 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3844 assembler allows semicolons as a line-breaking character. Note that some
3845 assembler dialects use semicolons to start a comment.
3846 The input operands are guaranteed not to use any of the clobbered
3847 registers, and neither will the output operands' addresses, so you can
3848 read and write the clobbered registers as many times as you like. Here
3849 is an example of multiple instructions in a template; it assumes the
3850 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3853 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3855 : "g" (from), "g" (to)
3859 Unless an output operand has the @samp{&} constraint modifier, GCC
3860 may allocate it in the same register as an unrelated input operand, on
3861 the assumption the inputs are consumed before the outputs are produced.
3862 This assumption may be false if the assembler code actually consists of
3863 more than one instruction. In such a case, use @samp{&} for each output
3864 operand that may not overlap an input. @xref{Modifiers}.
3866 If you want to test the condition code produced by an assembler
3867 instruction, you must include a branch and a label in the @code{asm}
3868 construct, as follows:
3871 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3877 This assumes your assembler supports local labels, as the GNU assembler
3878 and most Unix assemblers do.
3880 Speaking of labels, jumps from one @code{asm} to another are not
3881 supported. The compiler's optimizers do not know about these jumps, and
3882 therefore they cannot take account of them when deciding how to
3885 @cindex macros containing @code{asm}
3886 Usually the most convenient way to use these @code{asm} instructions is to
3887 encapsulate them in macros that look like functions. For example,
3891 (@{ double __value, __arg = (x); \
3892 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3897 Here the variable @code{__arg} is used to make sure that the instruction
3898 operates on a proper @code{double} value, and to accept only those
3899 arguments @code{x} which can convert automatically to a @code{double}.
3901 Another way to make sure the instruction operates on the correct data
3902 type is to use a cast in the @code{asm}. This is different from using a
3903 variable @code{__arg} in that it converts more different types. For
3904 example, if the desired type were @code{int}, casting the argument to
3905 @code{int} would accept a pointer with no complaint, while assigning the
3906 argument to an @code{int} variable named @code{__arg} would warn about
3907 using a pointer unless the caller explicitly casts it.
3909 If an @code{asm} has output operands, GCC assumes for optimization
3910 purposes the instruction has no side effects except to change the output
3911 operands. This does not mean instructions with a side effect cannot be
3912 used, but you must be careful, because the compiler may eliminate them
3913 if the output operands aren't used, or move them out of loops, or
3914 replace two with one if they constitute a common subexpression. Also,
3915 if your instruction does have a side effect on a variable that otherwise
3916 appears not to change, the old value of the variable may be reused later
3917 if it happens to be found in a register.
3919 You can prevent an @code{asm} instruction from being deleted
3920 by writing the keyword @code{volatile} after
3921 the @code{asm}. For example:
3924 #define get_and_set_priority(new) \
3926 asm volatile ("get_and_set_priority %0, %1" \
3927 : "=g" (__old) : "g" (new)); \
3932 The @code{volatile} keyword indicates that the instruction has
3933 important side-effects. GCC will not delete a volatile @code{asm} if
3934 it is reachable. (The instruction can still be deleted if GCC can
3935 prove that control-flow will never reach the location of the
3936 instruction.) Note that even a volatile @code{asm} instruction
3937 can be moved relative to other code, including across jump
3938 instructions. For example, on many targets there is a system
3939 register which can be set to control the rounding mode of
3940 floating point operations. You might try
3941 setting it with a volatile @code{asm}, like this PowerPC example:
3944 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
3949 This will not work reliably, as the compiler may move the addition back
3950 before the volatile @code{asm}. To make it work you need to add an
3951 artificial dependency to the @code{asm} referencing a variable in the code
3952 you don't want moved, for example:
3955 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
3959 Similarly, you can't expect a
3960 sequence of volatile @code{asm} instructions to remain perfectly
3961 consecutive. If you want consecutive output, use a single @code{asm}.
3962 Also, GCC will perform some optimizations across a volatile @code{asm}
3963 instruction; GCC does not ``forget everything'' when it encounters
3964 a volatile @code{asm} instruction the way some other compilers do.
3966 An @code{asm} instruction without any output operands will be treated
3967 identically to a volatile @code{asm} instruction.
3969 It is a natural idea to look for a way to give access to the condition
3970 code left by the assembler instruction. However, when we attempted to
3971 implement this, we found no way to make it work reliably. The problem
3972 is that output operands might need reloading, which would result in
3973 additional following ``store'' instructions. On most machines, these
3974 instructions would alter the condition code before there was time to
3975 test it. This problem doesn't arise for ordinary ``test'' and
3976 ``compare'' instructions because they don't have any output operands.
3978 For reasons similar to those described above, it is not possible to give
3979 an assembler instruction access to the condition code left by previous
3982 If you are writing a header file that should be includable in ISO C
3983 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3986 @subsection Size of an @code{asm}
3988 Some targets require that GCC track the size of each instruction used in
3989 order to generate correct code. Because the final length of an
3990 @code{asm} is only known by the assembler, GCC must make an estimate as
3991 to how big it will be. The estimate is formed by counting the number of
3992 statements in the pattern of the @code{asm} and multiplying that by the
3993 length of the longest instruction on that processor. Statements in the
3994 @code{asm} are identified by newline characters and whatever statement
3995 separator characters are supported by the assembler; on most processors
3996 this is the `@code{;}' character.
3998 Normally, GCC's estimate is perfectly adequate to ensure that correct
3999 code is generated, but it is possible to confuse the compiler if you use
4000 pseudo instructions or assembler macros that expand into multiple real
4001 instructions or if you use assembler directives that expand to more
4002 space in the object file than would be needed for a single instruction.
4003 If this happens then the assembler will produce a diagnostic saying that
4004 a label is unreachable.
4006 @subsection i386 floating point asm operands
4008 There are several rules on the usage of stack-like regs in
4009 asm_operands insns. These rules apply only to the operands that are
4014 Given a set of input regs that die in an asm_operands, it is
4015 necessary to know which are implicitly popped by the asm, and
4016 which must be explicitly popped by gcc.
4018 An input reg that is implicitly popped by the asm must be
4019 explicitly clobbered, unless it is constrained to match an
4023 For any input reg that is implicitly popped by an asm, it is
4024 necessary to know how to adjust the stack to compensate for the pop.
4025 If any non-popped input is closer to the top of the reg-stack than
4026 the implicitly popped reg, it would not be possible to know what the
4027 stack looked like---it's not clear how the rest of the stack ``slides
4030 All implicitly popped input regs must be closer to the top of
4031 the reg-stack than any input that is not implicitly popped.
4033 It is possible that if an input dies in an insn, reload might
4034 use the input reg for an output reload. Consider this example:
4037 asm ("foo" : "=t" (a) : "f" (b));
4040 This asm says that input B is not popped by the asm, and that
4041 the asm pushes a result onto the reg-stack, i.e., the stack is one
4042 deeper after the asm than it was before. But, it is possible that
4043 reload will think that it can use the same reg for both the input and
4044 the output, if input B dies in this insn.
4046 If any input operand uses the @code{f} constraint, all output reg
4047 constraints must use the @code{&} earlyclobber.
4049 The asm above would be written as
4052 asm ("foo" : "=&t" (a) : "f" (b));
4056 Some operands need to be in particular places on the stack. All
4057 output operands fall in this category---there is no other way to
4058 know which regs the outputs appear in unless the user indicates
4059 this in the constraints.
4061 Output operands must specifically indicate which reg an output
4062 appears in after an asm. @code{=f} is not allowed: the operand
4063 constraints must select a class with a single reg.
4066 Output operands may not be ``inserted'' between existing stack regs.
4067 Since no 387 opcode uses a read/write operand, all output operands
4068 are dead before the asm_operands, and are pushed by the asm_operands.
4069 It makes no sense to push anywhere but the top of the reg-stack.
4071 Output operands must start at the top of the reg-stack: output
4072 operands may not ``skip'' a reg.
4075 Some asm statements may need extra stack space for internal
4076 calculations. This can be guaranteed by clobbering stack registers
4077 unrelated to the inputs and outputs.
4081 Here are a couple of reasonable asms to want to write. This asm
4082 takes one input, which is internally popped, and produces two outputs.
4085 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
4088 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
4089 and replaces them with one output. The user must code the @code{st(1)}
4090 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
4093 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
4099 @section Controlling Names Used in Assembler Code
4100 @cindex assembler names for identifiers
4101 @cindex names used in assembler code
4102 @cindex identifiers, names in assembler code
4104 You can specify the name to be used in the assembler code for a C
4105 function or variable by writing the @code{asm} (or @code{__asm__})
4106 keyword after the declarator as follows:
4109 int foo asm ("myfoo") = 2;
4113 This specifies that the name to be used for the variable @code{foo} in
4114 the assembler code should be @samp{myfoo} rather than the usual
4117 On systems where an underscore is normally prepended to the name of a C
4118 function or variable, this feature allows you to define names for the
4119 linker that do not start with an underscore.
4121 It does not make sense to use this feature with a non-static local
4122 variable since such variables do not have assembler names. If you are
4123 trying to put the variable in a particular register, see @ref{Explicit
4124 Reg Vars}. GCC presently accepts such code with a warning, but will
4125 probably be changed to issue an error, rather than a warning, in the
4128 You cannot use @code{asm} in this way in a function @emph{definition}; but
4129 you can get the same effect by writing a declaration for the function
4130 before its definition and putting @code{asm} there, like this:
4133 extern func () asm ("FUNC");
4140 It is up to you to make sure that the assembler names you choose do not
4141 conflict with any other assembler symbols. Also, you must not use a
4142 register name; that would produce completely invalid assembler code. GCC
4143 does not as yet have the ability to store static variables in registers.
4144 Perhaps that will be added.
4146 @node Explicit Reg Vars
4147 @section Variables in Specified Registers
4148 @cindex explicit register variables
4149 @cindex variables in specified registers
4150 @cindex specified registers
4151 @cindex registers, global allocation
4153 GNU C allows you to put a few global variables into specified hardware
4154 registers. You can also specify the register in which an ordinary
4155 register variable should be allocated.
4159 Global register variables reserve registers throughout the program.
4160 This may be useful in programs such as programming language
4161 interpreters which have a couple of global variables that are accessed
4165 Local register variables in specific registers do not reserve the
4166 registers, except at the point where they are used as input or output
4167 operands in an @code{asm} statement and the @code{asm} statement itself is
4168 not deleted. The compiler's data flow analysis is capable of determining
4169 where the specified registers contain live values, and where they are
4170 available for other uses. Stores into local register variables may be deleted
4171 when they appear to be dead according to dataflow analysis. References
4172 to local register variables may be deleted or moved or simplified.
4174 These local variables are sometimes convenient for use with the extended
4175 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
4176 output of the assembler instruction directly into a particular register.
4177 (This will work provided the register you specify fits the constraints
4178 specified for that operand in the @code{asm}.)
4186 @node Global Reg Vars
4187 @subsection Defining Global Register Variables
4188 @cindex global register variables
4189 @cindex registers, global variables in
4191 You can define a global register variable in GNU C like this:
4194 register int *foo asm ("a5");
4198 Here @code{a5} is the name of the register which should be used. Choose a
4199 register which is normally saved and restored by function calls on your
4200 machine, so that library routines will not clobber it.
4202 Naturally the register name is cpu-dependent, so you would need to
4203 conditionalize your program according to cpu type. The register
4204 @code{a5} would be a good choice on a 68000 for a variable of pointer
4205 type. On machines with register windows, be sure to choose a ``global''
4206 register that is not affected magically by the function call mechanism.
4208 In addition, operating systems on one type of cpu may differ in how they
4209 name the registers; then you would need additional conditionals. For
4210 example, some 68000 operating systems call this register @code{%a5}.
4212 Eventually there may be a way of asking the compiler to choose a register
4213 automatically, but first we need to figure out how it should choose and
4214 how to enable you to guide the choice. No solution is evident.
4216 Defining a global register variable in a certain register reserves that
4217 register entirely for this use, at least within the current compilation.
4218 The register will not be allocated for any other purpose in the functions
4219 in the current compilation. The register will not be saved and restored by
4220 these functions. Stores into this register are never deleted even if they
4221 would appear to be dead, but references may be deleted or moved or
4224 It is not safe to access the global register variables from signal
4225 handlers, or from more than one thread of control, because the system
4226 library routines may temporarily use the register for other things (unless
4227 you recompile them specially for the task at hand).
4229 @cindex @code{qsort}, and global register variables
4230 It is not safe for one function that uses a global register variable to
4231 call another such function @code{foo} by way of a third function
4232 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4233 different source file in which the variable wasn't declared). This is
4234 because @code{lose} might save the register and put some other value there.
4235 For example, you can't expect a global register variable to be available in
4236 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4237 might have put something else in that register. (If you are prepared to
4238 recompile @code{qsort} with the same global register variable, you can
4239 solve this problem.)
4241 If you want to recompile @code{qsort} or other source files which do not
4242 actually use your global register variable, so that they will not use that
4243 register for any other purpose, then it suffices to specify the compiler
4244 option @option{-ffixed-@var{reg}}. You need not actually add a global
4245 register declaration to their source code.
4247 A function which can alter the value of a global register variable cannot
4248 safely be called from a function compiled without this variable, because it
4249 could clobber the value the caller expects to find there on return.
4250 Therefore, the function which is the entry point into the part of the
4251 program that uses the global register variable must explicitly save and
4252 restore the value which belongs to its caller.
4254 @cindex register variable after @code{longjmp}
4255 @cindex global register after @code{longjmp}
4256 @cindex value after @code{longjmp}
4259 On most machines, @code{longjmp} will restore to each global register
4260 variable the value it had at the time of the @code{setjmp}. On some
4261 machines, however, @code{longjmp} will not change the value of global
4262 register variables. To be portable, the function that called @code{setjmp}
4263 should make other arrangements to save the values of the global register
4264 variables, and to restore them in a @code{longjmp}. This way, the same
4265 thing will happen regardless of what @code{longjmp} does.
4267 All global register variable declarations must precede all function
4268 definitions. If such a declaration could appear after function
4269 definitions, the declaration would be too late to prevent the register from
4270 being used for other purposes in the preceding functions.
4272 Global register variables may not have initial values, because an
4273 executable file has no means to supply initial contents for a register.
4275 On the SPARC, there are reports that g3 @dots{} g7 are suitable
4276 registers, but certain library functions, such as @code{getwd}, as well
4277 as the subroutines for division and remainder, modify g3 and g4. g1 and
4278 g2 are local temporaries.
4280 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4281 Of course, it will not do to use more than a few of those.
4283 @node Local Reg Vars
4284 @subsection Specifying Registers for Local Variables
4285 @cindex local variables, specifying registers
4286 @cindex specifying registers for local variables
4287 @cindex registers for local variables
4289 You can define a local register variable with a specified register
4293 register int *foo asm ("a5");
4297 Here @code{a5} is the name of the register which should be used. Note
4298 that this is the same syntax used for defining global register
4299 variables, but for a local variable it would appear within a function.
4301 Naturally the register name is cpu-dependent, but this is not a
4302 problem, since specific registers are most often useful with explicit
4303 assembler instructions (@pxref{Extended Asm}). Both of these things
4304 generally require that you conditionalize your program according to
4307 In addition, operating systems on one type of cpu may differ in how they
4308 name the registers; then you would need additional conditionals. For
4309 example, some 68000 operating systems call this register @code{%a5}.
4311 Defining such a register variable does not reserve the register; it
4312 remains available for other uses in places where flow control determines
4313 the variable's value is not live.
4315 This option does not guarantee that GCC will generate code that has
4316 this variable in the register you specify at all times. You may not
4317 code an explicit reference to this register in the @emph{assembler
4318 instruction template} part of an @code{asm} statement and assume it will
4319 always refer to this variable. However, using the variable as an
4320 @code{asm} @emph{operand} guarantees that the specified register is used
4323 Stores into local register variables may be deleted when they appear to be dead
4324 according to dataflow analysis. References to local register variables may
4325 be deleted or moved or simplified.
4327 As for global register variables, it's recommended that you choose a
4328 register which is normally saved and restored by function calls on
4329 your machine, so that library routines will not clobber it. A common
4330 pitfall is to initialize multiple call-clobbered registers with
4331 arbitrary expressions, where a function call or library call for an
4332 arithmetic operator will overwrite a register value from a previous
4333 assignment, for example @code{r0} below:
4335 register int *p1 asm ("r0") = @dots{};
4336 register int *p2 asm ("r1") = @dots{};
4338 In those cases, a solution is to use a temporary variable for
4339 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
4341 @node Alternate Keywords
4342 @section Alternate Keywords
4343 @cindex alternate keywords
4344 @cindex keywords, alternate
4346 @option{-ansi} and the various @option{-std} options disable certain
4347 keywords. This causes trouble when you want to use GNU C extensions, or
4348 a general-purpose header file that should be usable by all programs,
4349 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4350 @code{inline} are not available in programs compiled with
4351 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4352 program compiled with @option{-std=c99}). The ISO C99 keyword
4353 @code{restrict} is only available when @option{-std=gnu99} (which will
4354 eventually be the default) or @option{-std=c99} (or the equivalent
4355 @option{-std=iso9899:1999}) is used.
4357 The way to solve these problems is to put @samp{__} at the beginning and
4358 end of each problematical keyword. For example, use @code{__asm__}
4359 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4361 Other C compilers won't accept these alternative keywords; if you want to
4362 compile with another compiler, you can define the alternate keywords as
4363 macros to replace them with the customary keywords. It looks like this:
4371 @findex __extension__
4373 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4375 prevent such warnings within one expression by writing
4376 @code{__extension__} before the expression. @code{__extension__} has no
4377 effect aside from this.
4379 @node Incomplete Enums
4380 @section Incomplete @code{enum} Types
4382 You can define an @code{enum} tag without specifying its possible values.
4383 This results in an incomplete type, much like what you get if you write
4384 @code{struct foo} without describing the elements. A later declaration
4385 which does specify the possible values completes the type.
4387 You can't allocate variables or storage using the type while it is
4388 incomplete. However, you can work with pointers to that type.
4390 This extension may not be very useful, but it makes the handling of
4391 @code{enum} more consistent with the way @code{struct} and @code{union}
4394 This extension is not supported by GNU C++.
4396 @node Function Names
4397 @section Function Names as Strings
4398 @cindex @code{__func__} identifier
4399 @cindex @code{__FUNCTION__} identifier
4400 @cindex @code{__PRETTY_FUNCTION__} identifier
4402 GCC provides three magic variables which hold the name of the current
4403 function, as a string. The first of these is @code{__func__}, which
4404 is part of the C99 standard:
4407 The identifier @code{__func__} is implicitly declared by the translator
4408 as if, immediately following the opening brace of each function
4409 definition, the declaration
4412 static const char __func__[] = "function-name";
4415 appeared, where function-name is the name of the lexically-enclosing
4416 function. This name is the unadorned name of the function.
4419 @code{__FUNCTION__} is another name for @code{__func__}. Older
4420 versions of GCC recognize only this name. However, it is not
4421 standardized. For maximum portability, we recommend you use
4422 @code{__func__}, but provide a fallback definition with the
4426 #if __STDC_VERSION__ < 199901L
4428 # define __func__ __FUNCTION__
4430 # define __func__ "<unknown>"
4435 In C, @code{__PRETTY_FUNCTION__} is yet another name for
4436 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
4437 the type signature of the function as well as its bare name. For
4438 example, this program:
4442 extern int printf (char *, ...);
4449 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4450 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4468 __PRETTY_FUNCTION__ = void a::sub(int)
4471 These identifiers are not preprocessor macros. In GCC 3.3 and
4472 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
4473 were treated as string literals; they could be used to initialize
4474 @code{char} arrays, and they could be concatenated with other string
4475 literals. GCC 3.4 and later treat them as variables, like
4476 @code{__func__}. In C++, @code{__FUNCTION__} and
4477 @code{__PRETTY_FUNCTION__} have always been variables.
4479 @node Return Address
4480 @section Getting the Return or Frame Address of a Function
4482 These functions may be used to get information about the callers of a
4485 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4486 This function returns the return address of the current function, or of
4487 one of its callers. The @var{level} argument is number of frames to
4488 scan up the call stack. A value of @code{0} yields the return address
4489 of the current function, a value of @code{1} yields the return address
4490 of the caller of the current function, and so forth. When inlining
4491 the expected behavior is that the function will return the address of
4492 the function that will be returned to. To work around this behavior use
4493 the @code{noinline} function attribute.
4495 The @var{level} argument must be a constant integer.
4497 On some machines it may be impossible to determine the return address of
4498 any function other than the current one; in such cases, or when the top
4499 of the stack has been reached, this function will return @code{0} or a
4500 random value. In addition, @code{__builtin_frame_address} may be used
4501 to determine if the top of the stack has been reached.
4503 This function should only be used with a nonzero argument for debugging
4507 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4508 This function is similar to @code{__builtin_return_address}, but it
4509 returns the address of the function frame rather than the return address
4510 of the function. Calling @code{__builtin_frame_address} with a value of
4511 @code{0} yields the frame address of the current function, a value of
4512 @code{1} yields the frame address of the caller of the current function,
4515 The frame is the area on the stack which holds local variables and saved
4516 registers. The frame address is normally the address of the first word
4517 pushed on to the stack by the function. However, the exact definition
4518 depends upon the processor and the calling convention. If the processor
4519 has a dedicated frame pointer register, and the function has a frame,
4520 then @code{__builtin_frame_address} will return the value of the frame
4523 On some machines it may be impossible to determine the frame address of
4524 any function other than the current one; in such cases, or when the top
4525 of the stack has been reached, this function will return @code{0} if
4526 the first frame pointer is properly initialized by the startup code.
4528 This function should only be used with a nonzero argument for debugging
4532 @node Vector Extensions
4533 @section Using vector instructions through built-in functions
4535 On some targets, the instruction set contains SIMD vector instructions that
4536 operate on multiple values contained in one large register at the same time.
4537 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4540 The first step in using these extensions is to provide the necessary data
4541 types. This should be done using an appropriate @code{typedef}:
4544 typedef int v4si __attribute__ ((vector_size (16)));
4547 The @code{int} type specifies the base type, while the attribute specifies
4548 the vector size for the variable, measured in bytes. For example, the
4549 declaration above causes the compiler to set the mode for the @code{v4si}
4550 type to be 16 bytes wide and divided into @code{int} sized units. For
4551 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
4552 corresponding mode of @code{foo} will be @acronym{V4SI}.
4554 The @code{vector_size} attribute is only applicable to integral and
4555 float scalars, although arrays, pointers, and function return values
4556 are allowed in conjunction with this construct.
4558 All the basic integer types can be used as base types, both as signed
4559 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
4560 @code{long long}. In addition, @code{float} and @code{double} can be
4561 used to build floating-point vector types.
4563 Specifying a combination that is not valid for the current architecture
4564 will cause GCC to synthesize the instructions using a narrower mode.
4565 For example, if you specify a variable of type @code{V4SI} and your
4566 architecture does not allow for this specific SIMD type, GCC will
4567 produce code that uses 4 @code{SIs}.
4569 The types defined in this manner can be used with a subset of normal C
4570 operations. Currently, GCC will allow using the following operators
4571 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~}@.
4573 The operations behave like C++ @code{valarrays}. Addition is defined as
4574 the addition of the corresponding elements of the operands. For
4575 example, in the code below, each of the 4 elements in @var{a} will be
4576 added to the corresponding 4 elements in @var{b} and the resulting
4577 vector will be stored in @var{c}.
4580 typedef int v4si __attribute__ ((vector_size (16)));
4587 Subtraction, multiplication, division, and the logical operations
4588 operate in a similar manner. Likewise, the result of using the unary
4589 minus or complement operators on a vector type is a vector whose
4590 elements are the negative or complemented values of the corresponding
4591 elements in the operand.
4593 You can declare variables and use them in function calls and returns, as
4594 well as in assignments and some casts. You can specify a vector type as
4595 a return type for a function. Vector types can also be used as function
4596 arguments. It is possible to cast from one vector type to another,
4597 provided they are of the same size (in fact, you can also cast vectors
4598 to and from other datatypes of the same size).
4600 You cannot operate between vectors of different lengths or different
4601 signedness without a cast.
4603 A port that supports hardware vector operations, usually provides a set
4604 of built-in functions that can be used to operate on vectors. For
4605 example, a function to add two vectors and multiply the result by a
4606 third could look like this:
4609 v4si f (v4si a, v4si b, v4si c)
4611 v4si tmp = __builtin_addv4si (a, b);
4612 return __builtin_mulv4si (tmp, c);
4619 @findex __builtin_offsetof
4621 GCC implements for both C and C++ a syntactic extension to implement
4622 the @code{offsetof} macro.
4626 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
4628 offsetof_member_designator:
4630 | offsetof_member_designator "." @code{identifier}
4631 | offsetof_member_designator "[" @code{expr} "]"
4634 This extension is sufficient such that
4637 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
4640 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
4641 may be dependent. In either case, @var{member} may consist of a single
4642 identifier, or a sequence of member accesses and array references.
4644 @node Atomic Builtins
4645 @section Built-in functions for atomic memory access
4647 The following builtins are intended to be compatible with those described
4648 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
4649 section 7.4. As such, they depart from the normal GCC practice of using
4650 the ``__builtin_'' prefix, and further that they are overloaded such that
4651 they work on multiple types.
4653 The definition given in the Intel documentation allows only for the use of
4654 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
4655 counterparts. GCC will allow any integral scalar or pointer type that is
4656 1, 2, 4 or 8 bytes in length.
4658 Not all operations are supported by all target processors. If a particular
4659 operation cannot be implemented on the target processor, a warning will be
4660 generated and a call an external function will be generated. The external
4661 function will carry the same name as the builtin, with an additional suffix
4662 @samp{_@var{n}} where @var{n} is the size of the data type.
4664 @c ??? Should we have a mechanism to suppress this warning? This is almost
4665 @c useful for implementing the operation under the control of an external
4668 In most cases, these builtins are considered a @dfn{full barrier}. That is,
4669 no memory operand will be moved across the operation, either forward or
4670 backward. Further, instructions will be issued as necessary to prevent the
4671 processor from speculating loads across the operation and from queuing stores
4672 after the operation.
4674 All of the routines are are described in the Intel documentation to take
4675 ``an optional list of variables protected by the memory barrier''. It's
4676 not clear what is meant by that; it could mean that @emph{only} the
4677 following variables are protected, or it could mean that these variables
4678 should in addition be protected. At present GCC ignores this list and
4679 protects all variables which are globally accessible. If in the future
4680 we make some use of this list, an empty list will continue to mean all
4681 globally accessible variables.
4684 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
4685 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
4686 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
4687 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
4688 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
4689 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
4690 @findex __sync_fetch_and_add
4691 @findex __sync_fetch_and_sub
4692 @findex __sync_fetch_and_or
4693 @findex __sync_fetch_and_and
4694 @findex __sync_fetch_and_xor
4695 @findex __sync_fetch_and_nand
4696 These builtins perform the operation suggested by the name, and
4697 returns the value that had previously been in memory. That is,
4700 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
4701 @{ tmp = *ptr; *ptr = ~tmp & value; return tmp; @} // nand
4704 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
4705 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
4706 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
4707 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
4708 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
4709 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
4710 @findex __sync_add_and_fetch
4711 @findex __sync_sub_and_fetch
4712 @findex __sync_or_and_fetch
4713 @findex __sync_and_and_fetch
4714 @findex __sync_xor_and_fetch
4715 @findex __sync_nand_and_fetch
4716 These builtins perform the operation suggested by the name, and
4717 return the new value. That is,
4720 @{ *ptr @var{op}= value; return *ptr; @}
4721 @{ *ptr = ~*ptr & value; return *ptr; @} // nand
4724 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
4725 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
4726 @findex __sync_bool_compare_and_swap
4727 @findex __sync_val_compare_and_swap
4728 These builtins perform an atomic compare and swap. That is, if the current
4729 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
4732 The ``bool'' version returns true if the comparison is successful and
4733 @var{newval} was written. The ``val'' version returns the contents
4734 of @code{*@var{ptr}} before the operation.
4736 @item __sync_synchronize (...)
4737 @findex __sync_synchronize
4738 This builtin issues a full memory barrier.
4740 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
4741 @findex __sync_lock_test_and_set
4742 This builtin, as described by Intel, is not a traditional test-and-set
4743 operation, but rather an atomic exchange operation. It writes @var{value}
4744 into @code{*@var{ptr}}, and returns the previous contents of
4747 Many targets have only minimal support for such locks, and do not support
4748 a full exchange operation. In this case, a target may support reduced
4749 functionality here by which the @emph{only} valid value to store is the
4750 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
4751 is implementation defined.
4753 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
4754 This means that references after the builtin cannot move to (or be
4755 speculated to) before the builtin, but previous memory stores may not
4756 be globally visible yet, and previous memory loads may not yet be
4759 @item void __sync_lock_release (@var{type} *ptr, ...)
4760 @findex __sync_lock_release
4761 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
4762 Normally this means writing the constant 0 to @code{*@var{ptr}}.
4764 This builtin is not a full barrier, but rather a @dfn{release barrier}.
4765 This means that all previous memory stores are globally visible, and all
4766 previous memory loads have been satisfied, but following memory reads
4767 are not prevented from being speculated to before the barrier.
4770 @node Object Size Checking
4771 @section Object Size Checking Builtins
4772 @findex __builtin_object_size
4773 @findex __builtin___memcpy_chk
4774 @findex __builtin___mempcpy_chk
4775 @findex __builtin___memmove_chk
4776 @findex __builtin___memset_chk
4777 @findex __builtin___strcpy_chk
4778 @findex __builtin___stpcpy_chk
4779 @findex __builtin___strncpy_chk
4780 @findex __builtin___strcat_chk
4781 @findex __builtin___strncat_chk
4782 @findex __builtin___sprintf_chk
4783 @findex __builtin___snprintf_chk
4784 @findex __builtin___vsprintf_chk
4785 @findex __builtin___vsnprintf_chk
4786 @findex __builtin___printf_chk
4787 @findex __builtin___vprintf_chk
4788 @findex __builtin___fprintf_chk
4789 @findex __builtin___vfprintf_chk
4791 GCC implements a limited buffer overflow protection mechanism
4792 that can prevent some buffer overflow attacks.
4794 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
4795 is a built-in construct that returns a constant number of bytes from
4796 @var{ptr} to the end of the object @var{ptr} pointer points to
4797 (if known at compile time). @code{__builtin_object_size} never evaluates
4798 its arguments for side-effects. If there are any side-effects in them, it
4799 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
4800 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
4801 point to and all of them are known at compile time, the returned number
4802 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
4803 0 and minimum if nonzero. If it is not possible to determine which objects
4804 @var{ptr} points to at compile time, @code{__builtin_object_size} should
4805 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
4806 for @var{type} 2 or 3.
4808 @var{type} is an integer constant from 0 to 3. If the least significant
4809 bit is clear, objects are whole variables, if it is set, a closest
4810 surrounding subobject is considered the object a pointer points to.
4811 The second bit determines if maximum or minimum of remaining bytes
4815 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
4816 char *p = &var.buf1[1], *q = &var.b;
4818 /* Here the object p points to is var. */
4819 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
4820 /* The subobject p points to is var.buf1. */
4821 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
4822 /* The object q points to is var. */
4823 assert (__builtin_object_size (q, 0)
4824 == (char *) (&var + 1) - (char *) &var.b);
4825 /* The subobject q points to is var.b. */
4826 assert (__builtin_object_size (q, 1) == sizeof (var.b));
4830 There are built-in functions added for many common string operation
4831 functions, e.g. for @code{memcpy} @code{__builtin___memcpy_chk}
4832 built-in is provided. This built-in has an additional last argument,
4833 which is the number of bytes remaining in object the @var{dest}
4834 argument points to or @code{(size_t) -1} if the size is not known.
4836 The built-in functions are optimized into the normal string functions
4837 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
4838 it is known at compile time that the destination object will not
4839 be overflown. If the compiler can determine at compile time the
4840 object will be always overflown, it issues a warning.
4842 The intended use can be e.g.
4846 #define bos0(dest) __builtin_object_size (dest, 0)
4847 #define memcpy(dest, src, n) \
4848 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
4852 /* It is unknown what object p points to, so this is optimized
4853 into plain memcpy - no checking is possible. */
4854 memcpy (p, "abcde", n);
4855 /* Destination is known and length too. It is known at compile
4856 time there will be no overflow. */
4857 memcpy (&buf[5], "abcde", 5);
4858 /* Destination is known, but the length is not known at compile time.
4859 This will result in __memcpy_chk call that can check for overflow
4861 memcpy (&buf[5], "abcde", n);
4862 /* Destination is known and it is known at compile time there will
4863 be overflow. There will be a warning and __memcpy_chk call that
4864 will abort the program at runtime. */
4865 memcpy (&buf[6], "abcde", 5);
4868 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
4869 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
4870 @code{strcat} and @code{strncat}.
4872 There are also checking built-in functions for formatted output functions.
4874 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
4875 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
4876 const char *fmt, ...);
4877 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
4879 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
4880 const char *fmt, va_list ap);
4883 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
4884 etc. functions and can contain implementation specific flags on what
4885 additional security measures the checking function might take, such as
4886 handling @code{%n} differently.
4888 The @var{os} argument is the object size @var{s} points to, like in the
4889 other built-in functions. There is a small difference in the behavior
4890 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
4891 optimized into the non-checking functions only if @var{flag} is 0, otherwise
4892 the checking function is called with @var{os} argument set to
4895 In addition to this, there are checking built-in functions
4896 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
4897 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
4898 These have just one additional argument, @var{flag}, right before
4899 format string @var{fmt}. If the compiler is able to optimize them to
4900 @code{fputc} etc. functions, it will, otherwise the checking function
4901 should be called and the @var{flag} argument passed to it.
4903 @node Other Builtins
4904 @section Other built-in functions provided by GCC
4905 @cindex built-in functions
4906 @findex __builtin_isgreater
4907 @findex __builtin_isgreaterequal
4908 @findex __builtin_isless
4909 @findex __builtin_islessequal
4910 @findex __builtin_islessgreater
4911 @findex __builtin_isunordered
4912 @findex __builtin_powi
4913 @findex __builtin_powif
4914 @findex __builtin_powil
5072 @findex fprintf_unlocked
5074 @findex fputs_unlocked
5184 @findex printf_unlocked
5213 @findex significandf
5214 @findex significandl
5285 GCC provides a large number of built-in functions other than the ones
5286 mentioned above. Some of these are for internal use in the processing
5287 of exceptions or variable-length argument lists and will not be
5288 documented here because they may change from time to time; we do not
5289 recommend general use of these functions.
5291 The remaining functions are provided for optimization purposes.
5293 @opindex fno-builtin
5294 GCC includes built-in versions of many of the functions in the standard
5295 C library. The versions prefixed with @code{__builtin_} will always be
5296 treated as having the same meaning as the C library function even if you
5297 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
5298 Many of these functions are only optimized in certain cases; if they are
5299 not optimized in a particular case, a call to the library function will
5304 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
5305 @option{-std=c99}), the functions
5306 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
5307 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
5308 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
5309 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked}, @code{fputs_unlocked},
5310 @code{gammaf}, @code{gammal}, @code{gamma}, @code{gettext},
5311 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
5312 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
5313 @code{mempcpy}, @code{pow10f}, @code{pow10l}, @code{pow10},
5314 @code{printf_unlocked}, @code{rindex}, @code{scalbf}, @code{scalbl},
5315 @code{scalb}, @code{signbit}, @code{signbitf}, @code{signbitl},
5316 @code{significandf}, @code{significandl}, @code{significand},
5317 @code{sincosf}, @code{sincosl}, @code{sincos}, @code{stpcpy},
5318 @code{stpncpy}, @code{strcasecmp}, @code{strdup}, @code{strfmon},
5319 @code{strncasecmp}, @code{strndup}, @code{toascii}, @code{y0f},
5320 @code{y0l}, @code{y0}, @code{y1f}, @code{y1l}, @code{y1}, @code{ynf},
5321 @code{ynl} and @code{yn}
5322 may be handled as built-in functions.
5323 All these functions have corresponding versions
5324 prefixed with @code{__builtin_}, which may be used even in strict C89
5327 The ISO C99 functions
5328 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
5329 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
5330 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
5331 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
5332 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
5333 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
5334 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
5335 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
5336 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
5337 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
5338 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
5339 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
5340 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
5341 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
5342 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
5343 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
5344 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
5345 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
5346 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
5347 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
5348 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
5349 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
5350 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
5351 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
5352 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
5353 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
5354 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
5355 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
5356 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
5357 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
5358 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
5359 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
5360 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
5361 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
5362 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
5363 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
5364 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
5365 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
5366 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
5367 are handled as built-in functions
5368 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5370 There are also built-in versions of the ISO C99 functions
5371 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
5372 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
5373 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
5374 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
5375 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
5376 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
5377 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
5378 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
5379 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
5380 that are recognized in any mode since ISO C90 reserves these names for
5381 the purpose to which ISO C99 puts them. All these functions have
5382 corresponding versions prefixed with @code{__builtin_}.
5384 The ISO C94 functions
5385 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
5386 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
5387 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
5389 are handled as built-in functions
5390 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5392 The ISO C90 functions
5393 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
5394 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
5395 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
5396 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
5397 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
5398 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
5399 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
5400 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
5401 @code{malloc}, @code{memcmp}, @code{memcpy}, @code{memset}, @code{modf},
5402 @code{pow}, @code{printf}, @code{putchar}, @code{puts}, @code{scanf},
5403 @code{sinh}, @code{sin}, @code{snprintf}, @code{sprintf}, @code{sqrt},
5404 @code{sscanf}, @code{strcat}, @code{strchr}, @code{strcmp},
5405 @code{strcpy}, @code{strcspn}, @code{strlen}, @code{strncat},
5406 @code{strncmp}, @code{strncpy}, @code{strpbrk}, @code{strrchr},
5407 @code{strspn}, @code{strstr}, @code{tanh}, @code{tan}, @code{vfprintf},
5408 @code{vprintf} and @code{vsprintf}
5409 are all recognized as built-in functions unless
5410 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
5411 is specified for an individual function). All of these functions have
5412 corresponding versions prefixed with @code{__builtin_}.
5414 GCC provides built-in versions of the ISO C99 floating point comparison
5415 macros that avoid raising exceptions for unordered operands. They have
5416 the same names as the standard macros ( @code{isgreater},
5417 @code{isgreaterequal}, @code{isless}, @code{islessequal},
5418 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
5419 prefixed. We intend for a library implementor to be able to simply
5420 @code{#define} each standard macro to its built-in equivalent.
5422 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
5424 You can use the built-in function @code{__builtin_types_compatible_p} to
5425 determine whether two types are the same.
5427 This built-in function returns 1 if the unqualified versions of the
5428 types @var{type1} and @var{type2} (which are types, not expressions) are
5429 compatible, 0 otherwise. The result of this built-in function can be
5430 used in integer constant expressions.
5432 This built-in function ignores top level qualifiers (e.g., @code{const},
5433 @code{volatile}). For example, @code{int} is equivalent to @code{const
5436 The type @code{int[]} and @code{int[5]} are compatible. On the other
5437 hand, @code{int} and @code{char *} are not compatible, even if the size
5438 of their types, on the particular architecture are the same. Also, the
5439 amount of pointer indirection is taken into account when determining
5440 similarity. Consequently, @code{short *} is not similar to
5441 @code{short **}. Furthermore, two types that are typedefed are
5442 considered compatible if their underlying types are compatible.
5444 An @code{enum} type is not considered to be compatible with another
5445 @code{enum} type even if both are compatible with the same integer
5446 type; this is what the C standard specifies.
5447 For example, @code{enum @{foo, bar@}} is not similar to
5448 @code{enum @{hot, dog@}}.
5450 You would typically use this function in code whose execution varies
5451 depending on the arguments' types. For example:
5457 if (__builtin_types_compatible_p (typeof (x), long double)) \
5458 tmp = foo_long_double (tmp); \
5459 else if (__builtin_types_compatible_p (typeof (x), double)) \
5460 tmp = foo_double (tmp); \
5461 else if (__builtin_types_compatible_p (typeof (x), float)) \
5462 tmp = foo_float (tmp); \
5469 @emph{Note:} This construct is only available for C@.
5473 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
5475 You can use the built-in function @code{__builtin_choose_expr} to
5476 evaluate code depending on the value of a constant expression. This
5477 built-in function returns @var{exp1} if @var{const_exp}, which is a
5478 constant expression that must be able to be determined at compile time,
5479 is nonzero. Otherwise it returns 0.
5481 This built-in function is analogous to the @samp{? :} operator in C,
5482 except that the expression returned has its type unaltered by promotion
5483 rules. Also, the built-in function does not evaluate the expression
5484 that was not chosen. For example, if @var{const_exp} evaluates to true,
5485 @var{exp2} is not evaluated even if it has side-effects.
5487 This built-in function can return an lvalue if the chosen argument is an
5490 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
5491 type. Similarly, if @var{exp2} is returned, its return type is the same
5498 __builtin_choose_expr ( \
5499 __builtin_types_compatible_p (typeof (x), double), \
5501 __builtin_choose_expr ( \
5502 __builtin_types_compatible_p (typeof (x), float), \
5504 /* @r{The void expression results in a compile-time error} \
5505 @r{when assigning the result to something.} */ \
5509 @emph{Note:} This construct is only available for C@. Furthermore, the
5510 unused expression (@var{exp1} or @var{exp2} depending on the value of
5511 @var{const_exp}) may still generate syntax errors. This may change in
5516 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
5517 You can use the built-in function @code{__builtin_constant_p} to
5518 determine if a value is known to be constant at compile-time and hence
5519 that GCC can perform constant-folding on expressions involving that
5520 value. The argument of the function is the value to test. The function
5521 returns the integer 1 if the argument is known to be a compile-time
5522 constant and 0 if it is not known to be a compile-time constant. A
5523 return of 0 does not indicate that the value is @emph{not} a constant,
5524 but merely that GCC cannot prove it is a constant with the specified
5525 value of the @option{-O} option.
5527 You would typically use this function in an embedded application where
5528 memory was a critical resource. If you have some complex calculation,
5529 you may want it to be folded if it involves constants, but need to call
5530 a function if it does not. For example:
5533 #define Scale_Value(X) \
5534 (__builtin_constant_p (X) \
5535 ? ((X) * SCALE + OFFSET) : Scale (X))
5538 You may use this built-in function in either a macro or an inline
5539 function. However, if you use it in an inlined function and pass an
5540 argument of the function as the argument to the built-in, GCC will
5541 never return 1 when you call the inline function with a string constant
5542 or compound literal (@pxref{Compound Literals}) and will not return 1
5543 when you pass a constant numeric value to the inline function unless you
5544 specify the @option{-O} option.
5546 You may also use @code{__builtin_constant_p} in initializers for static
5547 data. For instance, you can write
5550 static const int table[] = @{
5551 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
5557 This is an acceptable initializer even if @var{EXPRESSION} is not a
5558 constant expression. GCC must be more conservative about evaluating the
5559 built-in in this case, because it has no opportunity to perform
5562 Previous versions of GCC did not accept this built-in in data
5563 initializers. The earliest version where it is completely safe is
5567 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
5568 @opindex fprofile-arcs
5569 You may use @code{__builtin_expect} to provide the compiler with
5570 branch prediction information. In general, you should prefer to
5571 use actual profile feedback for this (@option{-fprofile-arcs}), as
5572 programmers are notoriously bad at predicting how their programs
5573 actually perform. However, there are applications in which this
5574 data is hard to collect.
5576 The return value is the value of @var{exp}, which should be an
5577 integral expression. The value of @var{c} must be a compile-time
5578 constant. The semantics of the built-in are that it is expected
5579 that @var{exp} == @var{c}. For example:
5582 if (__builtin_expect (x, 0))
5587 would indicate that we do not expect to call @code{foo}, since
5588 we expect @code{x} to be zero. Since you are limited to integral
5589 expressions for @var{exp}, you should use constructions such as
5592 if (__builtin_expect (ptr != NULL, 1))
5597 when testing pointer or floating-point values.
5600 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
5601 This function is used to minimize cache-miss latency by moving data into
5602 a cache before it is accessed.
5603 You can insert calls to @code{__builtin_prefetch} into code for which
5604 you know addresses of data in memory that is likely to be accessed soon.
5605 If the target supports them, data prefetch instructions will be generated.
5606 If the prefetch is done early enough before the access then the data will
5607 be in the cache by the time it is accessed.
5609 The value of @var{addr} is the address of the memory to prefetch.
5610 There are two optional arguments, @var{rw} and @var{locality}.
5611 The value of @var{rw} is a compile-time constant one or zero; one
5612 means that the prefetch is preparing for a write to the memory address
5613 and zero, the default, means that the prefetch is preparing for a read.
5614 The value @var{locality} must be a compile-time constant integer between
5615 zero and three. A value of zero means that the data has no temporal
5616 locality, so it need not be left in the cache after the access. A value
5617 of three means that the data has a high degree of temporal locality and
5618 should be left in all levels of cache possible. Values of one and two
5619 mean, respectively, a low or moderate degree of temporal locality. The
5623 for (i = 0; i < n; i++)
5626 __builtin_prefetch (&a[i+j], 1, 1);
5627 __builtin_prefetch (&b[i+j], 0, 1);
5632 Data prefetch does not generate faults if @var{addr} is invalid, but
5633 the address expression itself must be valid. For example, a prefetch
5634 of @code{p->next} will not fault if @code{p->next} is not a valid
5635 address, but evaluation will fault if @code{p} is not a valid address.
5637 If the target does not support data prefetch, the address expression
5638 is evaluated if it includes side effects but no other code is generated
5639 and GCC does not issue a warning.
5642 @deftypefn {Built-in Function} double __builtin_huge_val (void)
5643 Returns a positive infinity, if supported by the floating-point format,
5644 else @code{DBL_MAX}. This function is suitable for implementing the
5645 ISO C macro @code{HUGE_VAL}.
5648 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
5649 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
5652 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
5653 Similar to @code{__builtin_huge_val}, except the return
5654 type is @code{long double}.
5657 @deftypefn {Built-in Function} double __builtin_inf (void)
5658 Similar to @code{__builtin_huge_val}, except a warning is generated
5659 if the target floating-point format does not support infinities.
5662 @deftypefn {Built-in Function} float __builtin_inff (void)
5663 Similar to @code{__builtin_inf}, except the return type is @code{float}.
5664 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
5667 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
5668 Similar to @code{__builtin_inf}, except the return
5669 type is @code{long double}.
5672 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
5673 This is an implementation of the ISO C99 function @code{nan}.
5675 Since ISO C99 defines this function in terms of @code{strtod}, which we
5676 do not implement, a description of the parsing is in order. The string
5677 is parsed as by @code{strtol}; that is, the base is recognized by
5678 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
5679 in the significand such that the least significant bit of the number
5680 is at the least significant bit of the significand. The number is
5681 truncated to fit the significand field provided. The significand is
5682 forced to be a quiet NaN@.
5684 This function, if given a string literal, is evaluated early enough
5685 that it is considered a compile-time constant.
5688 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
5689 Similar to @code{__builtin_nan}, except the return type is @code{float}.
5692 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
5693 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
5696 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
5697 Similar to @code{__builtin_nan}, except the significand is forced
5698 to be a signaling NaN@. The @code{nans} function is proposed by
5699 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
5702 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
5703 Similar to @code{__builtin_nans}, except the return type is @code{float}.
5706 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
5707 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
5710 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
5711 Returns one plus the index of the least significant 1-bit of @var{x}, or
5712 if @var{x} is zero, returns zero.
5715 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
5716 Returns the number of leading 0-bits in @var{x}, starting at the most
5717 significant bit position. If @var{x} is 0, the result is undefined.
5720 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
5721 Returns the number of trailing 0-bits in @var{x}, starting at the least
5722 significant bit position. If @var{x} is 0, the result is undefined.
5725 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
5726 Returns the number of 1-bits in @var{x}.
5729 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
5730 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
5734 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
5735 Similar to @code{__builtin_ffs}, except the argument type is
5736 @code{unsigned long}.
5739 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
5740 Similar to @code{__builtin_clz}, except the argument type is
5741 @code{unsigned long}.
5744 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
5745 Similar to @code{__builtin_ctz}, except the argument type is
5746 @code{unsigned long}.
5749 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
5750 Similar to @code{__builtin_popcount}, except the argument type is
5751 @code{unsigned long}.
5754 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
5755 Similar to @code{__builtin_parity}, except the argument type is
5756 @code{unsigned long}.
5759 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
5760 Similar to @code{__builtin_ffs}, except the argument type is
5761 @code{unsigned long long}.
5764 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
5765 Similar to @code{__builtin_clz}, except the argument type is
5766 @code{unsigned long long}.
5769 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
5770 Similar to @code{__builtin_ctz}, except the argument type is
5771 @code{unsigned long long}.
5774 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
5775 Similar to @code{__builtin_popcount}, except the argument type is
5776 @code{unsigned long long}.
5779 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
5780 Similar to @code{__builtin_parity}, except the argument type is
5781 @code{unsigned long long}.
5784 @deftypefn {Built-in Function} double __builtin_powi (double, int)
5785 Returns the first argument raised to the power of the second. Unlike the
5786 @code{pow} function no guarantees about precision and rounding are made.
5789 @deftypefn {Built-in Function} float __builtin_powif (float, int)
5790 Similar to @code{__builtin_powi}, except the argument and return types
5794 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
5795 Similar to @code{__builtin_powi}, except the argument and return types
5796 are @code{long double}.
5800 @node Target Builtins
5801 @section Built-in Functions Specific to Particular Target Machines
5803 On some target machines, GCC supports many built-in functions specific
5804 to those machines. Generally these generate calls to specific machine
5805 instructions, but allow the compiler to schedule those calls.
5808 * Alpha Built-in Functions::
5809 * ARM Built-in Functions::
5810 * Blackfin Built-in Functions::
5811 * FR-V Built-in Functions::
5812 * X86 Built-in Functions::
5813 * MIPS DSP Built-in Functions::
5814 * MIPS Paired-Single Support::
5815 * PowerPC AltiVec Built-in Functions::
5816 * SPARC VIS Built-in Functions::
5819 @node Alpha Built-in Functions
5820 @subsection Alpha Built-in Functions
5822 These built-in functions are available for the Alpha family of
5823 processors, depending on the command-line switches used.
5825 The following built-in functions are always available. They
5826 all generate the machine instruction that is part of the name.
5829 long __builtin_alpha_implver (void)
5830 long __builtin_alpha_rpcc (void)
5831 long __builtin_alpha_amask (long)
5832 long __builtin_alpha_cmpbge (long, long)
5833 long __builtin_alpha_extbl (long, long)
5834 long __builtin_alpha_extwl (long, long)
5835 long __builtin_alpha_extll (long, long)
5836 long __builtin_alpha_extql (long, long)
5837 long __builtin_alpha_extwh (long, long)
5838 long __builtin_alpha_extlh (long, long)
5839 long __builtin_alpha_extqh (long, long)
5840 long __builtin_alpha_insbl (long, long)
5841 long __builtin_alpha_inswl (long, long)
5842 long __builtin_alpha_insll (long, long)
5843 long __builtin_alpha_insql (long, long)
5844 long __builtin_alpha_inswh (long, long)
5845 long __builtin_alpha_inslh (long, long)
5846 long __builtin_alpha_insqh (long, long)
5847 long __builtin_alpha_mskbl (long, long)
5848 long __builtin_alpha_mskwl (long, long)
5849 long __builtin_alpha_mskll (long, long)
5850 long __builtin_alpha_mskql (long, long)
5851 long __builtin_alpha_mskwh (long, long)
5852 long __builtin_alpha_msklh (long, long)
5853 long __builtin_alpha_mskqh (long, long)
5854 long __builtin_alpha_umulh (long, long)
5855 long __builtin_alpha_zap (long, long)
5856 long __builtin_alpha_zapnot (long, long)
5859 The following built-in functions are always with @option{-mmax}
5860 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
5861 later. They all generate the machine instruction that is part
5865 long __builtin_alpha_pklb (long)
5866 long __builtin_alpha_pkwb (long)
5867 long __builtin_alpha_unpkbl (long)
5868 long __builtin_alpha_unpkbw (long)
5869 long __builtin_alpha_minub8 (long, long)
5870 long __builtin_alpha_minsb8 (long, long)
5871 long __builtin_alpha_minuw4 (long, long)
5872 long __builtin_alpha_minsw4 (long, long)
5873 long __builtin_alpha_maxub8 (long, long)
5874 long __builtin_alpha_maxsb8 (long, long)
5875 long __builtin_alpha_maxuw4 (long, long)
5876 long __builtin_alpha_maxsw4 (long, long)
5877 long __builtin_alpha_perr (long, long)
5880 The following built-in functions are always with @option{-mcix}
5881 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
5882 later. They all generate the machine instruction that is part
5886 long __builtin_alpha_cttz (long)
5887 long __builtin_alpha_ctlz (long)
5888 long __builtin_alpha_ctpop (long)
5891 The following builtins are available on systems that use the OSF/1
5892 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
5893 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
5894 @code{rdval} and @code{wrval}.
5897 void *__builtin_thread_pointer (void)
5898 void __builtin_set_thread_pointer (void *)
5901 @node ARM Built-in Functions
5902 @subsection ARM Built-in Functions
5904 These built-in functions are available for the ARM family of
5905 processors, when the @option{-mcpu=iwmmxt} switch is used:
5908 typedef int v2si __attribute__ ((vector_size (8)));
5909 typedef short v4hi __attribute__ ((vector_size (8)));
5910 typedef char v8qi __attribute__ ((vector_size (8)));
5912 int __builtin_arm_getwcx (int)
5913 void __builtin_arm_setwcx (int, int)
5914 int __builtin_arm_textrmsb (v8qi, int)
5915 int __builtin_arm_textrmsh (v4hi, int)
5916 int __builtin_arm_textrmsw (v2si, int)
5917 int __builtin_arm_textrmub (v8qi, int)
5918 int __builtin_arm_textrmuh (v4hi, int)
5919 int __builtin_arm_textrmuw (v2si, int)
5920 v8qi __builtin_arm_tinsrb (v8qi, int)
5921 v4hi __builtin_arm_tinsrh (v4hi, int)
5922 v2si __builtin_arm_tinsrw (v2si, int)
5923 long long __builtin_arm_tmia (long long, int, int)
5924 long long __builtin_arm_tmiabb (long long, int, int)
5925 long long __builtin_arm_tmiabt (long long, int, int)
5926 long long __builtin_arm_tmiaph (long long, int, int)
5927 long long __builtin_arm_tmiatb (long long, int, int)
5928 long long __builtin_arm_tmiatt (long long, int, int)
5929 int __builtin_arm_tmovmskb (v8qi)
5930 int __builtin_arm_tmovmskh (v4hi)
5931 int __builtin_arm_tmovmskw (v2si)
5932 long long __builtin_arm_waccb (v8qi)
5933 long long __builtin_arm_wacch (v4hi)
5934 long long __builtin_arm_waccw (v2si)
5935 v8qi __builtin_arm_waddb (v8qi, v8qi)
5936 v8qi __builtin_arm_waddbss (v8qi, v8qi)
5937 v8qi __builtin_arm_waddbus (v8qi, v8qi)
5938 v4hi __builtin_arm_waddh (v4hi, v4hi)
5939 v4hi __builtin_arm_waddhss (v4hi, v4hi)
5940 v4hi __builtin_arm_waddhus (v4hi, v4hi)
5941 v2si __builtin_arm_waddw (v2si, v2si)
5942 v2si __builtin_arm_waddwss (v2si, v2si)
5943 v2si __builtin_arm_waddwus (v2si, v2si)
5944 v8qi __builtin_arm_walign (v8qi, v8qi, int)
5945 long long __builtin_arm_wand(long long, long long)
5946 long long __builtin_arm_wandn (long long, long long)
5947 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
5948 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
5949 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
5950 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
5951 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
5952 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
5953 v2si __builtin_arm_wcmpeqw (v2si, v2si)
5954 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
5955 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
5956 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
5957 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
5958 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
5959 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
5960 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
5961 long long __builtin_arm_wmacsz (v4hi, v4hi)
5962 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
5963 long long __builtin_arm_wmacuz (v4hi, v4hi)
5964 v4hi __builtin_arm_wmadds (v4hi, v4hi)
5965 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
5966 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
5967 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
5968 v2si __builtin_arm_wmaxsw (v2si, v2si)
5969 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
5970 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
5971 v2si __builtin_arm_wmaxuw (v2si, v2si)
5972 v8qi __builtin_arm_wminsb (v8qi, v8qi)
5973 v4hi __builtin_arm_wminsh (v4hi, v4hi)
5974 v2si __builtin_arm_wminsw (v2si, v2si)
5975 v8qi __builtin_arm_wminub (v8qi, v8qi)
5976 v4hi __builtin_arm_wminuh (v4hi, v4hi)
5977 v2si __builtin_arm_wminuw (v2si, v2si)
5978 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
5979 v4hi __builtin_arm_wmulul (v4hi, v4hi)
5980 v4hi __builtin_arm_wmulum (v4hi, v4hi)
5981 long long __builtin_arm_wor (long long, long long)
5982 v2si __builtin_arm_wpackdss (long long, long long)
5983 v2si __builtin_arm_wpackdus (long long, long long)
5984 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
5985 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
5986 v4hi __builtin_arm_wpackwss (v2si, v2si)
5987 v4hi __builtin_arm_wpackwus (v2si, v2si)
5988 long long __builtin_arm_wrord (long long, long long)
5989 long long __builtin_arm_wrordi (long long, int)
5990 v4hi __builtin_arm_wrorh (v4hi, long long)
5991 v4hi __builtin_arm_wrorhi (v4hi, int)
5992 v2si __builtin_arm_wrorw (v2si, long long)
5993 v2si __builtin_arm_wrorwi (v2si, int)
5994 v2si __builtin_arm_wsadb (v8qi, v8qi)
5995 v2si __builtin_arm_wsadbz (v8qi, v8qi)
5996 v2si __builtin_arm_wsadh (v4hi, v4hi)
5997 v2si __builtin_arm_wsadhz (v4hi, v4hi)
5998 v4hi __builtin_arm_wshufh (v4hi, int)
5999 long long __builtin_arm_wslld (long long, long long)
6000 long long __builtin_arm_wslldi (long long, int)
6001 v4hi __builtin_arm_wsllh (v4hi, long long)
6002 v4hi __builtin_arm_wsllhi (v4hi, int)
6003 v2si __builtin_arm_wsllw (v2si, long long)
6004 v2si __builtin_arm_wsllwi (v2si, int)
6005 long long __builtin_arm_wsrad (long long, long long)
6006 long long __builtin_arm_wsradi (long long, int)
6007 v4hi __builtin_arm_wsrah (v4hi, long long)
6008 v4hi __builtin_arm_wsrahi (v4hi, int)
6009 v2si __builtin_arm_wsraw (v2si, long long)
6010 v2si __builtin_arm_wsrawi (v2si, int)
6011 long long __builtin_arm_wsrld (long long, long long)
6012 long long __builtin_arm_wsrldi (long long, int)
6013 v4hi __builtin_arm_wsrlh (v4hi, long long)
6014 v4hi __builtin_arm_wsrlhi (v4hi, int)
6015 v2si __builtin_arm_wsrlw (v2si, long long)
6016 v2si __builtin_arm_wsrlwi (v2si, int)
6017 v8qi __builtin_arm_wsubb (v8qi, v8qi)
6018 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
6019 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
6020 v4hi __builtin_arm_wsubh (v4hi, v4hi)
6021 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
6022 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
6023 v2si __builtin_arm_wsubw (v2si, v2si)
6024 v2si __builtin_arm_wsubwss (v2si, v2si)
6025 v2si __builtin_arm_wsubwus (v2si, v2si)
6026 v4hi __builtin_arm_wunpckehsb (v8qi)
6027 v2si __builtin_arm_wunpckehsh (v4hi)
6028 long long __builtin_arm_wunpckehsw (v2si)
6029 v4hi __builtin_arm_wunpckehub (v8qi)
6030 v2si __builtin_arm_wunpckehuh (v4hi)
6031 long long __builtin_arm_wunpckehuw (v2si)
6032 v4hi __builtin_arm_wunpckelsb (v8qi)
6033 v2si __builtin_arm_wunpckelsh (v4hi)
6034 long long __builtin_arm_wunpckelsw (v2si)
6035 v4hi __builtin_arm_wunpckelub (v8qi)
6036 v2si __builtin_arm_wunpckeluh (v4hi)
6037 long long __builtin_arm_wunpckeluw (v2si)
6038 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
6039 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
6040 v2si __builtin_arm_wunpckihw (v2si, v2si)
6041 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
6042 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
6043 v2si __builtin_arm_wunpckilw (v2si, v2si)
6044 long long __builtin_arm_wxor (long long, long long)
6045 long long __builtin_arm_wzero ()
6048 @node Blackfin Built-in Functions
6049 @subsection Blackfin Built-in Functions
6051 Currently, there are two Blackfin-specific built-in functions. These are
6052 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
6053 using inline assembly; by using these built-in functions the compiler can
6054 automatically add workarounds for hardware errata involving these
6055 instructions. These functions are named as follows:
6058 void __builtin_bfin_csync (void)
6059 void __builtin_bfin_ssync (void)
6062 @node FR-V Built-in Functions
6063 @subsection FR-V Built-in Functions
6065 GCC provides many FR-V-specific built-in functions. In general,
6066 these functions are intended to be compatible with those described
6067 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
6068 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
6069 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
6070 pointer rather than by value.
6072 Most of the functions are named after specific FR-V instructions.
6073 Such functions are said to be ``directly mapped'' and are summarized
6074 here in tabular form.
6078 * Directly-mapped Integer Functions::
6079 * Directly-mapped Media Functions::
6080 * Raw read/write Functions::
6081 * Other Built-in Functions::
6084 @node Argument Types
6085 @subsubsection Argument Types
6087 The arguments to the built-in functions can be divided into three groups:
6088 register numbers, compile-time constants and run-time values. In order
6089 to make this classification clear at a glance, the arguments and return
6090 values are given the following pseudo types:
6092 @multitable @columnfractions .20 .30 .15 .35
6093 @item Pseudo type @tab Real C type @tab Constant? @tab Description
6094 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
6095 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
6096 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
6097 @item @code{uw2} @tab @code{unsigned long long} @tab No
6098 @tab an unsigned doubleword
6099 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
6100 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
6101 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
6102 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
6105 These pseudo types are not defined by GCC, they are simply a notational
6106 convenience used in this manual.
6108 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
6109 and @code{sw2} are evaluated at run time. They correspond to
6110 register operands in the underlying FR-V instructions.
6112 @code{const} arguments represent immediate operands in the underlying
6113 FR-V instructions. They must be compile-time constants.
6115 @code{acc} arguments are evaluated at compile time and specify the number
6116 of an accumulator register. For example, an @code{acc} argument of 2
6117 will select the ACC2 register.
6119 @code{iacc} arguments are similar to @code{acc} arguments but specify the
6120 number of an IACC register. See @pxref{Other Built-in Functions}
6123 @node Directly-mapped Integer Functions
6124 @subsubsection Directly-mapped Integer Functions
6126 The functions listed below map directly to FR-V I-type instructions.
6128 @multitable @columnfractions .45 .32 .23
6129 @item Function prototype @tab Example usage @tab Assembly output
6130 @item @code{sw1 __ADDSS (sw1, sw1)}
6131 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
6132 @tab @code{ADDSS @var{a},@var{b},@var{c}}
6133 @item @code{sw1 __SCAN (sw1, sw1)}
6134 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
6135 @tab @code{SCAN @var{a},@var{b},@var{c}}
6136 @item @code{sw1 __SCUTSS (sw1)}
6137 @tab @code{@var{b} = __SCUTSS (@var{a})}
6138 @tab @code{SCUTSS @var{a},@var{b}}
6139 @item @code{sw1 __SLASS (sw1, sw1)}
6140 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
6141 @tab @code{SLASS @var{a},@var{b},@var{c}}
6142 @item @code{void __SMASS (sw1, sw1)}
6143 @tab @code{__SMASS (@var{a}, @var{b})}
6144 @tab @code{SMASS @var{a},@var{b}}
6145 @item @code{void __SMSSS (sw1, sw1)}
6146 @tab @code{__SMSSS (@var{a}, @var{b})}
6147 @tab @code{SMSSS @var{a},@var{b}}
6148 @item @code{void __SMU (sw1, sw1)}
6149 @tab @code{__SMU (@var{a}, @var{b})}
6150 @tab @code{SMU @var{a},@var{b}}
6151 @item @code{sw2 __SMUL (sw1, sw1)}
6152 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
6153 @tab @code{SMUL @var{a},@var{b},@var{c}}
6154 @item @code{sw1 __SUBSS (sw1, sw1)}
6155 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
6156 @tab @code{SUBSS @var{a},@var{b},@var{c}}
6157 @item @code{uw2 __UMUL (uw1, uw1)}
6158 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
6159 @tab @code{UMUL @var{a},@var{b},@var{c}}
6162 @node Directly-mapped Media Functions
6163 @subsubsection Directly-mapped Media Functions
6165 The functions listed below map directly to FR-V M-type instructions.
6167 @multitable @columnfractions .45 .32 .23
6168 @item Function prototype @tab Example usage @tab Assembly output
6169 @item @code{uw1 __MABSHS (sw1)}
6170 @tab @code{@var{b} = __MABSHS (@var{a})}
6171 @tab @code{MABSHS @var{a},@var{b}}
6172 @item @code{void __MADDACCS (acc, acc)}
6173 @tab @code{__MADDACCS (@var{b}, @var{a})}
6174 @tab @code{MADDACCS @var{a},@var{b}}
6175 @item @code{sw1 __MADDHSS (sw1, sw1)}
6176 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
6177 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
6178 @item @code{uw1 __MADDHUS (uw1, uw1)}
6179 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
6180 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
6181 @item @code{uw1 __MAND (uw1, uw1)}
6182 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
6183 @tab @code{MAND @var{a},@var{b},@var{c}}
6184 @item @code{void __MASACCS (acc, acc)}
6185 @tab @code{__MASACCS (@var{b}, @var{a})}
6186 @tab @code{MASACCS @var{a},@var{b}}
6187 @item @code{uw1 __MAVEH (uw1, uw1)}
6188 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
6189 @tab @code{MAVEH @var{a},@var{b},@var{c}}
6190 @item @code{uw2 __MBTOH (uw1)}
6191 @tab @code{@var{b} = __MBTOH (@var{a})}
6192 @tab @code{MBTOH @var{a},@var{b}}
6193 @item @code{void __MBTOHE (uw1 *, uw1)}
6194 @tab @code{__MBTOHE (&@var{b}, @var{a})}
6195 @tab @code{MBTOHE @var{a},@var{b}}
6196 @item @code{void __MCLRACC (acc)}
6197 @tab @code{__MCLRACC (@var{a})}
6198 @tab @code{MCLRACC @var{a}}
6199 @item @code{void __MCLRACCA (void)}
6200 @tab @code{__MCLRACCA ()}
6201 @tab @code{MCLRACCA}
6202 @item @code{uw1 __Mcop1 (uw1, uw1)}
6203 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
6204 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
6205 @item @code{uw1 __Mcop2 (uw1, uw1)}
6206 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
6207 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
6208 @item @code{uw1 __MCPLHI (uw2, const)}
6209 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
6210 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
6211 @item @code{uw1 __MCPLI (uw2, const)}
6212 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
6213 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
6214 @item @code{void __MCPXIS (acc, sw1, sw1)}
6215 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
6216 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
6217 @item @code{void __MCPXIU (acc, uw1, uw1)}
6218 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
6219 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
6220 @item @code{void __MCPXRS (acc, sw1, sw1)}
6221 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
6222 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
6223 @item @code{void __MCPXRU (acc, uw1, uw1)}
6224 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
6225 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
6226 @item @code{uw1 __MCUT (acc, uw1)}
6227 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
6228 @tab @code{MCUT @var{a},@var{b},@var{c}}
6229 @item @code{uw1 __MCUTSS (acc, sw1)}
6230 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
6231 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
6232 @item @code{void __MDADDACCS (acc, acc)}
6233 @tab @code{__MDADDACCS (@var{b}, @var{a})}
6234 @tab @code{MDADDACCS @var{a},@var{b}}
6235 @item @code{void __MDASACCS (acc, acc)}
6236 @tab @code{__MDASACCS (@var{b}, @var{a})}
6237 @tab @code{MDASACCS @var{a},@var{b}}
6238 @item @code{uw2 __MDCUTSSI (acc, const)}
6239 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
6240 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
6241 @item @code{uw2 __MDPACKH (uw2, uw2)}
6242 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
6243 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
6244 @item @code{uw2 __MDROTLI (uw2, const)}
6245 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
6246 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
6247 @item @code{void __MDSUBACCS (acc, acc)}
6248 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
6249 @tab @code{MDSUBACCS @var{a},@var{b}}
6250 @item @code{void __MDUNPACKH (uw1 *, uw2)}
6251 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
6252 @tab @code{MDUNPACKH @var{a},@var{b}}
6253 @item @code{uw2 __MEXPDHD (uw1, const)}
6254 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
6255 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
6256 @item @code{uw1 __MEXPDHW (uw1, const)}
6257 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
6258 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
6259 @item @code{uw1 __MHDSETH (uw1, const)}
6260 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
6261 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
6262 @item @code{sw1 __MHDSETS (const)}
6263 @tab @code{@var{b} = __MHDSETS (@var{a})}
6264 @tab @code{MHDSETS #@var{a},@var{b}}
6265 @item @code{uw1 __MHSETHIH (uw1, const)}
6266 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
6267 @tab @code{MHSETHIH #@var{a},@var{b}}
6268 @item @code{sw1 __MHSETHIS (sw1, const)}
6269 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
6270 @tab @code{MHSETHIS #@var{a},@var{b}}
6271 @item @code{uw1 __MHSETLOH (uw1, const)}
6272 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
6273 @tab @code{MHSETLOH #@var{a},@var{b}}
6274 @item @code{sw1 __MHSETLOS (sw1, const)}
6275 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
6276 @tab @code{MHSETLOS #@var{a},@var{b}}
6277 @item @code{uw1 __MHTOB (uw2)}
6278 @tab @code{@var{b} = __MHTOB (@var{a})}
6279 @tab @code{MHTOB @var{a},@var{b}}
6280 @item @code{void __MMACHS (acc, sw1, sw1)}
6281 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
6282 @tab @code{MMACHS @var{a},@var{b},@var{c}}
6283 @item @code{void __MMACHU (acc, uw1, uw1)}
6284 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
6285 @tab @code{MMACHU @var{a},@var{b},@var{c}}
6286 @item @code{void __MMRDHS (acc, sw1, sw1)}
6287 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
6288 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
6289 @item @code{void __MMRDHU (acc, uw1, uw1)}
6290 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
6291 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
6292 @item @code{void __MMULHS (acc, sw1, sw1)}
6293 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
6294 @tab @code{MMULHS @var{a},@var{b},@var{c}}
6295 @item @code{void __MMULHU (acc, uw1, uw1)}
6296 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
6297 @tab @code{MMULHU @var{a},@var{b},@var{c}}
6298 @item @code{void __MMULXHS (acc, sw1, sw1)}
6299 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
6300 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
6301 @item @code{void __MMULXHU (acc, uw1, uw1)}
6302 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
6303 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
6304 @item @code{uw1 __MNOT (uw1)}
6305 @tab @code{@var{b} = __MNOT (@var{a})}
6306 @tab @code{MNOT @var{a},@var{b}}
6307 @item @code{uw1 __MOR (uw1, uw1)}
6308 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
6309 @tab @code{MOR @var{a},@var{b},@var{c}}
6310 @item @code{uw1 __MPACKH (uh, uh)}
6311 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
6312 @tab @code{MPACKH @var{a},@var{b},@var{c}}
6313 @item @code{sw2 __MQADDHSS (sw2, sw2)}
6314 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
6315 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
6316 @item @code{uw2 __MQADDHUS (uw2, uw2)}
6317 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
6318 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
6319 @item @code{void __MQCPXIS (acc, sw2, sw2)}
6320 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
6321 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
6322 @item @code{void __MQCPXIU (acc, uw2, uw2)}
6323 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
6324 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
6325 @item @code{void __MQCPXRS (acc, sw2, sw2)}
6326 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
6327 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
6328 @item @code{void __MQCPXRU (acc, uw2, uw2)}
6329 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
6330 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
6331 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
6332 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
6333 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
6334 @item @code{sw2 __MQLMTHS (sw2, sw2)}
6335 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
6336 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
6337 @item @code{void __MQMACHS (acc, sw2, sw2)}
6338 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
6339 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
6340 @item @code{void __MQMACHU (acc, uw2, uw2)}
6341 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
6342 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
6343 @item @code{void __MQMACXHS (acc, sw2, sw2)}
6344 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
6345 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
6346 @item @code{void __MQMULHS (acc, sw2, sw2)}
6347 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
6348 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
6349 @item @code{void __MQMULHU (acc, uw2, uw2)}
6350 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
6351 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
6352 @item @code{void __MQMULXHS (acc, sw2, sw2)}
6353 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
6354 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
6355 @item @code{void __MQMULXHU (acc, uw2, uw2)}
6356 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
6357 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
6358 @item @code{sw2 __MQSATHS (sw2, sw2)}
6359 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
6360 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
6361 @item @code{uw2 __MQSLLHI (uw2, int)}
6362 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
6363 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
6364 @item @code{sw2 __MQSRAHI (sw2, int)}
6365 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
6366 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
6367 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
6368 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
6369 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
6370 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
6371 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
6372 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
6373 @item @code{void __MQXMACHS (acc, sw2, sw2)}
6374 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
6375 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
6376 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
6377 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
6378 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
6379 @item @code{uw1 __MRDACC (acc)}
6380 @tab @code{@var{b} = __MRDACC (@var{a})}
6381 @tab @code{MRDACC @var{a},@var{b}}
6382 @item @code{uw1 __MRDACCG (acc)}
6383 @tab @code{@var{b} = __MRDACCG (@var{a})}
6384 @tab @code{MRDACCG @var{a},@var{b}}
6385 @item @code{uw1 __MROTLI (uw1, const)}
6386 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
6387 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
6388 @item @code{uw1 __MROTRI (uw1, const)}
6389 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
6390 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
6391 @item @code{sw1 __MSATHS (sw1, sw1)}
6392 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
6393 @tab @code{MSATHS @var{a},@var{b},@var{c}}
6394 @item @code{uw1 __MSATHU (uw1, uw1)}
6395 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
6396 @tab @code{MSATHU @var{a},@var{b},@var{c}}
6397 @item @code{uw1 __MSLLHI (uw1, const)}
6398 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
6399 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
6400 @item @code{sw1 __MSRAHI (sw1, const)}
6401 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
6402 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
6403 @item @code{uw1 __MSRLHI (uw1, const)}
6404 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
6405 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
6406 @item @code{void __MSUBACCS (acc, acc)}
6407 @tab @code{__MSUBACCS (@var{b}, @var{a})}
6408 @tab @code{MSUBACCS @var{a},@var{b}}
6409 @item @code{sw1 __MSUBHSS (sw1, sw1)}
6410 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
6411 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
6412 @item @code{uw1 __MSUBHUS (uw1, uw1)}
6413 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
6414 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
6415 @item @code{void __MTRAP (void)}
6416 @tab @code{__MTRAP ()}
6418 @item @code{uw2 __MUNPACKH (uw1)}
6419 @tab @code{@var{b} = __MUNPACKH (@var{a})}
6420 @tab @code{MUNPACKH @var{a},@var{b}}
6421 @item @code{uw1 __MWCUT (uw2, uw1)}
6422 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
6423 @tab @code{MWCUT @var{a},@var{b},@var{c}}
6424 @item @code{void __MWTACC (acc, uw1)}
6425 @tab @code{__MWTACC (@var{b}, @var{a})}
6426 @tab @code{MWTACC @var{a},@var{b}}
6427 @item @code{void __MWTACCG (acc, uw1)}
6428 @tab @code{__MWTACCG (@var{b}, @var{a})}
6429 @tab @code{MWTACCG @var{a},@var{b}}
6430 @item @code{uw1 __MXOR (uw1, uw1)}
6431 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
6432 @tab @code{MXOR @var{a},@var{b},@var{c}}
6435 @node Raw read/write Functions
6436 @subsubsection Raw read/write Functions
6438 This sections describes built-in functions related to read and write
6439 instructions to access memory. These functions generate
6440 @code{membar} instructions to flush the I/O load and stores where
6441 appropriate, as described in Fujitsu's manual described above.
6445 @item unsigned char __builtin_read8 (void *@var{data})
6446 @item unsigned short __builtin_read16 (void *@var{data})
6447 @item unsigned long __builtin_read32 (void *@var{data})
6448 @item unsigned long long __builtin_read64 (void *@var{data})
6450 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
6451 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
6452 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
6453 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
6456 @node Other Built-in Functions
6457 @subsubsection Other Built-in Functions
6459 This section describes built-in functions that are not named after
6460 a specific FR-V instruction.
6463 @item sw2 __IACCreadll (iacc @var{reg})
6464 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
6465 for future expansion and must be 0.
6467 @item sw1 __IACCreadl (iacc @var{reg})
6468 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
6469 Other values of @var{reg} are rejected as invalid.
6471 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
6472 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
6473 is reserved for future expansion and must be 0.
6475 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
6476 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
6477 is 1. Other values of @var{reg} are rejected as invalid.
6479 @item void __data_prefetch0 (const void *@var{x})
6480 Use the @code{dcpl} instruction to load the contents of address @var{x}
6481 into the data cache.
6483 @item void __data_prefetch (const void *@var{x})
6484 Use the @code{nldub} instruction to load the contents of address @var{x}
6485 into the data cache. The instruction will be issued in slot I1@.
6488 @node X86 Built-in Functions
6489 @subsection X86 Built-in Functions
6491 These built-in functions are available for the i386 and x86-64 family
6492 of computers, depending on the command-line switches used.
6494 Note that, if you specify command-line switches such as @option{-msse},
6495 the compiler could use the extended instruction sets even if the built-ins
6496 are not used explicitly in the program. For this reason, applications
6497 which perform runtime CPU detection must compile separate files for each
6498 supported architecture, using the appropriate flags. In particular,
6499 the file containing the CPU detection code should be compiled without
6502 The following machine modes are available for use with MMX built-in functions
6503 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
6504 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
6505 vector of eight 8-bit integers. Some of the built-in functions operate on
6506 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
6508 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
6509 of two 32-bit floating point values.
6511 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
6512 floating point values. Some instructions use a vector of four 32-bit
6513 integers, these use @code{V4SI}. Finally, some instructions operate on an
6514 entire vector register, interpreting it as a 128-bit integer, these use mode
6517 The following built-in functions are made available by @option{-mmmx}.
6518 All of them generate the machine instruction that is part of the name.
6521 v8qi __builtin_ia32_paddb (v8qi, v8qi)
6522 v4hi __builtin_ia32_paddw (v4hi, v4hi)
6523 v2si __builtin_ia32_paddd (v2si, v2si)
6524 v8qi __builtin_ia32_psubb (v8qi, v8qi)
6525 v4hi __builtin_ia32_psubw (v4hi, v4hi)
6526 v2si __builtin_ia32_psubd (v2si, v2si)
6527 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
6528 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
6529 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
6530 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
6531 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
6532 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
6533 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
6534 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
6535 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
6536 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
6537 di __builtin_ia32_pand (di, di)
6538 di __builtin_ia32_pandn (di,di)
6539 di __builtin_ia32_por (di, di)
6540 di __builtin_ia32_pxor (di, di)
6541 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
6542 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
6543 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
6544 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
6545 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
6546 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
6547 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
6548 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
6549 v2si __builtin_ia32_punpckhdq (v2si, v2si)
6550 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
6551 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
6552 v2si __builtin_ia32_punpckldq (v2si, v2si)
6553 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
6554 v4hi __builtin_ia32_packssdw (v2si, v2si)
6555 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
6558 The following built-in functions are made available either with
6559 @option{-msse}, or with a combination of @option{-m3dnow} and
6560 @option{-march=athlon}. All of them generate the machine
6561 instruction that is part of the name.
6564 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
6565 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
6566 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
6567 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
6568 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
6569 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
6570 v8qi __builtin_ia32_pminub (v8qi, v8qi)
6571 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
6572 int __builtin_ia32_pextrw (v4hi, int)
6573 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
6574 int __builtin_ia32_pmovmskb (v8qi)
6575 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
6576 void __builtin_ia32_movntq (di *, di)
6577 void __builtin_ia32_sfence (void)
6580 The following built-in functions are available when @option{-msse} is used.
6581 All of them generate the machine instruction that is part of the name.
6584 int __builtin_ia32_comieq (v4sf, v4sf)
6585 int __builtin_ia32_comineq (v4sf, v4sf)
6586 int __builtin_ia32_comilt (v4sf, v4sf)
6587 int __builtin_ia32_comile (v4sf, v4sf)
6588 int __builtin_ia32_comigt (v4sf, v4sf)
6589 int __builtin_ia32_comige (v4sf, v4sf)
6590 int __builtin_ia32_ucomieq (v4sf, v4sf)
6591 int __builtin_ia32_ucomineq (v4sf, v4sf)
6592 int __builtin_ia32_ucomilt (v4sf, v4sf)
6593 int __builtin_ia32_ucomile (v4sf, v4sf)
6594 int __builtin_ia32_ucomigt (v4sf, v4sf)
6595 int __builtin_ia32_ucomige (v4sf, v4sf)
6596 v4sf __builtin_ia32_addps (v4sf, v4sf)
6597 v4sf __builtin_ia32_subps (v4sf, v4sf)
6598 v4sf __builtin_ia32_mulps (v4sf, v4sf)
6599 v4sf __builtin_ia32_divps (v4sf, v4sf)
6600 v4sf __builtin_ia32_addss (v4sf, v4sf)
6601 v4sf __builtin_ia32_subss (v4sf, v4sf)
6602 v4sf __builtin_ia32_mulss (v4sf, v4sf)
6603 v4sf __builtin_ia32_divss (v4sf, v4sf)
6604 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
6605 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
6606 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
6607 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
6608 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
6609 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
6610 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
6611 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
6612 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
6613 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
6614 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
6615 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
6616 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
6617 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
6618 v4si __builtin_ia32_cmpless (v4sf, v4sf)
6619 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
6620 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
6621 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
6622 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
6623 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
6624 v4sf __builtin_ia32_maxps (v4sf, v4sf)
6625 v4sf __builtin_ia32_maxss (v4sf, v4sf)
6626 v4sf __builtin_ia32_minps (v4sf, v4sf)
6627 v4sf __builtin_ia32_minss (v4sf, v4sf)
6628 v4sf __builtin_ia32_andps (v4sf, v4sf)
6629 v4sf __builtin_ia32_andnps (v4sf, v4sf)
6630 v4sf __builtin_ia32_orps (v4sf, v4sf)
6631 v4sf __builtin_ia32_xorps (v4sf, v4sf)
6632 v4sf __builtin_ia32_movss (v4sf, v4sf)
6633 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
6634 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
6635 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
6636 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
6637 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
6638 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
6639 v2si __builtin_ia32_cvtps2pi (v4sf)
6640 int __builtin_ia32_cvtss2si (v4sf)
6641 v2si __builtin_ia32_cvttps2pi (v4sf)
6642 int __builtin_ia32_cvttss2si (v4sf)
6643 v4sf __builtin_ia32_rcpps (v4sf)
6644 v4sf __builtin_ia32_rsqrtps (v4sf)
6645 v4sf __builtin_ia32_sqrtps (v4sf)
6646 v4sf __builtin_ia32_rcpss (v4sf)
6647 v4sf __builtin_ia32_rsqrtss (v4sf)
6648 v4sf __builtin_ia32_sqrtss (v4sf)
6649 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
6650 void __builtin_ia32_movntps (float *, v4sf)
6651 int __builtin_ia32_movmskps (v4sf)
6654 The following built-in functions are available when @option{-msse} is used.
6657 @item v4sf __builtin_ia32_loadaps (float *)
6658 Generates the @code{movaps} machine instruction as a load from memory.
6659 @item void __builtin_ia32_storeaps (float *, v4sf)
6660 Generates the @code{movaps} machine instruction as a store to memory.
6661 @item v4sf __builtin_ia32_loadups (float *)
6662 Generates the @code{movups} machine instruction as a load from memory.
6663 @item void __builtin_ia32_storeups (float *, v4sf)
6664 Generates the @code{movups} machine instruction as a store to memory.
6665 @item v4sf __builtin_ia32_loadsss (float *)
6666 Generates the @code{movss} machine instruction as a load from memory.
6667 @item void __builtin_ia32_storess (float *, v4sf)
6668 Generates the @code{movss} machine instruction as a store to memory.
6669 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
6670 Generates the @code{movhps} machine instruction as a load from memory.
6671 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
6672 Generates the @code{movlps} machine instruction as a load from memory
6673 @item void __builtin_ia32_storehps (v4sf, v2si *)
6674 Generates the @code{movhps} machine instruction as a store to memory.
6675 @item void __builtin_ia32_storelps (v4sf, v2si *)
6676 Generates the @code{movlps} machine instruction as a store to memory.
6679 The following built-in functions are available when @option{-msse3} is used.
6680 All of them generate the machine instruction that is part of the name.
6683 v2df __builtin_ia32_addsubpd (v2df, v2df)
6684 v2df __builtin_ia32_addsubps (v2df, v2df)
6685 v2df __builtin_ia32_haddpd (v2df, v2df)
6686 v2df __builtin_ia32_haddps (v2df, v2df)
6687 v2df __builtin_ia32_hsubpd (v2df, v2df)
6688 v2df __builtin_ia32_hsubps (v2df, v2df)
6689 v16qi __builtin_ia32_lddqu (char const *)
6690 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
6691 v2df __builtin_ia32_movddup (v2df)
6692 v4sf __builtin_ia32_movshdup (v4sf)
6693 v4sf __builtin_ia32_movsldup (v4sf)
6694 void __builtin_ia32_mwait (unsigned int, unsigned int)
6697 The following built-in functions are available when @option{-msse3} is used.
6700 @item v2df __builtin_ia32_loadddup (double const *)
6701 Generates the @code{movddup} machine instruction as a load from memory.
6704 The following built-in functions are available when @option{-m3dnow} is used.
6705 All of them generate the machine instruction that is part of the name.
6708 void __builtin_ia32_femms (void)
6709 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
6710 v2si __builtin_ia32_pf2id (v2sf)
6711 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
6712 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
6713 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
6714 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
6715 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
6716 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
6717 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
6718 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
6719 v2sf __builtin_ia32_pfrcp (v2sf)
6720 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
6721 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
6722 v2sf __builtin_ia32_pfrsqrt (v2sf)
6723 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
6724 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
6725 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
6726 v2sf __builtin_ia32_pi2fd (v2si)
6727 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
6730 The following built-in functions are available when both @option{-m3dnow}
6731 and @option{-march=athlon} are used. All of them generate the machine
6732 instruction that is part of the name.
6735 v2si __builtin_ia32_pf2iw (v2sf)
6736 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
6737 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
6738 v2sf __builtin_ia32_pi2fw (v2si)
6739 v2sf __builtin_ia32_pswapdsf (v2sf)
6740 v2si __builtin_ia32_pswapdsi (v2si)
6743 @node MIPS DSP Built-in Functions
6744 @subsection MIPS DSP Built-in Functions
6746 The MIPS DSP Application-Specific Extension (ASE) includes new
6747 instructions that are designed to improve the performance of DSP and
6748 media applications. It provides instructions that operate on packed
6749 8-bit integer data, Q15 fractional data and Q31 fractional data.
6751 GCC supports MIPS DSP operations using both the generic
6752 vector extensions (@pxref{Vector Extensions}) and a collection of
6753 MIPS-specific built-in functions. Both kinds of support are
6754 enabled by the @option{-mdsp} command-line option.
6756 At present, GCC only provides support for operations on 32-bit
6757 vectors. The vector type associated with 8-bit integer data is
6758 usually called @code{v4i8} and the vector type associated with Q15 is
6759 usually called @code{v2q15}. They can be defined in C as follows:
6762 typedef char v4i8 __attribute__ ((vector_size(4)));
6763 typedef short v2q15 __attribute__ ((vector_size(4)));
6766 @code{v4i8} and @code{v2q15} values are initialized in the same way as
6767 aggregates. For example:
6770 v4i8 a = @{1, 2, 3, 4@};
6772 b = (v4i8) @{5, 6, 7, 8@};
6774 v2q15 c = @{0x0fcb, 0x3a75@};
6776 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
6779 @emph{Note:} The CPU's endianness determines the order in which values
6780 are packed. On little-endian targets, the first value is the least
6781 significant and the last value is the most significant. The opposite
6782 order applies to big-endian targets. For example, the code above will
6783 set the lowest byte of @code{a} to @code{1} on little-endian targets
6784 and @code{4} on big-endian targets.
6786 @emph{Note:} Q15 and Q31 values must be initialized with their integer
6787 representation. As shown in this example, the integer representation
6788 of a Q15 value can be obtained by multiplying the fractional value by
6789 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
6792 The table below lists the @code{v4i8} and @code{v2q15} operations for which
6793 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
6794 and @code{c} and @code{d} are @code{v2q15} values.
6796 @multitable @columnfractions .50 .50
6797 @item C code @tab MIPS instruction
6798 @item @code{a + b} @tab @code{addu.qb}
6799 @item @code{c + d} @tab @code{addq.ph}
6800 @item @code{a - b} @tab @code{subu.qb}
6801 @item @code{c - d} @tab @code{subq.ph}
6804 It is easier to describe the DSP built-in functions if we first define
6805 the following types:
6810 typedef long long a64;
6813 @code{q31} and @code{i32} are actually the same as @code{int}, but we
6814 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
6815 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
6816 @code{long long}, but we use @code{a64} to indicate values that will
6817 be placed in one of the four DSP accumulators (@code{$ac0},
6818 @code{$ac1}, @code{$ac2} or @code{$ac3}).
6820 Also, some built-in functions prefer or require immediate numbers as
6821 parameters, because the corresponding DSP instructions accept both immediate
6822 numbers and register operands, or accept immediate numbers only. The
6823 immediate parameters are listed as follows.
6831 imm_n32_31: -32 to 31.
6832 imm_n512_511: -512 to 511.
6835 The following built-in functions map directly to a particular MIPS DSP
6836 instruction. Please refer to the architecture specification
6837 for details on what each instruction does.
6840 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
6841 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
6842 q31 __builtin_mips_addq_s_w (q31, q31)
6843 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
6844 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
6845 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
6846 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
6847 q31 __builtin_mips_subq_s_w (q31, q31)
6848 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
6849 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
6850 i32 __builtin_mips_addsc (i32, i32)
6851 i32 __builtin_mips_addwc (i32, i32)
6852 i32 __builtin_mips_modsub (i32, i32)
6853 i32 __builtin_mips_raddu_w_qb (v4i8)
6854 v2q15 __builtin_mips_absq_s_ph (v2q15)
6855 q31 __builtin_mips_absq_s_w (q31)
6856 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
6857 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
6858 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
6859 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
6860 q31 __builtin_mips_preceq_w_phl (v2q15)
6861 q31 __builtin_mips_preceq_w_phr (v2q15)
6862 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
6863 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
6864 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
6865 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
6866 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
6867 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
6868 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
6869 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
6870 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
6871 v4i8 __builtin_mips_shll_qb (v4i8, i32)
6872 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
6873 v2q15 __builtin_mips_shll_ph (v2q15, i32)
6874 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
6875 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
6876 q31 __builtin_mips_shll_s_w (q31, imm0_31)
6877 q31 __builtin_mips_shll_s_w (q31, i32)
6878 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
6879 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
6880 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
6881 v2q15 __builtin_mips_shra_ph (v2q15, i32)
6882 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
6883 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
6884 q31 __builtin_mips_shra_r_w (q31, imm0_31)
6885 q31 __builtin_mips_shra_r_w (q31, i32)
6886 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
6887 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
6888 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
6889 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
6890 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
6891 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
6892 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
6893 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
6894 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
6895 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
6896 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
6897 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
6898 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
6899 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
6900 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
6901 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
6902 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
6903 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
6904 i32 __builtin_mips_bitrev (i32)
6905 i32 __builtin_mips_insv (i32, i32)
6906 v4i8 __builtin_mips_repl_qb (imm0_255)
6907 v4i8 __builtin_mips_repl_qb (i32)
6908 v2q15 __builtin_mips_repl_ph (imm_n512_511)
6909 v2q15 __builtin_mips_repl_ph (i32)
6910 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
6911 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
6912 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
6913 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
6914 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
6915 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
6916 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
6917 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
6918 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
6919 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
6920 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
6921 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
6922 i32 __builtin_mips_extr_w (a64, imm0_31)
6923 i32 __builtin_mips_extr_w (a64, i32)
6924 i32 __builtin_mips_extr_r_w (a64, imm0_31)
6925 i32 __builtin_mips_extr_s_h (a64, i32)
6926 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
6927 i32 __builtin_mips_extr_rs_w (a64, i32)
6928 i32 __builtin_mips_extr_s_h (a64, imm0_31)
6929 i32 __builtin_mips_extr_r_w (a64, i32)
6930 i32 __builtin_mips_extp (a64, imm0_31)
6931 i32 __builtin_mips_extp (a64, i32)
6932 i32 __builtin_mips_extpdp (a64, imm0_31)
6933 i32 __builtin_mips_extpdp (a64, i32)
6934 a64 __builtin_mips_shilo (a64, imm_n32_31)
6935 a64 __builtin_mips_shilo (a64, i32)
6936 a64 __builtin_mips_mthlip (a64, i32)
6937 void __builtin_mips_wrdsp (i32, imm0_63)
6938 i32 __builtin_mips_rddsp (imm0_63)
6939 i32 __builtin_mips_lbux (void *, i32)
6940 i32 __builtin_mips_lhx (void *, i32)
6941 i32 __builtin_mips_lwx (void *, i32)
6942 i32 __builtin_mips_bposge32 (void)
6945 @node MIPS Paired-Single Support
6946 @subsection MIPS Paired-Single Support
6948 The MIPS64 architecture includes a number of instructions that
6949 operate on pairs of single-precision floating-point values.
6950 Each pair is packed into a 64-bit floating-point register,
6951 with one element being designated the ``upper half'' and
6952 the other being designated the ``lower half''.
6954 GCC supports paired-single operations using both the generic
6955 vector extensions (@pxref{Vector Extensions}) and a collection of
6956 MIPS-specific built-in functions. Both kinds of support are
6957 enabled by the @option{-mpaired-single} command-line option.
6959 The vector type associated with paired-single values is usually
6960 called @code{v2sf}. It can be defined in C as follows:
6963 typedef float v2sf __attribute__ ((vector_size (8)));
6966 @code{v2sf} values are initialized in the same way as aggregates.
6970 v2sf a = @{1.5, 9.1@};
6973 b = (v2sf) @{e, f@};
6976 @emph{Note:} The CPU's endianness determines which value is stored in
6977 the upper half of a register and which value is stored in the lower half.
6978 On little-endian targets, the first value is the lower one and the second
6979 value is the upper one. The opposite order applies to big-endian targets.
6980 For example, the code above will set the lower half of @code{a} to
6981 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
6984 * Paired-Single Arithmetic::
6985 * Paired-Single Built-in Functions::
6986 * MIPS-3D Built-in Functions::
6989 @node Paired-Single Arithmetic
6990 @subsubsection Paired-Single Arithmetic
6992 The table below lists the @code{v2sf} operations for which hardware
6993 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
6994 values and @code{x} is an integral value.
6996 @multitable @columnfractions .50 .50
6997 @item C code @tab MIPS instruction
6998 @item @code{a + b} @tab @code{add.ps}
6999 @item @code{a - b} @tab @code{sub.ps}
7000 @item @code{-a} @tab @code{neg.ps}
7001 @item @code{a * b} @tab @code{mul.ps}
7002 @item @code{a * b + c} @tab @code{madd.ps}
7003 @item @code{a * b - c} @tab @code{msub.ps}
7004 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
7005 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
7006 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
7009 Note that the multiply-accumulate instructions can be disabled
7010 using the command-line option @code{-mno-fused-madd}.
7012 @node Paired-Single Built-in Functions
7013 @subsubsection Paired-Single Built-in Functions
7015 The following paired-single functions map directly to a particular
7016 MIPS instruction. Please refer to the architecture specification
7017 for details on what each instruction does.
7020 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
7021 Pair lower lower (@code{pll.ps}).
7023 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
7024 Pair upper lower (@code{pul.ps}).
7026 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
7027 Pair lower upper (@code{plu.ps}).
7029 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
7030 Pair upper upper (@code{puu.ps}).
7032 @item v2sf __builtin_mips_cvt_ps_s (float, float)
7033 Convert pair to paired single (@code{cvt.ps.s}).
7035 @item float __builtin_mips_cvt_s_pl (v2sf)
7036 Convert pair lower to single (@code{cvt.s.pl}).
7038 @item float __builtin_mips_cvt_s_pu (v2sf)
7039 Convert pair upper to single (@code{cvt.s.pu}).
7041 @item v2sf __builtin_mips_abs_ps (v2sf)
7042 Absolute value (@code{abs.ps}).
7044 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
7045 Align variable (@code{alnv.ps}).
7047 @emph{Note:} The value of the third parameter must be 0 or 4
7048 modulo 8, otherwise the result will be unpredictable. Please read the
7049 instruction description for details.
7052 The following multi-instruction functions are also available.
7053 In each case, @var{cond} can be any of the 16 floating-point conditions:
7054 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
7055 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
7056 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
7059 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7060 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7061 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
7062 @code{movt.ps}/@code{movf.ps}).
7064 The @code{movt} functions return the value @var{x} computed by:
7067 c.@var{cond}.ps @var{cc},@var{a},@var{b}
7068 mov.ps @var{x},@var{c}
7069 movt.ps @var{x},@var{d},@var{cc}
7072 The @code{movf} functions are similar but use @code{movf.ps} instead
7075 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7076 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7077 Comparison of two paired-single values (@code{c.@var{cond}.ps},
7078 @code{bc1t}/@code{bc1f}).
7080 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
7081 and return either the upper or lower half of the result. For example:
7085 if (__builtin_mips_upper_c_eq_ps (a, b))
7086 upper_halves_are_equal ();
7088 upper_halves_are_unequal ();
7090 if (__builtin_mips_lower_c_eq_ps (a, b))
7091 lower_halves_are_equal ();
7093 lower_halves_are_unequal ();
7097 @node MIPS-3D Built-in Functions
7098 @subsubsection MIPS-3D Built-in Functions
7100 The MIPS-3D Application-Specific Extension (ASE) includes additional
7101 paired-single instructions that are designed to improve the performance
7102 of 3D graphics operations. Support for these instructions is controlled
7103 by the @option{-mips3d} command-line option.
7105 The functions listed below map directly to a particular MIPS-3D
7106 instruction. Please refer to the architecture specification for
7107 more details on what each instruction does.
7110 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
7111 Reduction add (@code{addr.ps}).
7113 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
7114 Reduction multiply (@code{mulr.ps}).
7116 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
7117 Convert paired single to paired word (@code{cvt.pw.ps}).
7119 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
7120 Convert paired word to paired single (@code{cvt.ps.pw}).
7122 @item float __builtin_mips_recip1_s (float)
7123 @itemx double __builtin_mips_recip1_d (double)
7124 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
7125 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
7127 @item float __builtin_mips_recip2_s (float, float)
7128 @itemx double __builtin_mips_recip2_d (double, double)
7129 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
7130 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
7132 @item float __builtin_mips_rsqrt1_s (float)
7133 @itemx double __builtin_mips_rsqrt1_d (double)
7134 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
7135 Reduced precision reciprocal square root (sequence step 1)
7136 (@code{rsqrt1.@var{fmt}}).
7138 @item float __builtin_mips_rsqrt2_s (float, float)
7139 @itemx double __builtin_mips_rsqrt2_d (double, double)
7140 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
7141 Reduced precision reciprocal square root (sequence step 2)
7142 (@code{rsqrt2.@var{fmt}}).
7145 The following multi-instruction functions are also available.
7146 In each case, @var{cond} can be any of the 16 floating-point conditions:
7147 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
7148 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
7149 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
7152 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
7153 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
7154 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
7155 @code{bc1t}/@code{bc1f}).
7157 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
7158 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
7163 if (__builtin_mips_cabs_eq_s (a, b))
7169 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7170 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7171 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
7172 @code{bc1t}/@code{bc1f}).
7174 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
7175 and return either the upper or lower half of the result. For example:
7179 if (__builtin_mips_upper_cabs_eq_ps (a, b))
7180 upper_halves_are_equal ();
7182 upper_halves_are_unequal ();
7184 if (__builtin_mips_lower_cabs_eq_ps (a, b))
7185 lower_halves_are_equal ();
7187 lower_halves_are_unequal ();
7190 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7191 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7192 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
7193 @code{movt.ps}/@code{movf.ps}).
7195 The @code{movt} functions return the value @var{x} computed by:
7198 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
7199 mov.ps @var{x},@var{c}
7200 movt.ps @var{x},@var{d},@var{cc}
7203 The @code{movf} functions are similar but use @code{movf.ps} instead
7206 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7207 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7208 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7209 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
7210 Comparison of two paired-single values
7211 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
7212 @code{bc1any2t}/@code{bc1any2f}).
7214 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
7215 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
7216 result is true and the @code{all} forms return true if both results are true.
7221 if (__builtin_mips_any_c_eq_ps (a, b))
7226 if (__builtin_mips_all_c_eq_ps (a, b))
7232 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7233 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7234 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7235 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
7236 Comparison of four paired-single values
7237 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
7238 @code{bc1any4t}/@code{bc1any4f}).
7240 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
7241 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
7242 The @code{any} forms return true if any of the four results are true
7243 and the @code{all} forms return true if all four results are true.
7248 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
7253 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
7260 @node PowerPC AltiVec Built-in Functions
7261 @subsection PowerPC AltiVec Built-in Functions
7263 GCC provides an interface for the PowerPC family of processors to access
7264 the AltiVec operations described in Motorola's AltiVec Programming
7265 Interface Manual. The interface is made available by including
7266 @code{<altivec.h>} and using @option{-maltivec} and
7267 @option{-mabi=altivec}. The interface supports the following vector
7271 vector unsigned char
7275 vector unsigned short
7286 GCC's implementation of the high-level language interface available from
7287 C and C++ code differs from Motorola's documentation in several ways.
7292 A vector constant is a list of constant expressions within curly braces.
7295 A vector initializer requires no cast if the vector constant is of the
7296 same type as the variable it is initializing.
7299 If @code{signed} or @code{unsigned} is omitted, the signedness of the
7300 vector type is the default signedness of the base type. The default
7301 varies depending on the operating system, so a portable program should
7302 always specify the signedness.
7305 Compiling with @option{-maltivec} adds keywords @code{__vector},
7306 @code{__pixel}, and @code{__bool}. Macros @option{vector},
7307 @code{pixel}, and @code{bool} are defined in @code{<altivec.h>} and can
7311 GCC allows using a @code{typedef} name as the type specifier for a
7315 For C, overloaded functions are implemented with macros so the following
7319 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
7322 Since @code{vec_add} is a macro, the vector constant in the example
7323 is treated as four separate arguments. Wrap the entire argument in
7324 parentheses for this to work.
7327 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
7328 Internally, GCC uses built-in functions to achieve the functionality in
7329 the aforementioned header file, but they are not supported and are
7330 subject to change without notice.
7332 The following interfaces are supported for the generic and specific
7333 AltiVec operations and the AltiVec predicates. In cases where there
7334 is a direct mapping between generic and specific operations, only the
7335 generic names are shown here, although the specific operations can also
7338 Arguments that are documented as @code{const int} require literal
7339 integral values within the range required for that operation.
7342 vector signed char vec_abs (vector signed char);
7343 vector signed short vec_abs (vector signed short);
7344 vector signed int vec_abs (vector signed int);
7345 vector float vec_abs (vector float);
7347 vector signed char vec_abss (vector signed char);
7348 vector signed short vec_abss (vector signed short);
7349 vector signed int vec_abss (vector signed int);
7351 vector signed char vec_add (vector bool char, vector signed char);
7352 vector signed char vec_add (vector signed char, vector bool char);
7353 vector signed char vec_add (vector signed char, vector signed char);
7354 vector unsigned char vec_add (vector bool char, vector unsigned char);
7355 vector unsigned char vec_add (vector unsigned char, vector bool char);
7356 vector unsigned char vec_add (vector unsigned char,
7357 vector unsigned char);
7358 vector signed short vec_add (vector bool short, vector signed short);
7359 vector signed short vec_add (vector signed short, vector bool short);
7360 vector signed short vec_add (vector signed short, vector signed short);
7361 vector unsigned short vec_add (vector bool short,
7362 vector unsigned short);
7363 vector unsigned short vec_add (vector unsigned short,
7365 vector unsigned short vec_add (vector unsigned short,
7366 vector unsigned short);
7367 vector signed int vec_add (vector bool int, vector signed int);
7368 vector signed int vec_add (vector signed int, vector bool int);
7369 vector signed int vec_add (vector signed int, vector signed int);
7370 vector unsigned int vec_add (vector bool int, vector unsigned int);
7371 vector unsigned int vec_add (vector unsigned int, vector bool int);
7372 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
7373 vector float vec_add (vector float, vector float);
7375 vector float vec_vaddfp (vector float, vector float);
7377 vector signed int vec_vadduwm (vector bool int, vector signed int);
7378 vector signed int vec_vadduwm (vector signed int, vector bool int);
7379 vector signed int vec_vadduwm (vector signed int, vector signed int);
7380 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
7381 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
7382 vector unsigned int vec_vadduwm (vector unsigned int,
7383 vector unsigned int);
7385 vector signed short vec_vadduhm (vector bool short,
7386 vector signed short);
7387 vector signed short vec_vadduhm (vector signed short,
7389 vector signed short vec_vadduhm (vector signed short,
7390 vector signed short);
7391 vector unsigned short vec_vadduhm (vector bool short,
7392 vector unsigned short);
7393 vector unsigned short vec_vadduhm (vector unsigned short,
7395 vector unsigned short vec_vadduhm (vector unsigned short,
7396 vector unsigned short);
7398 vector signed char vec_vaddubm (vector bool char, vector signed char);
7399 vector signed char vec_vaddubm (vector signed char, vector bool char);
7400 vector signed char vec_vaddubm (vector signed char, vector signed char);
7401 vector unsigned char vec_vaddubm (vector bool char,
7402 vector unsigned char);
7403 vector unsigned char vec_vaddubm (vector unsigned char,
7405 vector unsigned char vec_vaddubm (vector unsigned char,
7406 vector unsigned char);
7408 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
7410 vector unsigned char vec_adds (vector bool char, vector unsigned char);
7411 vector unsigned char vec_adds (vector unsigned char, vector bool char);
7412 vector unsigned char vec_adds (vector unsigned char,
7413 vector unsigned char);
7414 vector signed char vec_adds (vector bool char, vector signed char);
7415 vector signed char vec_adds (vector signed char, vector bool char);
7416 vector signed char vec_adds (vector signed char, vector signed char);
7417 vector unsigned short vec_adds (vector bool short,
7418 vector unsigned short);
7419 vector unsigned short vec_adds (vector unsigned short,
7421 vector unsigned short vec_adds (vector unsigned short,
7422 vector unsigned short);
7423 vector signed short vec_adds (vector bool short, vector signed short);
7424 vector signed short vec_adds (vector signed short, vector bool short);
7425 vector signed short vec_adds (vector signed short, vector signed short);
7426 vector unsigned int vec_adds (vector bool int, vector unsigned int);
7427 vector unsigned int vec_adds (vector unsigned int, vector bool int);
7428 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
7429 vector signed int vec_adds (vector bool int, vector signed int);
7430 vector signed int vec_adds (vector signed int, vector bool int);
7431 vector signed int vec_adds (vector signed int, vector signed int);
7433 vector signed int vec_vaddsws (vector bool int, vector signed int);
7434 vector signed int vec_vaddsws (vector signed int, vector bool int);
7435 vector signed int vec_vaddsws (vector signed int, vector signed int);
7437 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
7438 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
7439 vector unsigned int vec_vadduws (vector unsigned int,
7440 vector unsigned int);
7442 vector signed short vec_vaddshs (vector bool short,
7443 vector signed short);
7444 vector signed short vec_vaddshs (vector signed short,
7446 vector signed short vec_vaddshs (vector signed short,
7447 vector signed short);
7449 vector unsigned short vec_vadduhs (vector bool short,
7450 vector unsigned short);
7451 vector unsigned short vec_vadduhs (vector unsigned short,
7453 vector unsigned short vec_vadduhs (vector unsigned short,
7454 vector unsigned short);
7456 vector signed char vec_vaddsbs (vector bool char, vector signed char);
7457 vector signed char vec_vaddsbs (vector signed char, vector bool char);
7458 vector signed char vec_vaddsbs (vector signed char, vector signed char);
7460 vector unsigned char vec_vaddubs (vector bool char,
7461 vector unsigned char);
7462 vector unsigned char vec_vaddubs (vector unsigned char,
7464 vector unsigned char vec_vaddubs (vector unsigned char,
7465 vector unsigned char);
7467 vector float vec_and (vector float, vector float);
7468 vector float vec_and (vector float, vector bool int);
7469 vector float vec_and (vector bool int, vector float);
7470 vector bool int vec_and (vector bool int, vector bool int);
7471 vector signed int vec_and (vector bool int, vector signed int);
7472 vector signed int vec_and (vector signed int, vector bool int);
7473 vector signed int vec_and (vector signed int, vector signed int);
7474 vector unsigned int vec_and (vector bool int, vector unsigned int);
7475 vector unsigned int vec_and (vector unsigned int, vector bool int);
7476 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
7477 vector bool short vec_and (vector bool short, vector bool short);
7478 vector signed short vec_and (vector bool short, vector signed short);
7479 vector signed short vec_and (vector signed short, vector bool short);
7480 vector signed short vec_and (vector signed short, vector signed short);
7481 vector unsigned short vec_and (vector bool short,
7482 vector unsigned short);
7483 vector unsigned short vec_and (vector unsigned short,
7485 vector unsigned short vec_and (vector unsigned short,
7486 vector unsigned short);
7487 vector signed char vec_and (vector bool char, vector signed char);
7488 vector bool char vec_and (vector bool char, vector bool char);
7489 vector signed char vec_and (vector signed char, vector bool char);
7490 vector signed char vec_and (vector signed char, vector signed char);
7491 vector unsigned char vec_and (vector bool char, vector unsigned char);
7492 vector unsigned char vec_and (vector unsigned char, vector bool char);
7493 vector unsigned char vec_and (vector unsigned char,
7494 vector unsigned char);
7496 vector float vec_andc (vector float, vector float);
7497 vector float vec_andc (vector float, vector bool int);
7498 vector float vec_andc (vector bool int, vector float);
7499 vector bool int vec_andc (vector bool int, vector bool int);
7500 vector signed int vec_andc (vector bool int, vector signed int);
7501 vector signed int vec_andc (vector signed int, vector bool int);
7502 vector signed int vec_andc (vector signed int, vector signed int);
7503 vector unsigned int vec_andc (vector bool int, vector unsigned int);
7504 vector unsigned int vec_andc (vector unsigned int, vector bool int);
7505 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
7506 vector bool short vec_andc (vector bool short, vector bool short);
7507 vector signed short vec_andc (vector bool short, vector signed short);
7508 vector signed short vec_andc (vector signed short, vector bool short);
7509 vector signed short vec_andc (vector signed short, vector signed short);
7510 vector unsigned short vec_andc (vector bool short,
7511 vector unsigned short);
7512 vector unsigned short vec_andc (vector unsigned short,
7514 vector unsigned short vec_andc (vector unsigned short,
7515 vector unsigned short);
7516 vector signed char vec_andc (vector bool char, vector signed char);
7517 vector bool char vec_andc (vector bool char, vector bool char);
7518 vector signed char vec_andc (vector signed char, vector bool char);
7519 vector signed char vec_andc (vector signed char, vector signed char);
7520 vector unsigned char vec_andc (vector bool char, vector unsigned char);
7521 vector unsigned char vec_andc (vector unsigned char, vector bool char);
7522 vector unsigned char vec_andc (vector unsigned char,
7523 vector unsigned char);
7525 vector unsigned char vec_avg (vector unsigned char,
7526 vector unsigned char);
7527 vector signed char vec_avg (vector signed char, vector signed char);
7528 vector unsigned short vec_avg (vector unsigned short,
7529 vector unsigned short);
7530 vector signed short vec_avg (vector signed short, vector signed short);
7531 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
7532 vector signed int vec_avg (vector signed int, vector signed int);
7534 vector signed int vec_vavgsw (vector signed int, vector signed int);
7536 vector unsigned int vec_vavguw (vector unsigned int,
7537 vector unsigned int);
7539 vector signed short vec_vavgsh (vector signed short,
7540 vector signed short);
7542 vector unsigned short vec_vavguh (vector unsigned short,
7543 vector unsigned short);
7545 vector signed char vec_vavgsb (vector signed char, vector signed char);
7547 vector unsigned char vec_vavgub (vector unsigned char,
7548 vector unsigned char);
7550 vector float vec_ceil (vector float);
7552 vector signed int vec_cmpb (vector float, vector float);
7554 vector bool char vec_cmpeq (vector signed char, vector signed char);
7555 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
7556 vector bool short vec_cmpeq (vector signed short, vector signed short);
7557 vector bool short vec_cmpeq (vector unsigned short,
7558 vector unsigned short);
7559 vector bool int vec_cmpeq (vector signed int, vector signed int);
7560 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
7561 vector bool int vec_cmpeq (vector float, vector float);
7563 vector bool int vec_vcmpeqfp (vector float, vector float);
7565 vector bool int vec_vcmpequw (vector signed int, vector signed int);
7566 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
7568 vector bool short vec_vcmpequh (vector signed short,
7569 vector signed short);
7570 vector bool short vec_vcmpequh (vector unsigned short,
7571 vector unsigned short);
7573 vector bool char vec_vcmpequb (vector signed char, vector signed char);
7574 vector bool char vec_vcmpequb (vector unsigned char,
7575 vector unsigned char);
7577 vector bool int vec_cmpge (vector float, vector float);
7579 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
7580 vector bool char vec_cmpgt (vector signed char, vector signed char);
7581 vector bool short vec_cmpgt (vector unsigned short,
7582 vector unsigned short);
7583 vector bool short vec_cmpgt (vector signed short, vector signed short);
7584 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
7585 vector bool int vec_cmpgt (vector signed int, vector signed int);
7586 vector bool int vec_cmpgt (vector float, vector float);
7588 vector bool int vec_vcmpgtfp (vector float, vector float);
7590 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
7592 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
7594 vector bool short vec_vcmpgtsh (vector signed short,
7595 vector signed short);
7597 vector bool short vec_vcmpgtuh (vector unsigned short,
7598 vector unsigned short);
7600 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
7602 vector bool char vec_vcmpgtub (vector unsigned char,
7603 vector unsigned char);
7605 vector bool int vec_cmple (vector float, vector float);
7607 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
7608 vector bool char vec_cmplt (vector signed char, vector signed char);
7609 vector bool short vec_cmplt (vector unsigned short,
7610 vector unsigned short);
7611 vector bool short vec_cmplt (vector signed short, vector signed short);
7612 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
7613 vector bool int vec_cmplt (vector signed int, vector signed int);
7614 vector bool int vec_cmplt (vector float, vector float);
7616 vector float vec_ctf (vector unsigned int, const int);
7617 vector float vec_ctf (vector signed int, const int);
7619 vector float vec_vcfsx (vector signed int, const int);
7621 vector float vec_vcfux (vector unsigned int, const int);
7623 vector signed int vec_cts (vector float, const int);
7625 vector unsigned int vec_ctu (vector float, const int);
7627 void vec_dss (const int);
7629 void vec_dssall (void);
7631 void vec_dst (const vector unsigned char *, int, const int);
7632 void vec_dst (const vector signed char *, int, const int);
7633 void vec_dst (const vector bool char *, int, const int);
7634 void vec_dst (const vector unsigned short *, int, const int);
7635 void vec_dst (const vector signed short *, int, const int);
7636 void vec_dst (const vector bool short *, int, const int);
7637 void vec_dst (const vector pixel *, int, const int);
7638 void vec_dst (const vector unsigned int *, int, const int);
7639 void vec_dst (const vector signed int *, int, const int);
7640 void vec_dst (const vector bool int *, int, const int);
7641 void vec_dst (const vector float *, int, const int);
7642 void vec_dst (const unsigned char *, int, const int);
7643 void vec_dst (const signed char *, int, const int);
7644 void vec_dst (const unsigned short *, int, const int);
7645 void vec_dst (const short *, int, const int);
7646 void vec_dst (const unsigned int *, int, const int);
7647 void vec_dst (const int *, int, const int);
7648 void vec_dst (const unsigned long *, int, const int);
7649 void vec_dst (const long *, int, const int);
7650 void vec_dst (const float *, int, const int);
7652 void vec_dstst (const vector unsigned char *, int, const int);
7653 void vec_dstst (const vector signed char *, int, const int);
7654 void vec_dstst (const vector bool char *, int, const int);
7655 void vec_dstst (const vector unsigned short *, int, const int);
7656 void vec_dstst (const vector signed short *, int, const int);
7657 void vec_dstst (const vector bool short *, int, const int);
7658 void vec_dstst (const vector pixel *, int, const int);
7659 void vec_dstst (const vector unsigned int *, int, const int);
7660 void vec_dstst (const vector signed int *, int, const int);
7661 void vec_dstst (const vector bool int *, int, const int);
7662 void vec_dstst (const vector float *, int, const int);
7663 void vec_dstst (const unsigned char *, int, const int);
7664 void vec_dstst (const signed char *, int, const int);
7665 void vec_dstst (const unsigned short *, int, const int);
7666 void vec_dstst (const short *, int, const int);
7667 void vec_dstst (const unsigned int *, int, const int);
7668 void vec_dstst (const int *, int, const int);
7669 void vec_dstst (const unsigned long *, int, const int);
7670 void vec_dstst (const long *, int, const int);
7671 void vec_dstst (const float *, int, const int);
7673 void vec_dststt (const vector unsigned char *, int, const int);
7674 void vec_dststt (const vector signed char *, int, const int);
7675 void vec_dststt (const vector bool char *, int, const int);
7676 void vec_dststt (const vector unsigned short *, int, const int);
7677 void vec_dststt (const vector signed short *, int, const int);
7678 void vec_dststt (const vector bool short *, int, const int);
7679 void vec_dststt (const vector pixel *, int, const int);
7680 void vec_dststt (const vector unsigned int *, int, const int);
7681 void vec_dststt (const vector signed int *, int, const int);
7682 void vec_dststt (const vector bool int *, int, const int);
7683 void vec_dststt (const vector float *, int, const int);
7684 void vec_dststt (const unsigned char *, int, const int);
7685 void vec_dststt (const signed char *, int, const int);
7686 void vec_dststt (const unsigned short *, int, const int);
7687 void vec_dststt (const short *, int, const int);
7688 void vec_dststt (const unsigned int *, int, const int);
7689 void vec_dststt (const int *, int, const int);
7690 void vec_dststt (const unsigned long *, int, const int);
7691 void vec_dststt (const long *, int, const int);
7692 void vec_dststt (const float *, int, const int);
7694 void vec_dstt (const vector unsigned char *, int, const int);
7695 void vec_dstt (const vector signed char *, int, const int);
7696 void vec_dstt (const vector bool char *, int, const int);
7697 void vec_dstt (const vector unsigned short *, int, const int);
7698 void vec_dstt (const vector signed short *, int, const int);
7699 void vec_dstt (const vector bool short *, int, const int);
7700 void vec_dstt (const vector pixel *, int, const int);
7701 void vec_dstt (const vector unsigned int *, int, const int);
7702 void vec_dstt (const vector signed int *, int, const int);
7703 void vec_dstt (const vector bool int *, int, const int);
7704 void vec_dstt (const vector float *, int, const int);
7705 void vec_dstt (const unsigned char *, int, const int);
7706 void vec_dstt (const signed char *, int, const int);
7707 void vec_dstt (const unsigned short *, int, const int);
7708 void vec_dstt (const short *, int, const int);
7709 void vec_dstt (const unsigned int *, int, const int);
7710 void vec_dstt (const int *, int, const int);
7711 void vec_dstt (const unsigned long *, int, const int);
7712 void vec_dstt (const long *, int, const int);
7713 void vec_dstt (const float *, int, const int);
7715 vector float vec_expte (vector float);
7717 vector float vec_floor (vector float);
7719 vector float vec_ld (int, const vector float *);
7720 vector float vec_ld (int, const float *);
7721 vector bool int vec_ld (int, const vector bool int *);
7722 vector signed int vec_ld (int, const vector signed int *);
7723 vector signed int vec_ld (int, const int *);
7724 vector signed int vec_ld (int, const long *);
7725 vector unsigned int vec_ld (int, const vector unsigned int *);
7726 vector unsigned int vec_ld (int, const unsigned int *);
7727 vector unsigned int vec_ld (int, const unsigned long *);
7728 vector bool short vec_ld (int, const vector bool short *);
7729 vector pixel vec_ld (int, const vector pixel *);
7730 vector signed short vec_ld (int, const vector signed short *);
7731 vector signed short vec_ld (int, const short *);
7732 vector unsigned short vec_ld (int, const vector unsigned short *);
7733 vector unsigned short vec_ld (int, const unsigned short *);
7734 vector bool char vec_ld (int, const vector bool char *);
7735 vector signed char vec_ld (int, const vector signed char *);
7736 vector signed char vec_ld (int, const signed char *);
7737 vector unsigned char vec_ld (int, const vector unsigned char *);
7738 vector unsigned char vec_ld (int, const unsigned char *);
7740 vector signed char vec_lde (int, const signed char *);
7741 vector unsigned char vec_lde (int, const unsigned char *);
7742 vector signed short vec_lde (int, const short *);
7743 vector unsigned short vec_lde (int, const unsigned short *);
7744 vector float vec_lde (int, const float *);
7745 vector signed int vec_lde (int, const int *);
7746 vector unsigned int vec_lde (int, const unsigned int *);
7747 vector signed int vec_lde (int, const long *);
7748 vector unsigned int vec_lde (int, const unsigned long *);
7750 vector float vec_lvewx (int, float *);
7751 vector signed int vec_lvewx (int, int *);
7752 vector unsigned int vec_lvewx (int, unsigned int *);
7753 vector signed int vec_lvewx (int, long *);
7754 vector unsigned int vec_lvewx (int, unsigned long *);
7756 vector signed short vec_lvehx (int, short *);
7757 vector unsigned short vec_lvehx (int, unsigned short *);
7759 vector signed char vec_lvebx (int, char *);
7760 vector unsigned char vec_lvebx (int, unsigned char *);
7762 vector float vec_ldl (int, const vector float *);
7763 vector float vec_ldl (int, const float *);
7764 vector bool int vec_ldl (int, const vector bool int *);
7765 vector signed int vec_ldl (int, const vector signed int *);
7766 vector signed int vec_ldl (int, const int *);
7767 vector signed int vec_ldl (int, const long *);
7768 vector unsigned int vec_ldl (int, const vector unsigned int *);
7769 vector unsigned int vec_ldl (int, const unsigned int *);
7770 vector unsigned int vec_ldl (int, const unsigned long *);
7771 vector bool short vec_ldl (int, const vector bool short *);
7772 vector pixel vec_ldl (int, const vector pixel *);
7773 vector signed short vec_ldl (int, const vector signed short *);
7774 vector signed short vec_ldl (int, const short *);
7775 vector unsigned short vec_ldl (int, const vector unsigned short *);
7776 vector unsigned short vec_ldl (int, const unsigned short *);
7777 vector bool char vec_ldl (int, const vector bool char *);
7778 vector signed char vec_ldl (int, const vector signed char *);
7779 vector signed char vec_ldl (int, const signed char *);
7780 vector unsigned char vec_ldl (int, const vector unsigned char *);
7781 vector unsigned char vec_ldl (int, const unsigned char *);
7783 vector float vec_loge (vector float);
7785 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
7786 vector unsigned char vec_lvsl (int, const volatile signed char *);
7787 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
7788 vector unsigned char vec_lvsl (int, const volatile short *);
7789 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
7790 vector unsigned char vec_lvsl (int, const volatile int *);
7791 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
7792 vector unsigned char vec_lvsl (int, const volatile long *);
7793 vector unsigned char vec_lvsl (int, const volatile float *);
7795 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
7796 vector unsigned char vec_lvsr (int, const volatile signed char *);
7797 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
7798 vector unsigned char vec_lvsr (int, const volatile short *);
7799 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
7800 vector unsigned char vec_lvsr (int, const volatile int *);
7801 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
7802 vector unsigned char vec_lvsr (int, const volatile long *);
7803 vector unsigned char vec_lvsr (int, const volatile float *);
7805 vector float vec_madd (vector float, vector float, vector float);
7807 vector signed short vec_madds (vector signed short,
7808 vector signed short,
7809 vector signed short);
7811 vector unsigned char vec_max (vector bool char, vector unsigned char);
7812 vector unsigned char vec_max (vector unsigned char, vector bool char);
7813 vector unsigned char vec_max (vector unsigned char,
7814 vector unsigned char);
7815 vector signed char vec_max (vector bool char, vector signed char);
7816 vector signed char vec_max (vector signed char, vector bool char);
7817 vector signed char vec_max (vector signed char, vector signed char);
7818 vector unsigned short vec_max (vector bool short,
7819 vector unsigned short);
7820 vector unsigned short vec_max (vector unsigned short,
7822 vector unsigned short vec_max (vector unsigned short,
7823 vector unsigned short);
7824 vector signed short vec_max (vector bool short, vector signed short);
7825 vector signed short vec_max (vector signed short, vector bool short);
7826 vector signed short vec_max (vector signed short, vector signed short);
7827 vector unsigned int vec_max (vector bool int, vector unsigned int);
7828 vector unsigned int vec_max (vector unsigned int, vector bool int);
7829 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
7830 vector signed int vec_max (vector bool int, vector signed int);
7831 vector signed int vec_max (vector signed int, vector bool int);
7832 vector signed int vec_max (vector signed int, vector signed int);
7833 vector float vec_max (vector float, vector float);
7835 vector float vec_vmaxfp (vector float, vector float);
7837 vector signed int vec_vmaxsw (vector bool int, vector signed int);
7838 vector signed int vec_vmaxsw (vector signed int, vector bool int);
7839 vector signed int vec_vmaxsw (vector signed int, vector signed int);
7841 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
7842 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
7843 vector unsigned int vec_vmaxuw (vector unsigned int,
7844 vector unsigned int);
7846 vector signed short vec_vmaxsh (vector bool short, vector signed short);
7847 vector signed short vec_vmaxsh (vector signed short, vector bool short);
7848 vector signed short vec_vmaxsh (vector signed short,
7849 vector signed short);
7851 vector unsigned short vec_vmaxuh (vector bool short,
7852 vector unsigned short);
7853 vector unsigned short vec_vmaxuh (vector unsigned short,
7855 vector unsigned short vec_vmaxuh (vector unsigned short,
7856 vector unsigned short);
7858 vector signed char vec_vmaxsb (vector bool char, vector signed char);
7859 vector signed char vec_vmaxsb (vector signed char, vector bool char);
7860 vector signed char vec_vmaxsb (vector signed char, vector signed char);
7862 vector unsigned char vec_vmaxub (vector bool char,
7863 vector unsigned char);
7864 vector unsigned char vec_vmaxub (vector unsigned char,
7866 vector unsigned char vec_vmaxub (vector unsigned char,
7867 vector unsigned char);
7869 vector bool char vec_mergeh (vector bool char, vector bool char);
7870 vector signed char vec_mergeh (vector signed char, vector signed char);
7871 vector unsigned char vec_mergeh (vector unsigned char,
7872 vector unsigned char);
7873 vector bool short vec_mergeh (vector bool short, vector bool short);
7874 vector pixel vec_mergeh (vector pixel, vector pixel);
7875 vector signed short vec_mergeh (vector signed short,
7876 vector signed short);
7877 vector unsigned short vec_mergeh (vector unsigned short,
7878 vector unsigned short);
7879 vector float vec_mergeh (vector float, vector float);
7880 vector bool int vec_mergeh (vector bool int, vector bool int);
7881 vector signed int vec_mergeh (vector signed int, vector signed int);
7882 vector unsigned int vec_mergeh (vector unsigned int,
7883 vector unsigned int);
7885 vector float vec_vmrghw (vector float, vector float);
7886 vector bool int vec_vmrghw (vector bool int, vector bool int);
7887 vector signed int vec_vmrghw (vector signed int, vector signed int);
7888 vector unsigned int vec_vmrghw (vector unsigned int,
7889 vector unsigned int);
7891 vector bool short vec_vmrghh (vector bool short, vector bool short);
7892 vector signed short vec_vmrghh (vector signed short,
7893 vector signed short);
7894 vector unsigned short vec_vmrghh (vector unsigned short,
7895 vector unsigned short);
7896 vector pixel vec_vmrghh (vector pixel, vector pixel);
7898 vector bool char vec_vmrghb (vector bool char, vector bool char);
7899 vector signed char vec_vmrghb (vector signed char, vector signed char);
7900 vector unsigned char vec_vmrghb (vector unsigned char,
7901 vector unsigned char);
7903 vector bool char vec_mergel (vector bool char, vector bool char);
7904 vector signed char vec_mergel (vector signed char, vector signed char);
7905 vector unsigned char vec_mergel (vector unsigned char,
7906 vector unsigned char);
7907 vector bool short vec_mergel (vector bool short, vector bool short);
7908 vector pixel vec_mergel (vector pixel, vector pixel);
7909 vector signed short vec_mergel (vector signed short,
7910 vector signed short);
7911 vector unsigned short vec_mergel (vector unsigned short,
7912 vector unsigned short);
7913 vector float vec_mergel (vector float, vector float);
7914 vector bool int vec_mergel (vector bool int, vector bool int);
7915 vector signed int vec_mergel (vector signed int, vector signed int);
7916 vector unsigned int vec_mergel (vector unsigned int,
7917 vector unsigned int);
7919 vector float vec_vmrglw (vector float, vector float);
7920 vector signed int vec_vmrglw (vector signed int, vector signed int);
7921 vector unsigned int vec_vmrglw (vector unsigned int,
7922 vector unsigned int);
7923 vector bool int vec_vmrglw (vector bool int, vector bool int);
7925 vector bool short vec_vmrglh (vector bool short, vector bool short);
7926 vector signed short vec_vmrglh (vector signed short,
7927 vector signed short);
7928 vector unsigned short vec_vmrglh (vector unsigned short,
7929 vector unsigned short);
7930 vector pixel vec_vmrglh (vector pixel, vector pixel);
7932 vector bool char vec_vmrglb (vector bool char, vector bool char);
7933 vector signed char vec_vmrglb (vector signed char, vector signed char);
7934 vector unsigned char vec_vmrglb (vector unsigned char,
7935 vector unsigned char);
7937 vector unsigned short vec_mfvscr (void);
7939 vector unsigned char vec_min (vector bool char, vector unsigned char);
7940 vector unsigned char vec_min (vector unsigned char, vector bool char);
7941 vector unsigned char vec_min (vector unsigned char,
7942 vector unsigned char);
7943 vector signed char vec_min (vector bool char, vector signed char);
7944 vector signed char vec_min (vector signed char, vector bool char);
7945 vector signed char vec_min (vector signed char, vector signed char);
7946 vector unsigned short vec_min (vector bool short,
7947 vector unsigned short);
7948 vector unsigned short vec_min (vector unsigned short,
7950 vector unsigned short vec_min (vector unsigned short,
7951 vector unsigned short);
7952 vector signed short vec_min (vector bool short, vector signed short);
7953 vector signed short vec_min (vector signed short, vector bool short);
7954 vector signed short vec_min (vector signed short, vector signed short);
7955 vector unsigned int vec_min (vector bool int, vector unsigned int);
7956 vector unsigned int vec_min (vector unsigned int, vector bool int);
7957 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
7958 vector signed int vec_min (vector bool int, vector signed int);
7959 vector signed int vec_min (vector signed int, vector bool int);
7960 vector signed int vec_min (vector signed int, vector signed int);
7961 vector float vec_min (vector float, vector float);
7963 vector float vec_vminfp (vector float, vector float);
7965 vector signed int vec_vminsw (vector bool int, vector signed int);
7966 vector signed int vec_vminsw (vector signed int, vector bool int);
7967 vector signed int vec_vminsw (vector signed int, vector signed int);
7969 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
7970 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
7971 vector unsigned int vec_vminuw (vector unsigned int,
7972 vector unsigned int);
7974 vector signed short vec_vminsh (vector bool short, vector signed short);
7975 vector signed short vec_vminsh (vector signed short, vector bool short);
7976 vector signed short vec_vminsh (vector signed short,
7977 vector signed short);
7979 vector unsigned short vec_vminuh (vector bool short,
7980 vector unsigned short);
7981 vector unsigned short vec_vminuh (vector unsigned short,
7983 vector unsigned short vec_vminuh (vector unsigned short,
7984 vector unsigned short);
7986 vector signed char vec_vminsb (vector bool char, vector signed char);
7987 vector signed char vec_vminsb (vector signed char, vector bool char);
7988 vector signed char vec_vminsb (vector signed char, vector signed char);
7990 vector unsigned char vec_vminub (vector bool char,
7991 vector unsigned char);
7992 vector unsigned char vec_vminub (vector unsigned char,
7994 vector unsigned char vec_vminub (vector unsigned char,
7995 vector unsigned char);
7997 vector signed short vec_mladd (vector signed short,
7998 vector signed short,
7999 vector signed short);
8000 vector signed short vec_mladd (vector signed short,
8001 vector unsigned short,
8002 vector unsigned short);
8003 vector signed short vec_mladd (vector unsigned short,
8004 vector signed short,
8005 vector signed short);
8006 vector unsigned short vec_mladd (vector unsigned short,
8007 vector unsigned short,
8008 vector unsigned short);
8010 vector signed short vec_mradds (vector signed short,
8011 vector signed short,
8012 vector signed short);
8014 vector unsigned int vec_msum (vector unsigned char,
8015 vector unsigned char,
8016 vector unsigned int);
8017 vector signed int vec_msum (vector signed char,
8018 vector unsigned char,
8020 vector unsigned int vec_msum (vector unsigned short,
8021 vector unsigned short,
8022 vector unsigned int);
8023 vector signed int vec_msum (vector signed short,
8024 vector signed short,
8027 vector signed int vec_vmsumshm (vector signed short,
8028 vector signed short,
8031 vector unsigned int vec_vmsumuhm (vector unsigned short,
8032 vector unsigned short,
8033 vector unsigned int);
8035 vector signed int vec_vmsummbm (vector signed char,
8036 vector unsigned char,
8039 vector unsigned int vec_vmsumubm (vector unsigned char,
8040 vector unsigned char,
8041 vector unsigned int);
8043 vector unsigned int vec_msums (vector unsigned short,
8044 vector unsigned short,
8045 vector unsigned int);
8046 vector signed int vec_msums (vector signed short,
8047 vector signed short,
8050 vector signed int vec_vmsumshs (vector signed short,
8051 vector signed short,
8054 vector unsigned int vec_vmsumuhs (vector unsigned short,
8055 vector unsigned short,
8056 vector unsigned int);
8058 void vec_mtvscr (vector signed int);
8059 void vec_mtvscr (vector unsigned int);
8060 void vec_mtvscr (vector bool int);
8061 void vec_mtvscr (vector signed short);
8062 void vec_mtvscr (vector unsigned short);
8063 void vec_mtvscr (vector bool short);
8064 void vec_mtvscr (vector pixel);
8065 void vec_mtvscr (vector signed char);
8066 void vec_mtvscr (vector unsigned char);
8067 void vec_mtvscr (vector bool char);
8069 vector unsigned short vec_mule (vector unsigned char,
8070 vector unsigned char);
8071 vector signed short vec_mule (vector signed char,
8072 vector signed char);
8073 vector unsigned int vec_mule (vector unsigned short,
8074 vector unsigned short);
8075 vector signed int vec_mule (vector signed short, vector signed short);
8077 vector signed int vec_vmulesh (vector signed short,
8078 vector signed short);
8080 vector unsigned int vec_vmuleuh (vector unsigned short,
8081 vector unsigned short);
8083 vector signed short vec_vmulesb (vector signed char,
8084 vector signed char);
8086 vector unsigned short vec_vmuleub (vector unsigned char,
8087 vector unsigned char);
8089 vector unsigned short vec_mulo (vector unsigned char,
8090 vector unsigned char);
8091 vector signed short vec_mulo (vector signed char, vector signed char);
8092 vector unsigned int vec_mulo (vector unsigned short,
8093 vector unsigned short);
8094 vector signed int vec_mulo (vector signed short, vector signed short);
8096 vector signed int vec_vmulosh (vector signed short,
8097 vector signed short);
8099 vector unsigned int vec_vmulouh (vector unsigned short,
8100 vector unsigned short);
8102 vector signed short vec_vmulosb (vector signed char,
8103 vector signed char);
8105 vector unsigned short vec_vmuloub (vector unsigned char,
8106 vector unsigned char);
8108 vector float vec_nmsub (vector float, vector float, vector float);
8110 vector float vec_nor (vector float, vector float);
8111 vector signed int vec_nor (vector signed int, vector signed int);
8112 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
8113 vector bool int vec_nor (vector bool int, vector bool int);
8114 vector signed short vec_nor (vector signed short, vector signed short);
8115 vector unsigned short vec_nor (vector unsigned short,
8116 vector unsigned short);
8117 vector bool short vec_nor (vector bool short, vector bool short);
8118 vector signed char vec_nor (vector signed char, vector signed char);
8119 vector unsigned char vec_nor (vector unsigned char,
8120 vector unsigned char);
8121 vector bool char vec_nor (vector bool char, vector bool char);
8123 vector float vec_or (vector float, vector float);
8124 vector float vec_or (vector float, vector bool int);
8125 vector float vec_or (vector bool int, vector float);
8126 vector bool int vec_or (vector bool int, vector bool int);
8127 vector signed int vec_or (vector bool int, vector signed int);
8128 vector signed int vec_or (vector signed int, vector bool int);
8129 vector signed int vec_or (vector signed int, vector signed int);
8130 vector unsigned int vec_or (vector bool int, vector unsigned int);
8131 vector unsigned int vec_or (vector unsigned int, vector bool int);
8132 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
8133 vector bool short vec_or (vector bool short, vector bool short);
8134 vector signed short vec_or (vector bool short, vector signed short);
8135 vector signed short vec_or (vector signed short, vector bool short);
8136 vector signed short vec_or (vector signed short, vector signed short);
8137 vector unsigned short vec_or (vector bool short, vector unsigned short);
8138 vector unsigned short vec_or (vector unsigned short, vector bool short);
8139 vector unsigned short vec_or (vector unsigned short,
8140 vector unsigned short);
8141 vector signed char vec_or (vector bool char, vector signed char);
8142 vector bool char vec_or (vector bool char, vector bool char);
8143 vector signed char vec_or (vector signed char, vector bool char);
8144 vector signed char vec_or (vector signed char, vector signed char);
8145 vector unsigned char vec_or (vector bool char, vector unsigned char);
8146 vector unsigned char vec_or (vector unsigned char, vector bool char);
8147 vector unsigned char vec_or (vector unsigned char,
8148 vector unsigned char);
8150 vector signed char vec_pack (vector signed short, vector signed short);
8151 vector unsigned char vec_pack (vector unsigned short,
8152 vector unsigned short);
8153 vector bool char vec_pack (vector bool short, vector bool short);
8154 vector signed short vec_pack (vector signed int, vector signed int);
8155 vector unsigned short vec_pack (vector unsigned int,
8156 vector unsigned int);
8157 vector bool short vec_pack (vector bool int, vector bool int);
8159 vector bool short vec_vpkuwum (vector bool int, vector bool int);
8160 vector signed short vec_vpkuwum (vector signed int, vector signed int);
8161 vector unsigned short vec_vpkuwum (vector unsigned int,
8162 vector unsigned int);
8164 vector bool char vec_vpkuhum (vector bool short, vector bool short);
8165 vector signed char vec_vpkuhum (vector signed short,
8166 vector signed short);
8167 vector unsigned char vec_vpkuhum (vector unsigned short,
8168 vector unsigned short);
8170 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
8172 vector unsigned char vec_packs (vector unsigned short,
8173 vector unsigned short);
8174 vector signed char vec_packs (vector signed short, vector signed short);
8175 vector unsigned short vec_packs (vector unsigned int,
8176 vector unsigned int);
8177 vector signed short vec_packs (vector signed int, vector signed int);
8179 vector signed short vec_vpkswss (vector signed int, vector signed int);
8181 vector unsigned short vec_vpkuwus (vector unsigned int,
8182 vector unsigned int);
8184 vector signed char vec_vpkshss (vector signed short,
8185 vector signed short);
8187 vector unsigned char vec_vpkuhus (vector unsigned short,
8188 vector unsigned short);
8190 vector unsigned char vec_packsu (vector unsigned short,
8191 vector unsigned short);
8192 vector unsigned char vec_packsu (vector signed short,
8193 vector signed short);
8194 vector unsigned short vec_packsu (vector unsigned int,
8195 vector unsigned int);
8196 vector unsigned short vec_packsu (vector signed int, vector signed int);
8198 vector unsigned short vec_vpkswus (vector signed int,
8201 vector unsigned char vec_vpkshus (vector signed short,
8202 vector signed short);
8204 vector float vec_perm (vector float,
8206 vector unsigned char);
8207 vector signed int vec_perm (vector signed int,
8209 vector unsigned char);
8210 vector unsigned int vec_perm (vector unsigned int,
8211 vector unsigned int,
8212 vector unsigned char);
8213 vector bool int vec_perm (vector bool int,
8215 vector unsigned char);
8216 vector signed short vec_perm (vector signed short,
8217 vector signed short,
8218 vector unsigned char);
8219 vector unsigned short vec_perm (vector unsigned short,
8220 vector unsigned short,
8221 vector unsigned char);
8222 vector bool short vec_perm (vector bool short,
8224 vector unsigned char);
8225 vector pixel vec_perm (vector pixel,
8227 vector unsigned char);
8228 vector signed char vec_perm (vector signed char,
8230 vector unsigned char);
8231 vector unsigned char vec_perm (vector unsigned char,
8232 vector unsigned char,
8233 vector unsigned char);
8234 vector bool char vec_perm (vector bool char,
8236 vector unsigned char);
8238 vector float vec_re (vector float);
8240 vector signed char vec_rl (vector signed char,
8241 vector unsigned char);
8242 vector unsigned char vec_rl (vector unsigned char,
8243 vector unsigned char);
8244 vector signed short vec_rl (vector signed short, vector unsigned short);
8245 vector unsigned short vec_rl (vector unsigned short,
8246 vector unsigned short);
8247 vector signed int vec_rl (vector signed int, vector unsigned int);
8248 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
8250 vector signed int vec_vrlw (vector signed int, vector unsigned int);
8251 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
8253 vector signed short vec_vrlh (vector signed short,
8254 vector unsigned short);
8255 vector unsigned short vec_vrlh (vector unsigned short,
8256 vector unsigned short);
8258 vector signed char vec_vrlb (vector signed char, vector unsigned char);
8259 vector unsigned char vec_vrlb (vector unsigned char,
8260 vector unsigned char);
8262 vector float vec_round (vector float);
8264 vector float vec_rsqrte (vector float);
8266 vector float vec_sel (vector float, vector float, vector bool int);
8267 vector float vec_sel (vector float, vector float, vector unsigned int);
8268 vector signed int vec_sel (vector signed int,
8271 vector signed int vec_sel (vector signed int,
8273 vector unsigned int);
8274 vector unsigned int vec_sel (vector unsigned int,
8275 vector unsigned int,
8277 vector unsigned int vec_sel (vector unsigned int,
8278 vector unsigned int,
8279 vector unsigned int);
8280 vector bool int vec_sel (vector bool int,
8283 vector bool int vec_sel (vector bool int,
8285 vector unsigned int);
8286 vector signed short vec_sel (vector signed short,
8287 vector signed short,
8289 vector signed short vec_sel (vector signed short,
8290 vector signed short,
8291 vector unsigned short);
8292 vector unsigned short vec_sel (vector unsigned short,
8293 vector unsigned short,
8295 vector unsigned short vec_sel (vector unsigned short,
8296 vector unsigned short,
8297 vector unsigned short);
8298 vector bool short vec_sel (vector bool short,
8301 vector bool short vec_sel (vector bool short,
8303 vector unsigned short);
8304 vector signed char vec_sel (vector signed char,
8307 vector signed char vec_sel (vector signed char,
8309 vector unsigned char);
8310 vector unsigned char vec_sel (vector unsigned char,
8311 vector unsigned char,
8313 vector unsigned char vec_sel (vector unsigned char,
8314 vector unsigned char,
8315 vector unsigned char);
8316 vector bool char vec_sel (vector bool char,
8319 vector bool char vec_sel (vector bool char,
8321 vector unsigned char);
8323 vector signed char vec_sl (vector signed char,
8324 vector unsigned char);
8325 vector unsigned char vec_sl (vector unsigned char,
8326 vector unsigned char);
8327 vector signed short vec_sl (vector signed short, vector unsigned short);
8328 vector unsigned short vec_sl (vector unsigned short,
8329 vector unsigned short);
8330 vector signed int vec_sl (vector signed int, vector unsigned int);
8331 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
8333 vector signed int vec_vslw (vector signed int, vector unsigned int);
8334 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
8336 vector signed short vec_vslh (vector signed short,
8337 vector unsigned short);
8338 vector unsigned short vec_vslh (vector unsigned short,
8339 vector unsigned short);
8341 vector signed char vec_vslb (vector signed char, vector unsigned char);
8342 vector unsigned char vec_vslb (vector unsigned char,
8343 vector unsigned char);
8345 vector float vec_sld (vector float, vector float, const int);
8346 vector signed int vec_sld (vector signed int,
8349 vector unsigned int vec_sld (vector unsigned int,
8350 vector unsigned int,
8352 vector bool int vec_sld (vector bool int,
8355 vector signed short vec_sld (vector signed short,
8356 vector signed short,
8358 vector unsigned short vec_sld (vector unsigned short,
8359 vector unsigned short,
8361 vector bool short vec_sld (vector bool short,
8364 vector pixel vec_sld (vector pixel,
8367 vector signed char vec_sld (vector signed char,
8370 vector unsigned char vec_sld (vector unsigned char,
8371 vector unsigned char,
8373 vector bool char vec_sld (vector bool char,
8377 vector signed int vec_sll (vector signed int,
8378 vector unsigned int);
8379 vector signed int vec_sll (vector signed int,
8380 vector unsigned short);
8381 vector signed int vec_sll (vector signed int,
8382 vector unsigned char);
8383 vector unsigned int vec_sll (vector unsigned int,
8384 vector unsigned int);
8385 vector unsigned int vec_sll (vector unsigned int,
8386 vector unsigned short);
8387 vector unsigned int vec_sll (vector unsigned int,
8388 vector unsigned char);
8389 vector bool int vec_sll (vector bool int,
8390 vector unsigned int);
8391 vector bool int vec_sll (vector bool int,
8392 vector unsigned short);
8393 vector bool int vec_sll (vector bool int,
8394 vector unsigned char);
8395 vector signed short vec_sll (vector signed short,
8396 vector unsigned int);
8397 vector signed short vec_sll (vector signed short,
8398 vector unsigned short);
8399 vector signed short vec_sll (vector signed short,
8400 vector unsigned char);
8401 vector unsigned short vec_sll (vector unsigned short,
8402 vector unsigned int);
8403 vector unsigned short vec_sll (vector unsigned short,
8404 vector unsigned short);
8405 vector unsigned short vec_sll (vector unsigned short,
8406 vector unsigned char);
8407 vector bool short vec_sll (vector bool short, vector unsigned int);
8408 vector bool short vec_sll (vector bool short, vector unsigned short);
8409 vector bool short vec_sll (vector bool short, vector unsigned char);
8410 vector pixel vec_sll (vector pixel, vector unsigned int);
8411 vector pixel vec_sll (vector pixel, vector unsigned short);
8412 vector pixel vec_sll (vector pixel, vector unsigned char);
8413 vector signed char vec_sll (vector signed char, vector unsigned int);
8414 vector signed char vec_sll (vector signed char, vector unsigned short);
8415 vector signed char vec_sll (vector signed char, vector unsigned char);
8416 vector unsigned char vec_sll (vector unsigned char,
8417 vector unsigned int);
8418 vector unsigned char vec_sll (vector unsigned char,
8419 vector unsigned short);
8420 vector unsigned char vec_sll (vector unsigned char,
8421 vector unsigned char);
8422 vector bool char vec_sll (vector bool char, vector unsigned int);
8423 vector bool char vec_sll (vector bool char, vector unsigned short);
8424 vector bool char vec_sll (vector bool char, vector unsigned char);
8426 vector float vec_slo (vector float, vector signed char);
8427 vector float vec_slo (vector float, vector unsigned char);
8428 vector signed int vec_slo (vector signed int, vector signed char);
8429 vector signed int vec_slo (vector signed int, vector unsigned char);
8430 vector unsigned int vec_slo (vector unsigned int, vector signed char);
8431 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
8432 vector signed short vec_slo (vector signed short, vector signed char);
8433 vector signed short vec_slo (vector signed short, vector unsigned char);
8434 vector unsigned short vec_slo (vector unsigned short,
8435 vector signed char);
8436 vector unsigned short vec_slo (vector unsigned short,
8437 vector unsigned char);
8438 vector pixel vec_slo (vector pixel, vector signed char);
8439 vector pixel vec_slo (vector pixel, vector unsigned char);
8440 vector signed char vec_slo (vector signed char, vector signed char);
8441 vector signed char vec_slo (vector signed char, vector unsigned char);
8442 vector unsigned char vec_slo (vector unsigned char, vector signed char);
8443 vector unsigned char vec_slo (vector unsigned char,
8444 vector unsigned char);
8446 vector signed char vec_splat (vector signed char, const int);
8447 vector unsigned char vec_splat (vector unsigned char, const int);
8448 vector bool char vec_splat (vector bool char, const int);
8449 vector signed short vec_splat (vector signed short, const int);
8450 vector unsigned short vec_splat (vector unsigned short, const int);
8451 vector bool short vec_splat (vector bool short, const int);
8452 vector pixel vec_splat (vector pixel, const int);
8453 vector float vec_splat (vector float, const int);
8454 vector signed int vec_splat (vector signed int, const int);
8455 vector unsigned int vec_splat (vector unsigned int, const int);
8456 vector bool int vec_splat (vector bool int, const int);
8458 vector float vec_vspltw (vector float, const int);
8459 vector signed int vec_vspltw (vector signed int, const int);
8460 vector unsigned int vec_vspltw (vector unsigned int, const int);
8461 vector bool int vec_vspltw (vector bool int, const int);
8463 vector bool short vec_vsplth (vector bool short, const int);
8464 vector signed short vec_vsplth (vector signed short, const int);
8465 vector unsigned short vec_vsplth (vector unsigned short, const int);
8466 vector pixel vec_vsplth (vector pixel, const int);
8468 vector signed char vec_vspltb (vector signed char, const int);
8469 vector unsigned char vec_vspltb (vector unsigned char, const int);
8470 vector bool char vec_vspltb (vector bool char, const int);
8472 vector signed char vec_splat_s8 (const int);
8474 vector signed short vec_splat_s16 (const int);
8476 vector signed int vec_splat_s32 (const int);
8478 vector unsigned char vec_splat_u8 (const int);
8480 vector unsigned short vec_splat_u16 (const int);
8482 vector unsigned int vec_splat_u32 (const int);
8484 vector signed char vec_sr (vector signed char, vector unsigned char);
8485 vector unsigned char vec_sr (vector unsigned char,
8486 vector unsigned char);
8487 vector signed short vec_sr (vector signed short,
8488 vector unsigned short);
8489 vector unsigned short vec_sr (vector unsigned short,
8490 vector unsigned short);
8491 vector signed int vec_sr (vector signed int, vector unsigned int);
8492 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
8494 vector signed int vec_vsrw (vector signed int, vector unsigned int);
8495 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
8497 vector signed short vec_vsrh (vector signed short,
8498 vector unsigned short);
8499 vector unsigned short vec_vsrh (vector unsigned short,
8500 vector unsigned short);
8502 vector signed char vec_vsrb (vector signed char, vector unsigned char);
8503 vector unsigned char vec_vsrb (vector unsigned char,
8504 vector unsigned char);
8506 vector signed char vec_sra (vector signed char, vector unsigned char);
8507 vector unsigned char vec_sra (vector unsigned char,
8508 vector unsigned char);
8509 vector signed short vec_sra (vector signed short,
8510 vector unsigned short);
8511 vector unsigned short vec_sra (vector unsigned short,
8512 vector unsigned short);
8513 vector signed int vec_sra (vector signed int, vector unsigned int);
8514 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
8516 vector signed int vec_vsraw (vector signed int, vector unsigned int);
8517 vector unsigned int vec_vsraw (vector unsigned int,
8518 vector unsigned int);
8520 vector signed short vec_vsrah (vector signed short,
8521 vector unsigned short);
8522 vector unsigned short vec_vsrah (vector unsigned short,
8523 vector unsigned short);
8525 vector signed char vec_vsrab (vector signed char, vector unsigned char);
8526 vector unsigned char vec_vsrab (vector unsigned char,
8527 vector unsigned char);
8529 vector signed int vec_srl (vector signed int, vector unsigned int);
8530 vector signed int vec_srl (vector signed int, vector unsigned short);
8531 vector signed int vec_srl (vector signed int, vector unsigned char);
8532 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
8533 vector unsigned int vec_srl (vector unsigned int,
8534 vector unsigned short);
8535 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
8536 vector bool int vec_srl (vector bool int, vector unsigned int);
8537 vector bool int vec_srl (vector bool int, vector unsigned short);
8538 vector bool int vec_srl (vector bool int, vector unsigned char);
8539 vector signed short vec_srl (vector signed short, vector unsigned int);
8540 vector signed short vec_srl (vector signed short,
8541 vector unsigned short);
8542 vector signed short vec_srl (vector signed short, vector unsigned char);
8543 vector unsigned short vec_srl (vector unsigned short,
8544 vector unsigned int);
8545 vector unsigned short vec_srl (vector unsigned short,
8546 vector unsigned short);
8547 vector unsigned short vec_srl (vector unsigned short,
8548 vector unsigned char);
8549 vector bool short vec_srl (vector bool short, vector unsigned int);
8550 vector bool short vec_srl (vector bool short, vector unsigned short);
8551 vector bool short vec_srl (vector bool short, vector unsigned char);
8552 vector pixel vec_srl (vector pixel, vector unsigned int);
8553 vector pixel vec_srl (vector pixel, vector unsigned short);
8554 vector pixel vec_srl (vector pixel, vector unsigned char);
8555 vector signed char vec_srl (vector signed char, vector unsigned int);
8556 vector signed char vec_srl (vector signed char, vector unsigned short);
8557 vector signed char vec_srl (vector signed char, vector unsigned char);
8558 vector unsigned char vec_srl (vector unsigned char,
8559 vector unsigned int);
8560 vector unsigned char vec_srl (vector unsigned char,
8561 vector unsigned short);
8562 vector unsigned char vec_srl (vector unsigned char,
8563 vector unsigned char);
8564 vector bool char vec_srl (vector bool char, vector unsigned int);
8565 vector bool char vec_srl (vector bool char, vector unsigned short);
8566 vector bool char vec_srl (vector bool char, vector unsigned char);
8568 vector float vec_sro (vector float, vector signed char);
8569 vector float vec_sro (vector float, vector unsigned char);
8570 vector signed int vec_sro (vector signed int, vector signed char);
8571 vector signed int vec_sro (vector signed int, vector unsigned char);
8572 vector unsigned int vec_sro (vector unsigned int, vector signed char);
8573 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
8574 vector signed short vec_sro (vector signed short, vector signed char);
8575 vector signed short vec_sro (vector signed short, vector unsigned char);
8576 vector unsigned short vec_sro (vector unsigned short,
8577 vector signed char);
8578 vector unsigned short vec_sro (vector unsigned short,
8579 vector unsigned char);
8580 vector pixel vec_sro (vector pixel, vector signed char);
8581 vector pixel vec_sro (vector pixel, vector unsigned char);
8582 vector signed char vec_sro (vector signed char, vector signed char);
8583 vector signed char vec_sro (vector signed char, vector unsigned char);
8584 vector unsigned char vec_sro (vector unsigned char, vector signed char);
8585 vector unsigned char vec_sro (vector unsigned char,
8586 vector unsigned char);
8588 void vec_st (vector float, int, vector float *);
8589 void vec_st (vector float, int, float *);
8590 void vec_st (vector signed int, int, vector signed int *);
8591 void vec_st (vector signed int, int, int *);
8592 void vec_st (vector unsigned int, int, vector unsigned int *);
8593 void vec_st (vector unsigned int, int, unsigned int *);
8594 void vec_st (vector bool int, int, vector bool int *);
8595 void vec_st (vector bool int, int, unsigned int *);
8596 void vec_st (vector bool int, int, int *);
8597 void vec_st (vector signed short, int, vector signed short *);
8598 void vec_st (vector signed short, int, short *);
8599 void vec_st (vector unsigned short, int, vector unsigned short *);
8600 void vec_st (vector unsigned short, int, unsigned short *);
8601 void vec_st (vector bool short, int, vector bool short *);
8602 void vec_st (vector bool short, int, unsigned short *);
8603 void vec_st (vector pixel, int, vector pixel *);
8604 void vec_st (vector pixel, int, unsigned short *);
8605 void vec_st (vector pixel, int, short *);
8606 void vec_st (vector bool short, int, short *);
8607 void vec_st (vector signed char, int, vector signed char *);
8608 void vec_st (vector signed char, int, signed char *);
8609 void vec_st (vector unsigned char, int, vector unsigned char *);
8610 void vec_st (vector unsigned char, int, unsigned char *);
8611 void vec_st (vector bool char, int, vector bool char *);
8612 void vec_st (vector bool char, int, unsigned char *);
8613 void vec_st (vector bool char, int, signed char *);
8615 void vec_ste (vector signed char, int, signed char *);
8616 void vec_ste (vector unsigned char, int, unsigned char *);
8617 void vec_ste (vector bool char, int, signed char *);
8618 void vec_ste (vector bool char, int, unsigned char *);
8619 void vec_ste (vector signed short, int, short *);
8620 void vec_ste (vector unsigned short, int, unsigned short *);
8621 void vec_ste (vector bool short, int, short *);
8622 void vec_ste (vector bool short, int, unsigned short *);
8623 void vec_ste (vector pixel, int, short *);
8624 void vec_ste (vector pixel, int, unsigned short *);
8625 void vec_ste (vector float, int, float *);
8626 void vec_ste (vector signed int, int, int *);
8627 void vec_ste (vector unsigned int, int, unsigned int *);
8628 void vec_ste (vector bool int, int, int *);
8629 void vec_ste (vector bool int, int, unsigned int *);
8631 void vec_stvewx (vector float, int, float *);
8632 void vec_stvewx (vector signed int, int, int *);
8633 void vec_stvewx (vector unsigned int, int, unsigned int *);
8634 void vec_stvewx (vector bool int, int, int *);
8635 void vec_stvewx (vector bool int, int, unsigned int *);
8637 void vec_stvehx (vector signed short, int, short *);
8638 void vec_stvehx (vector unsigned short, int, unsigned short *);
8639 void vec_stvehx (vector bool short, int, short *);
8640 void vec_stvehx (vector bool short, int, unsigned short *);
8641 void vec_stvehx (vector pixel, int, short *);
8642 void vec_stvehx (vector pixel, int, unsigned short *);
8644 void vec_stvebx (vector signed char, int, signed char *);
8645 void vec_stvebx (vector unsigned char, int, unsigned char *);
8646 void vec_stvebx (vector bool char, int, signed char *);
8647 void vec_stvebx (vector bool char, int, unsigned char *);
8649 void vec_stl (vector float, int, vector float *);
8650 void vec_stl (vector float, int, float *);
8651 void vec_stl (vector signed int, int, vector signed int *);
8652 void vec_stl (vector signed int, int, int *);
8653 void vec_stl (vector unsigned int, int, vector unsigned int *);
8654 void vec_stl (vector unsigned int, int, unsigned int *);
8655 void vec_stl (vector bool int, int, vector bool int *);
8656 void vec_stl (vector bool int, int, unsigned int *);
8657 void vec_stl (vector bool int, int, int *);
8658 void vec_stl (vector signed short, int, vector signed short *);
8659 void vec_stl (vector signed short, int, short *);
8660 void vec_stl (vector unsigned short, int, vector unsigned short *);
8661 void vec_stl (vector unsigned short, int, unsigned short *);
8662 void vec_stl (vector bool short, int, vector bool short *);
8663 void vec_stl (vector bool short, int, unsigned short *);
8664 void vec_stl (vector bool short, int, short *);
8665 void vec_stl (vector pixel, int, vector pixel *);
8666 void vec_stl (vector pixel, int, unsigned short *);
8667 void vec_stl (vector pixel, int, short *);
8668 void vec_stl (vector signed char, int, vector signed char *);
8669 void vec_stl (vector signed char, int, signed char *);
8670 void vec_stl (vector unsigned char, int, vector unsigned char *);
8671 void vec_stl (vector unsigned char, int, unsigned char *);
8672 void vec_stl (vector bool char, int, vector bool char *);
8673 void vec_stl (vector bool char, int, unsigned char *);
8674 void vec_stl (vector bool char, int, signed char *);
8676 vector signed char vec_sub (vector bool char, vector signed char);
8677 vector signed char vec_sub (vector signed char, vector bool char);
8678 vector signed char vec_sub (vector signed char, vector signed char);
8679 vector unsigned char vec_sub (vector bool char, vector unsigned char);
8680 vector unsigned char vec_sub (vector unsigned char, vector bool char);
8681 vector unsigned char vec_sub (vector unsigned char,
8682 vector unsigned char);
8683 vector signed short vec_sub (vector bool short, vector signed short);
8684 vector signed short vec_sub (vector signed short, vector bool short);
8685 vector signed short vec_sub (vector signed short, vector signed short);
8686 vector unsigned short vec_sub (vector bool short,
8687 vector unsigned short);
8688 vector unsigned short vec_sub (vector unsigned short,
8690 vector unsigned short vec_sub (vector unsigned short,
8691 vector unsigned short);
8692 vector signed int vec_sub (vector bool int, vector signed int);
8693 vector signed int vec_sub (vector signed int, vector bool int);
8694 vector signed int vec_sub (vector signed int, vector signed int);
8695 vector unsigned int vec_sub (vector bool int, vector unsigned int);
8696 vector unsigned int vec_sub (vector unsigned int, vector bool int);
8697 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
8698 vector float vec_sub (vector float, vector float);
8700 vector float vec_vsubfp (vector float, vector float);
8702 vector signed int vec_vsubuwm (vector bool int, vector signed int);
8703 vector signed int vec_vsubuwm (vector signed int, vector bool int);
8704 vector signed int vec_vsubuwm (vector signed int, vector signed int);
8705 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
8706 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
8707 vector unsigned int vec_vsubuwm (vector unsigned int,
8708 vector unsigned int);
8710 vector signed short vec_vsubuhm (vector bool short,
8711 vector signed short);
8712 vector signed short vec_vsubuhm (vector signed short,
8714 vector signed short vec_vsubuhm (vector signed short,
8715 vector signed short);
8716 vector unsigned short vec_vsubuhm (vector bool short,
8717 vector unsigned short);
8718 vector unsigned short vec_vsubuhm (vector unsigned short,
8720 vector unsigned short vec_vsubuhm (vector unsigned short,
8721 vector unsigned short);
8723 vector signed char vec_vsububm (vector bool char, vector signed char);
8724 vector signed char vec_vsububm (vector signed char, vector bool char);
8725 vector signed char vec_vsububm (vector signed char, vector signed char);
8726 vector unsigned char vec_vsububm (vector bool char,
8727 vector unsigned char);
8728 vector unsigned char vec_vsububm (vector unsigned char,
8730 vector unsigned char vec_vsububm (vector unsigned char,
8731 vector unsigned char);
8733 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
8735 vector unsigned char vec_subs (vector bool char, vector unsigned char);
8736 vector unsigned char vec_subs (vector unsigned char, vector bool char);
8737 vector unsigned char vec_subs (vector unsigned char,
8738 vector unsigned char);
8739 vector signed char vec_subs (vector bool char, vector signed char);
8740 vector signed char vec_subs (vector signed char, vector bool char);
8741 vector signed char vec_subs (vector signed char, vector signed char);
8742 vector unsigned short vec_subs (vector bool short,
8743 vector unsigned short);
8744 vector unsigned short vec_subs (vector unsigned short,
8746 vector unsigned short vec_subs (vector unsigned short,
8747 vector unsigned short);
8748 vector signed short vec_subs (vector bool short, vector signed short);
8749 vector signed short vec_subs (vector signed short, vector bool short);
8750 vector signed short vec_subs (vector signed short, vector signed short);
8751 vector unsigned int vec_subs (vector bool int, vector unsigned int);
8752 vector unsigned int vec_subs (vector unsigned int, vector bool int);
8753 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
8754 vector signed int vec_subs (vector bool int, vector signed int);
8755 vector signed int vec_subs (vector signed int, vector bool int);
8756 vector signed int vec_subs (vector signed int, vector signed int);
8758 vector signed int vec_vsubsws (vector bool int, vector signed int);
8759 vector signed int vec_vsubsws (vector signed int, vector bool int);
8760 vector signed int vec_vsubsws (vector signed int, vector signed int);
8762 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
8763 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
8764 vector unsigned int vec_vsubuws (vector unsigned int,
8765 vector unsigned int);
8767 vector signed short vec_vsubshs (vector bool short,
8768 vector signed short);
8769 vector signed short vec_vsubshs (vector signed short,
8771 vector signed short vec_vsubshs (vector signed short,
8772 vector signed short);
8774 vector unsigned short vec_vsubuhs (vector bool short,
8775 vector unsigned short);
8776 vector unsigned short vec_vsubuhs (vector unsigned short,
8778 vector unsigned short vec_vsubuhs (vector unsigned short,
8779 vector unsigned short);
8781 vector signed char vec_vsubsbs (vector bool char, vector signed char);
8782 vector signed char vec_vsubsbs (vector signed char, vector bool char);
8783 vector signed char vec_vsubsbs (vector signed char, vector signed char);
8785 vector unsigned char vec_vsububs (vector bool char,
8786 vector unsigned char);
8787 vector unsigned char vec_vsububs (vector unsigned char,
8789 vector unsigned char vec_vsububs (vector unsigned char,
8790 vector unsigned char);
8792 vector unsigned int vec_sum4s (vector unsigned char,
8793 vector unsigned int);
8794 vector signed int vec_sum4s (vector signed char, vector signed int);
8795 vector signed int vec_sum4s (vector signed short, vector signed int);
8797 vector signed int vec_vsum4shs (vector signed short, vector signed int);
8799 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
8801 vector unsigned int vec_vsum4ubs (vector unsigned char,
8802 vector unsigned int);
8804 vector signed int vec_sum2s (vector signed int, vector signed int);
8806 vector signed int vec_sums (vector signed int, vector signed int);
8808 vector float vec_trunc (vector float);
8810 vector signed short vec_unpackh (vector signed char);
8811 vector bool short vec_unpackh (vector bool char);
8812 vector signed int vec_unpackh (vector signed short);
8813 vector bool int vec_unpackh (vector bool short);
8814 vector unsigned int vec_unpackh (vector pixel);
8816 vector bool int vec_vupkhsh (vector bool short);
8817 vector signed int vec_vupkhsh (vector signed short);
8819 vector unsigned int vec_vupkhpx (vector pixel);
8821 vector bool short vec_vupkhsb (vector bool char);
8822 vector signed short vec_vupkhsb (vector signed char);
8824 vector signed short vec_unpackl (vector signed char);
8825 vector bool short vec_unpackl (vector bool char);
8826 vector unsigned int vec_unpackl (vector pixel);
8827 vector signed int vec_unpackl (vector signed short);
8828 vector bool int vec_unpackl (vector bool short);
8830 vector unsigned int vec_vupklpx (vector pixel);
8832 vector bool int vec_vupklsh (vector bool short);
8833 vector signed int vec_vupklsh (vector signed short);
8835 vector bool short vec_vupklsb (vector bool char);
8836 vector signed short vec_vupklsb (vector signed char);
8838 vector float vec_xor (vector float, vector float);
8839 vector float vec_xor (vector float, vector bool int);
8840 vector float vec_xor (vector bool int, vector float);
8841 vector bool int vec_xor (vector bool int, vector bool int);
8842 vector signed int vec_xor (vector bool int, vector signed int);
8843 vector signed int vec_xor (vector signed int, vector bool int);
8844 vector signed int vec_xor (vector signed int, vector signed int);
8845 vector unsigned int vec_xor (vector bool int, vector unsigned int);
8846 vector unsigned int vec_xor (vector unsigned int, vector bool int);
8847 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
8848 vector bool short vec_xor (vector bool short, vector bool short);
8849 vector signed short vec_xor (vector bool short, vector signed short);
8850 vector signed short vec_xor (vector signed short, vector bool short);
8851 vector signed short vec_xor (vector signed short, vector signed short);
8852 vector unsigned short vec_xor (vector bool short,
8853 vector unsigned short);
8854 vector unsigned short vec_xor (vector unsigned short,
8856 vector unsigned short vec_xor (vector unsigned short,
8857 vector unsigned short);
8858 vector signed char vec_xor (vector bool char, vector signed char);
8859 vector bool char vec_xor (vector bool char, vector bool char);
8860 vector signed char vec_xor (vector signed char, vector bool char);
8861 vector signed char vec_xor (vector signed char, vector signed char);
8862 vector unsigned char vec_xor (vector bool char, vector unsigned char);
8863 vector unsigned char vec_xor (vector unsigned char, vector bool char);
8864 vector unsigned char vec_xor (vector unsigned char,
8865 vector unsigned char);
8867 int vec_all_eq (vector signed char, vector bool char);
8868 int vec_all_eq (vector signed char, vector signed char);
8869 int vec_all_eq (vector unsigned char, vector bool char);
8870 int vec_all_eq (vector unsigned char, vector unsigned char);
8871 int vec_all_eq (vector bool char, vector bool char);
8872 int vec_all_eq (vector bool char, vector unsigned char);
8873 int vec_all_eq (vector bool char, vector signed char);
8874 int vec_all_eq (vector signed short, vector bool short);
8875 int vec_all_eq (vector signed short, vector signed short);
8876 int vec_all_eq (vector unsigned short, vector bool short);
8877 int vec_all_eq (vector unsigned short, vector unsigned short);
8878 int vec_all_eq (vector bool short, vector bool short);
8879 int vec_all_eq (vector bool short, vector unsigned short);
8880 int vec_all_eq (vector bool short, vector signed short);
8881 int vec_all_eq (vector pixel, vector pixel);
8882 int vec_all_eq (vector signed int, vector bool int);
8883 int vec_all_eq (vector signed int, vector signed int);
8884 int vec_all_eq (vector unsigned int, vector bool int);
8885 int vec_all_eq (vector unsigned int, vector unsigned int);
8886 int vec_all_eq (vector bool int, vector bool int);
8887 int vec_all_eq (vector bool int, vector unsigned int);
8888 int vec_all_eq (vector bool int, vector signed int);
8889 int vec_all_eq (vector float, vector float);
8891 int vec_all_ge (vector bool char, vector unsigned char);
8892 int vec_all_ge (vector unsigned char, vector bool char);
8893 int vec_all_ge (vector unsigned char, vector unsigned char);
8894 int vec_all_ge (vector bool char, vector signed char);
8895 int vec_all_ge (vector signed char, vector bool char);
8896 int vec_all_ge (vector signed char, vector signed char);
8897 int vec_all_ge (vector bool short, vector unsigned short);
8898 int vec_all_ge (vector unsigned short, vector bool short);
8899 int vec_all_ge (vector unsigned short, vector unsigned short);
8900 int vec_all_ge (vector signed short, vector signed short);
8901 int vec_all_ge (vector bool short, vector signed short);
8902 int vec_all_ge (vector signed short, vector bool short);
8903 int vec_all_ge (vector bool int, vector unsigned int);
8904 int vec_all_ge (vector unsigned int, vector bool int);
8905 int vec_all_ge (vector unsigned int, vector unsigned int);
8906 int vec_all_ge (vector bool int, vector signed int);
8907 int vec_all_ge (vector signed int, vector bool int);
8908 int vec_all_ge (vector signed int, vector signed int);
8909 int vec_all_ge (vector float, vector float);
8911 int vec_all_gt (vector bool char, vector unsigned char);
8912 int vec_all_gt (vector unsigned char, vector bool char);
8913 int vec_all_gt (vector unsigned char, vector unsigned char);
8914 int vec_all_gt (vector bool char, vector signed char);
8915 int vec_all_gt (vector signed char, vector bool char);
8916 int vec_all_gt (vector signed char, vector signed char);
8917 int vec_all_gt (vector bool short, vector unsigned short);
8918 int vec_all_gt (vector unsigned short, vector bool short);
8919 int vec_all_gt (vector unsigned short, vector unsigned short);
8920 int vec_all_gt (vector bool short, vector signed short);
8921 int vec_all_gt (vector signed short, vector bool short);
8922 int vec_all_gt (vector signed short, vector signed short);
8923 int vec_all_gt (vector bool int, vector unsigned int);
8924 int vec_all_gt (vector unsigned int, vector bool int);
8925 int vec_all_gt (vector unsigned int, vector unsigned int);
8926 int vec_all_gt (vector bool int, vector signed int);
8927 int vec_all_gt (vector signed int, vector bool int);
8928 int vec_all_gt (vector signed int, vector signed int);
8929 int vec_all_gt (vector float, vector float);
8931 int vec_all_in (vector float, vector float);
8933 int vec_all_le (vector bool char, vector unsigned char);
8934 int vec_all_le (vector unsigned char, vector bool char);
8935 int vec_all_le (vector unsigned char, vector unsigned char);
8936 int vec_all_le (vector bool char, vector signed char);
8937 int vec_all_le (vector signed char, vector bool char);
8938 int vec_all_le (vector signed char, vector signed char);
8939 int vec_all_le (vector bool short, vector unsigned short);
8940 int vec_all_le (vector unsigned short, vector bool short);
8941 int vec_all_le (vector unsigned short, vector unsigned short);
8942 int vec_all_le (vector bool short, vector signed short);
8943 int vec_all_le (vector signed short, vector bool short);
8944 int vec_all_le (vector signed short, vector signed short);
8945 int vec_all_le (vector bool int, vector unsigned int);
8946 int vec_all_le (vector unsigned int, vector bool int);
8947 int vec_all_le (vector unsigned int, vector unsigned int);
8948 int vec_all_le (vector bool int, vector signed int);
8949 int vec_all_le (vector signed int, vector bool int);
8950 int vec_all_le (vector signed int, vector signed int);
8951 int vec_all_le (vector float, vector float);
8953 int vec_all_lt (vector bool char, vector unsigned char);
8954 int vec_all_lt (vector unsigned char, vector bool char);
8955 int vec_all_lt (vector unsigned char, vector unsigned char);
8956 int vec_all_lt (vector bool char, vector signed char);
8957 int vec_all_lt (vector signed char, vector bool char);
8958 int vec_all_lt (vector signed char, vector signed char);
8959 int vec_all_lt (vector bool short, vector unsigned short);
8960 int vec_all_lt (vector unsigned short, vector bool short);
8961 int vec_all_lt (vector unsigned short, vector unsigned short);
8962 int vec_all_lt (vector bool short, vector signed short);
8963 int vec_all_lt (vector signed short, vector bool short);
8964 int vec_all_lt (vector signed short, vector signed short);
8965 int vec_all_lt (vector bool int, vector unsigned int);
8966 int vec_all_lt (vector unsigned int, vector bool int);
8967 int vec_all_lt (vector unsigned int, vector unsigned int);
8968 int vec_all_lt (vector bool int, vector signed int);
8969 int vec_all_lt (vector signed int, vector bool int);
8970 int vec_all_lt (vector signed int, vector signed int);
8971 int vec_all_lt (vector float, vector float);
8973 int vec_all_nan (vector float);
8975 int vec_all_ne (vector signed char, vector bool char);
8976 int vec_all_ne (vector signed char, vector signed char);
8977 int vec_all_ne (vector unsigned char, vector bool char);
8978 int vec_all_ne (vector unsigned char, vector unsigned char);
8979 int vec_all_ne (vector bool char, vector bool char);
8980 int vec_all_ne (vector bool char, vector unsigned char);
8981 int vec_all_ne (vector bool char, vector signed char);
8982 int vec_all_ne (vector signed short, vector bool short);
8983 int vec_all_ne (vector signed short, vector signed short);
8984 int vec_all_ne (vector unsigned short, vector bool short);
8985 int vec_all_ne (vector unsigned short, vector unsigned short);
8986 int vec_all_ne (vector bool short, vector bool short);
8987 int vec_all_ne (vector bool short, vector unsigned short);
8988 int vec_all_ne (vector bool short, vector signed short);
8989 int vec_all_ne (vector pixel, vector pixel);
8990 int vec_all_ne (vector signed int, vector bool int);
8991 int vec_all_ne (vector signed int, vector signed int);
8992 int vec_all_ne (vector unsigned int, vector bool int);
8993 int vec_all_ne (vector unsigned int, vector unsigned int);
8994 int vec_all_ne (vector bool int, vector bool int);
8995 int vec_all_ne (vector bool int, vector unsigned int);
8996 int vec_all_ne (vector bool int, vector signed int);
8997 int vec_all_ne (vector float, vector float);
8999 int vec_all_nge (vector float, vector float);
9001 int vec_all_ngt (vector float, vector float);
9003 int vec_all_nle (vector float, vector float);
9005 int vec_all_nlt (vector float, vector float);
9007 int vec_all_numeric (vector float);
9009 int vec_any_eq (vector signed char, vector bool char);
9010 int vec_any_eq (vector signed char, vector signed char);
9011 int vec_any_eq (vector unsigned char, vector bool char);
9012 int vec_any_eq (vector unsigned char, vector unsigned char);
9013 int vec_any_eq (vector bool char, vector bool char);
9014 int vec_any_eq (vector bool char, vector unsigned char);
9015 int vec_any_eq (vector bool char, vector signed char);
9016 int vec_any_eq (vector signed short, vector bool short);
9017 int vec_any_eq (vector signed short, vector signed short);
9018 int vec_any_eq (vector unsigned short, vector bool short);
9019 int vec_any_eq (vector unsigned short, vector unsigned short);
9020 int vec_any_eq (vector bool short, vector bool short);
9021 int vec_any_eq (vector bool short, vector unsigned short);
9022 int vec_any_eq (vector bool short, vector signed short);
9023 int vec_any_eq (vector pixel, vector pixel);
9024 int vec_any_eq (vector signed int, vector bool int);
9025 int vec_any_eq (vector signed int, vector signed int);
9026 int vec_any_eq (vector unsigned int, vector bool int);
9027 int vec_any_eq (vector unsigned int, vector unsigned int);
9028 int vec_any_eq (vector bool int, vector bool int);
9029 int vec_any_eq (vector bool int, vector unsigned int);
9030 int vec_any_eq (vector bool int, vector signed int);
9031 int vec_any_eq (vector float, vector float);
9033 int vec_any_ge (vector signed char, vector bool char);
9034 int vec_any_ge (vector unsigned char, vector bool char);
9035 int vec_any_ge (vector unsigned char, vector unsigned char);
9036 int vec_any_ge (vector signed char, vector signed char);
9037 int vec_any_ge (vector bool char, vector unsigned char);
9038 int vec_any_ge (vector bool char, vector signed char);
9039 int vec_any_ge (vector unsigned short, vector bool short);
9040 int vec_any_ge (vector unsigned short, vector unsigned short);
9041 int vec_any_ge (vector signed short, vector signed short);
9042 int vec_any_ge (vector signed short, vector bool short);
9043 int vec_any_ge (vector bool short, vector unsigned short);
9044 int vec_any_ge (vector bool short, vector signed short);
9045 int vec_any_ge (vector signed int, vector bool int);
9046 int vec_any_ge (vector unsigned int, vector bool int);
9047 int vec_any_ge (vector unsigned int, vector unsigned int);
9048 int vec_any_ge (vector signed int, vector signed int);
9049 int vec_any_ge (vector bool int, vector unsigned int);
9050 int vec_any_ge (vector bool int, vector signed int);
9051 int vec_any_ge (vector float, vector float);
9053 int vec_any_gt (vector bool char, vector unsigned char);
9054 int vec_any_gt (vector unsigned char, vector bool char);
9055 int vec_any_gt (vector unsigned char, vector unsigned char);
9056 int vec_any_gt (vector bool char, vector signed char);
9057 int vec_any_gt (vector signed char, vector bool char);
9058 int vec_any_gt (vector signed char, vector signed char);
9059 int vec_any_gt (vector bool short, vector unsigned short);
9060 int vec_any_gt (vector unsigned short, vector bool short);
9061 int vec_any_gt (vector unsigned short, vector unsigned short);
9062 int vec_any_gt (vector bool short, vector signed short);
9063 int vec_any_gt (vector signed short, vector bool short);
9064 int vec_any_gt (vector signed short, vector signed short);
9065 int vec_any_gt (vector bool int, vector unsigned int);
9066 int vec_any_gt (vector unsigned int, vector bool int);
9067 int vec_any_gt (vector unsigned int, vector unsigned int);
9068 int vec_any_gt (vector bool int, vector signed int);
9069 int vec_any_gt (vector signed int, vector bool int);
9070 int vec_any_gt (vector signed int, vector signed int);
9071 int vec_any_gt (vector float, vector float);
9073 int vec_any_le (vector bool char, vector unsigned char);
9074 int vec_any_le (vector unsigned char, vector bool char);
9075 int vec_any_le (vector unsigned char, vector unsigned char);
9076 int vec_any_le (vector bool char, vector signed char);
9077 int vec_any_le (vector signed char, vector bool char);
9078 int vec_any_le (vector signed char, vector signed char);
9079 int vec_any_le (vector bool short, vector unsigned short);
9080 int vec_any_le (vector unsigned short, vector bool short);
9081 int vec_any_le (vector unsigned short, vector unsigned short);
9082 int vec_any_le (vector bool short, vector signed short);
9083 int vec_any_le (vector signed short, vector bool short);
9084 int vec_any_le (vector signed short, vector signed short);
9085 int vec_any_le (vector bool int, vector unsigned int);
9086 int vec_any_le (vector unsigned int, vector bool int);
9087 int vec_any_le (vector unsigned int, vector unsigned int);
9088 int vec_any_le (vector bool int, vector signed int);
9089 int vec_any_le (vector signed int, vector bool int);
9090 int vec_any_le (vector signed int, vector signed int);
9091 int vec_any_le (vector float, vector float);
9093 int vec_any_lt (vector bool char, vector unsigned char);
9094 int vec_any_lt (vector unsigned char, vector bool char);
9095 int vec_any_lt (vector unsigned char, vector unsigned char);
9096 int vec_any_lt (vector bool char, vector signed char);
9097 int vec_any_lt (vector signed char, vector bool char);
9098 int vec_any_lt (vector signed char, vector signed char);
9099 int vec_any_lt (vector bool short, vector unsigned short);
9100 int vec_any_lt (vector unsigned short, vector bool short);
9101 int vec_any_lt (vector unsigned short, vector unsigned short);
9102 int vec_any_lt (vector bool short, vector signed short);
9103 int vec_any_lt (vector signed short, vector bool short);
9104 int vec_any_lt (vector signed short, vector signed short);
9105 int vec_any_lt (vector bool int, vector unsigned int);
9106 int vec_any_lt (vector unsigned int, vector bool int);
9107 int vec_any_lt (vector unsigned int, vector unsigned int);
9108 int vec_any_lt (vector bool int, vector signed int);
9109 int vec_any_lt (vector signed int, vector bool int);
9110 int vec_any_lt (vector signed int, vector signed int);
9111 int vec_any_lt (vector float, vector float);
9113 int vec_any_nan (vector float);
9115 int vec_any_ne (vector signed char, vector bool char);
9116 int vec_any_ne (vector signed char, vector signed char);
9117 int vec_any_ne (vector unsigned char, vector bool char);
9118 int vec_any_ne (vector unsigned char, vector unsigned char);
9119 int vec_any_ne (vector bool char, vector bool char);
9120 int vec_any_ne (vector bool char, vector unsigned char);
9121 int vec_any_ne (vector bool char, vector signed char);
9122 int vec_any_ne (vector signed short, vector bool short);
9123 int vec_any_ne (vector signed short, vector signed short);
9124 int vec_any_ne (vector unsigned short, vector bool short);
9125 int vec_any_ne (vector unsigned short, vector unsigned short);
9126 int vec_any_ne (vector bool short, vector bool short);
9127 int vec_any_ne (vector bool short, vector unsigned short);
9128 int vec_any_ne (vector bool short, vector signed short);
9129 int vec_any_ne (vector pixel, vector pixel);
9130 int vec_any_ne (vector signed int, vector bool int);
9131 int vec_any_ne (vector signed int, vector signed int);
9132 int vec_any_ne (vector unsigned int, vector bool int);
9133 int vec_any_ne (vector unsigned int, vector unsigned int);
9134 int vec_any_ne (vector bool int, vector bool int);
9135 int vec_any_ne (vector bool int, vector unsigned int);
9136 int vec_any_ne (vector bool int, vector signed int);
9137 int vec_any_ne (vector float, vector float);
9139 int vec_any_nge (vector float, vector float);
9141 int vec_any_ngt (vector float, vector float);
9143 int vec_any_nle (vector float, vector float);
9145 int vec_any_nlt (vector float, vector float);
9147 int vec_any_numeric (vector float);
9149 int vec_any_out (vector float, vector float);
9152 @node SPARC VIS Built-in Functions
9153 @subsection SPARC VIS Built-in Functions
9155 GCC supports SIMD operations on the SPARC using both the generic vector
9156 extensions (@pxref{Vector Extensions}) as well as built-in functions for
9157 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
9158 switch, the VIS extension is exposed as the following built-in functions:
9161 typedef int v2si __attribute__ ((vector_size (8)));
9162 typedef short v4hi __attribute__ ((vector_size (8)));
9163 typedef short v2hi __attribute__ ((vector_size (4)));
9164 typedef char v8qi __attribute__ ((vector_size (8)));
9165 typedef char v4qi __attribute__ ((vector_size (4)));
9167 void * __builtin_vis_alignaddr (void *, long);
9168 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
9169 v2si __builtin_vis_faligndatav2si (v2si, v2si);
9170 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
9171 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
9173 v4hi __builtin_vis_fexpand (v4qi);
9175 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
9176 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
9177 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
9178 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
9179 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
9180 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
9181 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
9183 v4qi __builtin_vis_fpack16 (v4hi);
9184 v8qi __builtin_vis_fpack32 (v2si, v2si);
9185 v2hi __builtin_vis_fpackfix (v2si);
9186 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
9188 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
9191 @node Target Format Checks
9192 @section Format Checks Specific to Particular Target Machines
9194 For some target machines, GCC supports additional options to the
9196 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
9199 * Solaris Format Checks::
9202 @node Solaris Format Checks
9203 @subsection Solaris Format Checks
9205 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
9206 check. @code{cmn_err} accepts a subset of the standard @code{printf}
9207 conversions, and the two-argument @code{%b} conversion for displaying
9208 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
9211 @section Pragmas Accepted by GCC
9215 GCC supports several types of pragmas, primarily in order to compile
9216 code originally written for other compilers. Note that in general
9217 we do not recommend the use of pragmas; @xref{Function Attributes},
9218 for further explanation.
9223 * RS/6000 and PowerPC Pragmas::
9226 * Symbol-Renaming Pragmas::
9227 * Structure-Packing Pragmas::
9232 @subsection ARM Pragmas
9234 The ARM target defines pragmas for controlling the default addition of
9235 @code{long_call} and @code{short_call} attributes to functions.
9236 @xref{Function Attributes}, for information about the effects of these
9241 @cindex pragma, long_calls
9242 Set all subsequent functions to have the @code{long_call} attribute.
9245 @cindex pragma, no_long_calls
9246 Set all subsequent functions to have the @code{short_call} attribute.
9248 @item long_calls_off
9249 @cindex pragma, long_calls_off
9250 Do not affect the @code{long_call} or @code{short_call} attributes of
9251 subsequent functions.
9255 @subsection M32C Pragmas
9258 @item memregs @var{number}
9259 @cindex pragma, memregs
9260 Overrides the command line option @code{-memregs=} for the current
9261 file. Use with care! This pragma must be before any function in the
9262 file, and mixing different memregs values in different objects may
9263 make them incompatible. This pragma is useful when a
9264 performance-critical function uses a memreg for temporary values,
9265 as it may allow you to reduce the number of memregs used.
9269 @node RS/6000 and PowerPC Pragmas
9270 @subsection RS/6000 and PowerPC Pragmas
9272 The RS/6000 and PowerPC targets define one pragma for controlling
9273 whether or not the @code{longcall} attribute is added to function
9274 declarations by default. This pragma overrides the @option{-mlongcall}
9275 option, but not the @code{longcall} and @code{shortcall} attributes.
9276 @xref{RS/6000 and PowerPC Options}, for more information about when long
9277 calls are and are not necessary.
9281 @cindex pragma, longcall
9282 Apply the @code{longcall} attribute to all subsequent function
9286 Do not apply the @code{longcall} attribute to subsequent function
9290 @c Describe c4x pragmas here.
9291 @c Describe h8300 pragmas here.
9292 @c Describe sh pragmas here.
9293 @c Describe v850 pragmas here.
9295 @node Darwin Pragmas
9296 @subsection Darwin Pragmas
9298 The following pragmas are available for all architectures running the
9299 Darwin operating system. These are useful for compatibility with other
9303 @item mark @var{tokens}@dots{}
9304 @cindex pragma, mark
9305 This pragma is accepted, but has no effect.
9307 @item options align=@var{alignment}
9308 @cindex pragma, options align
9309 This pragma sets the alignment of fields in structures. The values of
9310 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
9311 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
9312 properly; to restore the previous setting, use @code{reset} for the
9315 @item segment @var{tokens}@dots{}
9316 @cindex pragma, segment
9317 This pragma is accepted, but has no effect.
9319 @item unused (@var{var} [, @var{var}]@dots{})
9320 @cindex pragma, unused
9321 This pragma declares variables to be possibly unused. GCC will not
9322 produce warnings for the listed variables. The effect is similar to
9323 that of the @code{unused} attribute, except that this pragma may appear
9324 anywhere within the variables' scopes.
9327 @node Solaris Pragmas
9328 @subsection Solaris Pragmas
9330 The Solaris target supports @code{#pragma redefine_extname}
9331 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
9332 @code{#pragma} directives for compatibility with the system compiler.
9335 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
9336 @cindex pragma, align
9338 Increase the minimum alignment of each @var{variable} to @var{alignment}.
9339 This is the same as GCC's @code{aligned} attribute @pxref{Variable
9340 Attributes}). Macro expansion occurs on the arguments to this pragma
9341 when compiling C and Objective-C. It does not currently occur when
9342 compiling C++, but this is a bug which may be fixed in a future
9345 @item fini (@var{function} [, @var{function}]...)
9346 @cindex pragma, fini
9348 This pragma causes each listed @var{function} to be called after
9349 main, or during shared module unloading, by adding a call to the
9350 @code{.fini} section.
9352 @item init (@var{function} [, @var{function}]...)
9353 @cindex pragma, init
9355 This pragma causes each listed @var{function} to be called during
9356 initialization (before @code{main}) or during shared module loading, by
9357 adding a call to the @code{.init} section.
9361 @node Symbol-Renaming Pragmas
9362 @subsection Symbol-Renaming Pragmas
9364 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
9365 supports two @code{#pragma} directives which change the name used in
9366 assembly for a given declaration. These pragmas are only available on
9367 platforms whose system headers need them. To get this effect on all
9368 platforms supported by GCC, use the asm labels extension (@pxref{Asm
9372 @item redefine_extname @var{oldname} @var{newname}
9373 @cindex pragma, redefine_extname
9375 This pragma gives the C function @var{oldname} the assembly symbol
9376 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
9377 will be defined if this pragma is available (currently only on
9380 @item extern_prefix @var{string}
9381 @cindex pragma, extern_prefix
9383 This pragma causes all subsequent external function and variable
9384 declarations to have @var{string} prepended to their assembly symbols.
9385 This effect may be terminated with another @code{extern_prefix} pragma
9386 whose argument is an empty string. The preprocessor macro
9387 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
9388 available (currently only on Tru64 UNIX)@.
9391 These pragmas and the asm labels extension interact in a complicated
9392 manner. Here are some corner cases you may want to be aware of.
9395 @item Both pragmas silently apply only to declarations with external
9396 linkage. Asm labels do not have this restriction.
9398 @item In C++, both pragmas silently apply only to declarations with
9399 ``C'' linkage. Again, asm labels do not have this restriction.
9401 @item If any of the three ways of changing the assembly name of a
9402 declaration is applied to a declaration whose assembly name has
9403 already been determined (either by a previous use of one of these
9404 features, or because the compiler needed the assembly name in order to
9405 generate code), and the new name is different, a warning issues and
9406 the name does not change.
9408 @item The @var{oldname} used by @code{#pragma redefine_extname} is
9409 always the C-language name.
9411 @item If @code{#pragma extern_prefix} is in effect, and a declaration
9412 occurs with an asm label attached, the prefix is silently ignored for
9415 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
9416 apply to the same declaration, whichever triggered first wins, and a
9417 warning issues if they contradict each other. (We would like to have
9418 @code{#pragma redefine_extname} always win, for consistency with asm
9419 labels, but if @code{#pragma extern_prefix} triggers first we have no
9420 way of knowing that that happened.)
9423 @node Structure-Packing Pragmas
9424 @subsection Structure-Packing Pragmas
9426 For compatibility with Win32, GCC supports a set of @code{#pragma}
9427 directives which change the maximum alignment of members of structures
9428 (other than zero-width bitfields), unions, and classes subsequently
9429 defined. The @var{n} value below always is required to be a small power
9430 of two and specifies the new alignment in bytes.
9433 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
9434 @item @code{#pragma pack()} sets the alignment to the one that was in
9435 effect when compilation started (see also command line option
9436 @option{-fpack-struct[=<n>]} @pxref{Code Gen Options}).
9437 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
9438 setting on an internal stack and then optionally sets the new alignment.
9439 @item @code{#pragma pack(pop)} restores the alignment setting to the one
9440 saved at the top of the internal stack (and removes that stack entry).
9441 Note that @code{#pragma pack([@var{n}])} does not influence this internal
9442 stack; thus it is possible to have @code{#pragma pack(push)} followed by
9443 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
9444 @code{#pragma pack(pop)}.
9448 @subsection Weak Pragmas
9450 For compatibility with SVR4, GCC supports a set of @code{#pragma}
9451 directives for declaring symbols to be weak, and defining weak
9455 @item #pragma weak @var{symbol}
9456 @cindex pragma, weak
9457 This pragma declares @var{symbol} to be weak, as if the declaration
9458 had the attribute of the same name. The pragma may appear before
9459 or after the declaration of @var{symbol}, but must appear before
9460 either its first use or its definition. It is not an error for
9461 @var{symbol} to never be defined at all.
9463 @item #pragma weak @var{symbol1} = @var{symbol2}
9464 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
9465 It is an error if @var{symbol2} is not defined in the current
9469 @node Unnamed Fields
9470 @section Unnamed struct/union fields within structs/unions
9474 For compatibility with other compilers, GCC allows you to define
9475 a structure or union that contains, as fields, structures and unions
9476 without names. For example:
9489 In this example, the user would be able to access members of the unnamed
9490 union with code like @samp{foo.b}. Note that only unnamed structs and
9491 unions are allowed, you may not have, for example, an unnamed
9494 You must never create such structures that cause ambiguous field definitions.
9495 For example, this structure:
9506 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
9507 Such constructs are not supported and must be avoided. In the future,
9508 such constructs may be detected and treated as compilation errors.
9510 @opindex fms-extensions
9511 Unless @option{-fms-extensions} is used, the unnamed field must be a
9512 structure or union definition without a tag (for example, @samp{struct
9513 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
9514 also be a definition with a tag such as @samp{struct foo @{ int a;
9515 @};}, a reference to a previously defined structure or union such as
9516 @samp{struct foo;}, or a reference to a @code{typedef} name for a
9517 previously defined structure or union type.
9520 @section Thread-Local Storage
9521 @cindex Thread-Local Storage
9522 @cindex @acronym{TLS}
9525 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
9526 are allocated such that there is one instance of the variable per extant
9527 thread. The run-time model GCC uses to implement this originates
9528 in the IA-64 processor-specific ABI, but has since been migrated
9529 to other processors as well. It requires significant support from
9530 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
9531 system libraries (@file{libc.so} and @file{libpthread.so}), so it
9532 is not available everywhere.
9534 At the user level, the extension is visible with a new storage
9535 class keyword: @code{__thread}. For example:
9539 extern __thread struct state s;
9540 static __thread char *p;
9543 The @code{__thread} specifier may be used alone, with the @code{extern}
9544 or @code{static} specifiers, but with no other storage class specifier.
9545 When used with @code{extern} or @code{static}, @code{__thread} must appear
9546 immediately after the other storage class specifier.
9548 The @code{__thread} specifier may be applied to any global, file-scoped
9549 static, function-scoped static, or static data member of a class. It may
9550 not be applied to block-scoped automatic or non-static data member.
9552 When the address-of operator is applied to a thread-local variable, it is
9553 evaluated at run-time and returns the address of the current thread's
9554 instance of that variable. An address so obtained may be used by any
9555 thread. When a thread terminates, any pointers to thread-local variables
9556 in that thread become invalid.
9558 No static initialization may refer to the address of a thread-local variable.
9560 In C++, if an initializer is present for a thread-local variable, it must
9561 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
9564 See @uref{http://people.redhat.com/drepper/tls.pdf,
9565 ELF Handling For Thread-Local Storage} for a detailed explanation of
9566 the four thread-local storage addressing models, and how the run-time
9567 is expected to function.
9570 * C99 Thread-Local Edits::
9571 * C++98 Thread-Local Edits::
9574 @node C99 Thread-Local Edits
9575 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
9577 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
9578 that document the exact semantics of the language extension.
9582 @cite{5.1.2 Execution environments}
9584 Add new text after paragraph 1
9587 Within either execution environment, a @dfn{thread} is a flow of
9588 control within a program. It is implementation defined whether
9589 or not there may be more than one thread associated with a program.
9590 It is implementation defined how threads beyond the first are
9591 created, the name and type of the function called at thread
9592 startup, and how threads may be terminated. However, objects
9593 with thread storage duration shall be initialized before thread
9598 @cite{6.2.4 Storage durations of objects}
9600 Add new text before paragraph 3
9603 An object whose identifier is declared with the storage-class
9604 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
9605 Its lifetime is the entire execution of the thread, and its
9606 stored value is initialized only once, prior to thread startup.
9610 @cite{6.4.1 Keywords}
9612 Add @code{__thread}.
9615 @cite{6.7.1 Storage-class specifiers}
9617 Add @code{__thread} to the list of storage class specifiers in
9620 Change paragraph 2 to
9623 With the exception of @code{__thread}, at most one storage-class
9624 specifier may be given [@dots{}]. The @code{__thread} specifier may
9625 be used alone, or immediately following @code{extern} or
9629 Add new text after paragraph 6
9632 The declaration of an identifier for a variable that has
9633 block scope that specifies @code{__thread} shall also
9634 specify either @code{extern} or @code{static}.
9636 The @code{__thread} specifier shall be used only with
9641 @node C++98 Thread-Local Edits
9642 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
9644 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
9645 that document the exact semantics of the language extension.
9649 @b{[intro.execution]}
9651 New text after paragraph 4
9654 A @dfn{thread} is a flow of control within the abstract machine.
9655 It is implementation defined whether or not there may be more than
9659 New text after paragraph 7
9662 It is unspecified whether additional action must be taken to
9663 ensure when and whether side effects are visible to other threads.
9669 Add @code{__thread}.
9672 @b{[basic.start.main]}
9674 Add after paragraph 5
9677 The thread that begins execution at the @code{main} function is called
9678 the @dfn{main thread}. It is implementation defined how functions
9679 beginning threads other than the main thread are designated or typed.
9680 A function so designated, as well as the @code{main} function, is called
9681 a @dfn{thread startup function}. It is implementation defined what
9682 happens if a thread startup function returns. It is implementation
9683 defined what happens to other threads when any thread calls @code{exit}.
9687 @b{[basic.start.init]}
9689 Add after paragraph 4
9692 The storage for an object of thread storage duration shall be
9693 statically initialized before the first statement of the thread startup
9694 function. An object of thread storage duration shall not require
9695 dynamic initialization.
9699 @b{[basic.start.term]}
9701 Add after paragraph 3
9704 The type of an object with thread storage duration shall not have a
9705 non-trivial destructor, nor shall it be an array type whose elements
9706 (directly or indirectly) have non-trivial destructors.
9712 Add ``thread storage duration'' to the list in paragraph 1.
9717 Thread, static, and automatic storage durations are associated with
9718 objects introduced by declarations [@dots{}].
9721 Add @code{__thread} to the list of specifiers in paragraph 3.
9724 @b{[basic.stc.thread]}
9726 New section before @b{[basic.stc.static]}
9729 The keyword @code{__thread} applied to a non-local object gives the
9730 object thread storage duration.
9732 A local variable or class data member declared both @code{static}
9733 and @code{__thread} gives the variable or member thread storage
9738 @b{[basic.stc.static]}
9743 All objects which have neither thread storage duration, dynamic
9744 storage duration nor are local [@dots{}].
9750 Add @code{__thread} to the list in paragraph 1.
9755 With the exception of @code{__thread}, at most one
9756 @var{storage-class-specifier} shall appear in a given
9757 @var{decl-specifier-seq}. The @code{__thread} specifier may
9758 be used alone, or immediately following the @code{extern} or
9759 @code{static} specifiers. [@dots{}]
9762 Add after paragraph 5
9765 The @code{__thread} specifier can be applied only to the names of objects
9766 and to anonymous unions.
9772 Add after paragraph 6
9775 Non-@code{static} members shall not be @code{__thread}.
9779 @node C++ Extensions
9780 @chapter Extensions to the C++ Language
9781 @cindex extensions, C++ language
9782 @cindex C++ language extensions
9784 The GNU compiler provides these extensions to the C++ language (and you
9785 can also use most of the C language extensions in your C++ programs). If you
9786 want to write code that checks whether these features are available, you can
9787 test for the GNU compiler the same way as for C programs: check for a
9788 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
9789 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
9790 Predefined Macros,cpp,The GNU C Preprocessor}).
9793 * Volatiles:: What constitutes an access to a volatile object.
9794 * Restricted Pointers:: C99 restricted pointers and references.
9795 * Vague Linkage:: Where G++ puts inlines, vtables and such.
9796 * C++ Interface:: You can use a single C++ header file for both
9797 declarations and definitions.
9798 * Template Instantiation:: Methods for ensuring that exactly one copy of
9799 each needed template instantiation is emitted.
9800 * Bound member functions:: You can extract a function pointer to the
9801 method denoted by a @samp{->*} or @samp{.*} expression.
9802 * C++ Attributes:: Variable, function, and type attributes for C++ only.
9803 * Strong Using:: Strong using-directives for namespace composition.
9804 * Java Exceptions:: Tweaking exception handling to work with Java.
9805 * Deprecated Features:: Things will disappear from g++.
9806 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
9810 @section When is a Volatile Object Accessed?
9811 @cindex accessing volatiles
9812 @cindex volatile read
9813 @cindex volatile write
9814 @cindex volatile access
9816 Both the C and C++ standard have the concept of volatile objects. These
9817 are normally accessed by pointers and used for accessing hardware. The
9818 standards encourage compilers to refrain from optimizations
9819 concerning accesses to volatile objects that it might perform on
9820 non-volatile objects. The C standard leaves it implementation defined
9821 as to what constitutes a volatile access. The C++ standard omits to
9822 specify this, except to say that C++ should behave in a similar manner
9823 to C with respect to volatiles, where possible. The minimum either
9824 standard specifies is that at a sequence point all previous accesses to
9825 volatile objects have stabilized and no subsequent accesses have
9826 occurred. Thus an implementation is free to reorder and combine
9827 volatile accesses which occur between sequence points, but cannot do so
9828 for accesses across a sequence point. The use of volatiles does not
9829 allow you to violate the restriction on updating objects multiple times
9830 within a sequence point.
9832 In most expressions, it is intuitively obvious what is a read and what is
9833 a write. For instance
9836 volatile int *dst = @var{somevalue};
9837 volatile int *src = @var{someothervalue};
9842 will cause a read of the volatile object pointed to by @var{src} and stores the
9843 value into the volatile object pointed to by @var{dst}. There is no
9844 guarantee that these reads and writes are atomic, especially for objects
9845 larger than @code{int}.
9847 Less obvious expressions are where something which looks like an access
9848 is used in a void context. An example would be,
9851 volatile int *src = @var{somevalue};
9855 With C, such expressions are rvalues, and as rvalues cause a read of
9856 the object, GCC interprets this as a read of the volatile being pointed
9857 to. The C++ standard specifies that such expressions do not undergo
9858 lvalue to rvalue conversion, and that the type of the dereferenced
9859 object may be incomplete. The C++ standard does not specify explicitly
9860 that it is this lvalue to rvalue conversion which is responsible for
9861 causing an access. However, there is reason to believe that it is,
9862 because otherwise certain simple expressions become undefined. However,
9863 because it would surprise most programmers, G++ treats dereferencing a
9864 pointer to volatile object of complete type in a void context as a read
9865 of the object. When the object has incomplete type, G++ issues a
9870 struct T @{int m;@};
9871 volatile S *ptr1 = @var{somevalue};
9872 volatile T *ptr2 = @var{somevalue};
9877 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
9878 causes a read of the object pointed to. If you wish to force an error on
9879 the first case, you must force a conversion to rvalue with, for instance
9880 a static cast, @code{static_cast<S>(*ptr1)}.
9882 When using a reference to volatile, G++ does not treat equivalent
9883 expressions as accesses to volatiles, but instead issues a warning that
9884 no volatile is accessed. The rationale for this is that otherwise it
9885 becomes difficult to determine where volatile access occur, and not
9886 possible to ignore the return value from functions returning volatile
9887 references. Again, if you wish to force a read, cast the reference to
9890 @node Restricted Pointers
9891 @section Restricting Pointer Aliasing
9892 @cindex restricted pointers
9893 @cindex restricted references
9894 @cindex restricted this pointer
9896 As with the C front end, G++ understands the C99 feature of restricted pointers,
9897 specified with the @code{__restrict__}, or @code{__restrict} type
9898 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
9899 language flag, @code{restrict} is not a keyword in C++.
9901 In addition to allowing restricted pointers, you can specify restricted
9902 references, which indicate that the reference is not aliased in the local
9906 void fn (int *__restrict__ rptr, int &__restrict__ rref)
9913 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
9914 @var{rref} refers to a (different) unaliased integer.
9916 You may also specify whether a member function's @var{this} pointer is
9917 unaliased by using @code{__restrict__} as a member function qualifier.
9920 void T::fn () __restrict__
9927 Within the body of @code{T::fn}, @var{this} will have the effective
9928 definition @code{T *__restrict__ const this}. Notice that the
9929 interpretation of a @code{__restrict__} member function qualifier is
9930 different to that of @code{const} or @code{volatile} qualifier, in that it
9931 is applied to the pointer rather than the object. This is consistent with
9932 other compilers which implement restricted pointers.
9934 As with all outermost parameter qualifiers, @code{__restrict__} is
9935 ignored in function definition matching. This means you only need to
9936 specify @code{__restrict__} in a function definition, rather than
9937 in a function prototype as well.
9940 @section Vague Linkage
9941 @cindex vague linkage
9943 There are several constructs in C++ which require space in the object
9944 file but are not clearly tied to a single translation unit. We say that
9945 these constructs have ``vague linkage''. Typically such constructs are
9946 emitted wherever they are needed, though sometimes we can be more
9950 @item Inline Functions
9951 Inline functions are typically defined in a header file which can be
9952 included in many different compilations. Hopefully they can usually be
9953 inlined, but sometimes an out-of-line copy is necessary, if the address
9954 of the function is taken or if inlining fails. In general, we emit an
9955 out-of-line copy in all translation units where one is needed. As an
9956 exception, we only emit inline virtual functions with the vtable, since
9957 it will always require a copy.
9959 Local static variables and string constants used in an inline function
9960 are also considered to have vague linkage, since they must be shared
9961 between all inlined and out-of-line instances of the function.
9965 C++ virtual functions are implemented in most compilers using a lookup
9966 table, known as a vtable. The vtable contains pointers to the virtual
9967 functions provided by a class, and each object of the class contains a
9968 pointer to its vtable (or vtables, in some multiple-inheritance
9969 situations). If the class declares any non-inline, non-pure virtual
9970 functions, the first one is chosen as the ``key method'' for the class,
9971 and the vtable is only emitted in the translation unit where the key
9974 @emph{Note:} If the chosen key method is later defined as inline, the
9975 vtable will still be emitted in every translation unit which defines it.
9976 Make sure that any inline virtuals are declared inline in the class
9977 body, even if they are not defined there.
9979 @item type_info objects
9982 C++ requires information about types to be written out in order to
9983 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
9984 For polymorphic classes (classes with virtual functions), the type_info
9985 object is written out along with the vtable so that @samp{dynamic_cast}
9986 can determine the dynamic type of a class object at runtime. For all
9987 other types, we write out the type_info object when it is used: when
9988 applying @samp{typeid} to an expression, throwing an object, or
9989 referring to a type in a catch clause or exception specification.
9991 @item Template Instantiations
9992 Most everything in this section also applies to template instantiations,
9993 but there are other options as well.
9994 @xref{Template Instantiation,,Where's the Template?}.
9998 When used with GNU ld version 2.8 or later on an ELF system such as
9999 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
10000 these constructs will be discarded at link time. This is known as
10003 On targets that don't support COMDAT, but do support weak symbols, GCC
10004 will use them. This way one copy will override all the others, but
10005 the unused copies will still take up space in the executable.
10007 For targets which do not support either COMDAT or weak symbols,
10008 most entities with vague linkage will be emitted as local symbols to
10009 avoid duplicate definition errors from the linker. This will not happen
10010 for local statics in inlines, however, as having multiple copies will
10011 almost certainly break things.
10013 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
10014 another way to control placement of these constructs.
10016 @node C++ Interface
10017 @section #pragma interface and implementation
10019 @cindex interface and implementation headers, C++
10020 @cindex C++ interface and implementation headers
10021 @cindex pragmas, interface and implementation
10023 @code{#pragma interface} and @code{#pragma implementation} provide the
10024 user with a way of explicitly directing the compiler to emit entities
10025 with vague linkage (and debugging information) in a particular
10028 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
10029 most cases, because of COMDAT support and the ``key method'' heuristic
10030 mentioned in @ref{Vague Linkage}. Using them can actually cause your
10031 program to grow due to unnecessary out-of-line copies of inline
10032 functions. Currently (3.4) the only benefit of these
10033 @code{#pragma}s is reduced duplication of debugging information, and
10034 that should be addressed soon on DWARF 2 targets with the use of
10038 @item #pragma interface
10039 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
10040 @kindex #pragma interface
10041 Use this directive in @emph{header files} that define object classes, to save
10042 space in most of the object files that use those classes. Normally,
10043 local copies of certain information (backup copies of inline member
10044 functions, debugging information, and the internal tables that implement
10045 virtual functions) must be kept in each object file that includes class
10046 definitions. You can use this pragma to avoid such duplication. When a
10047 header file containing @samp{#pragma interface} is included in a
10048 compilation, this auxiliary information will not be generated (unless
10049 the main input source file itself uses @samp{#pragma implementation}).
10050 Instead, the object files will contain references to be resolved at link
10053 The second form of this directive is useful for the case where you have
10054 multiple headers with the same name in different directories. If you
10055 use this form, you must specify the same string to @samp{#pragma
10058 @item #pragma implementation
10059 @itemx #pragma implementation "@var{objects}.h"
10060 @kindex #pragma implementation
10061 Use this pragma in a @emph{main input file}, when you want full output from
10062 included header files to be generated (and made globally visible). The
10063 included header file, in turn, should use @samp{#pragma interface}.
10064 Backup copies of inline member functions, debugging information, and the
10065 internal tables used to implement virtual functions are all generated in
10066 implementation files.
10068 @cindex implied @code{#pragma implementation}
10069 @cindex @code{#pragma implementation}, implied
10070 @cindex naming convention, implementation headers
10071 If you use @samp{#pragma implementation} with no argument, it applies to
10072 an include file with the same basename@footnote{A file's @dfn{basename}
10073 was the name stripped of all leading path information and of trailing
10074 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
10075 file. For example, in @file{allclass.cc}, giving just
10076 @samp{#pragma implementation}
10077 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
10079 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
10080 an implementation file whenever you would include it from
10081 @file{allclass.cc} even if you never specified @samp{#pragma
10082 implementation}. This was deemed to be more trouble than it was worth,
10083 however, and disabled.
10085 Use the string argument if you want a single implementation file to
10086 include code from multiple header files. (You must also use
10087 @samp{#include} to include the header file; @samp{#pragma
10088 implementation} only specifies how to use the file---it doesn't actually
10091 There is no way to split up the contents of a single header file into
10092 multiple implementation files.
10095 @cindex inlining and C++ pragmas
10096 @cindex C++ pragmas, effect on inlining
10097 @cindex pragmas in C++, effect on inlining
10098 @samp{#pragma implementation} and @samp{#pragma interface} also have an
10099 effect on function inlining.
10101 If you define a class in a header file marked with @samp{#pragma
10102 interface}, the effect on an inline function defined in that class is
10103 similar to an explicit @code{extern} declaration---the compiler emits
10104 no code at all to define an independent version of the function. Its
10105 definition is used only for inlining with its callers.
10107 @opindex fno-implement-inlines
10108 Conversely, when you include the same header file in a main source file
10109 that declares it as @samp{#pragma implementation}, the compiler emits
10110 code for the function itself; this defines a version of the function
10111 that can be found via pointers (or by callers compiled without
10112 inlining). If all calls to the function can be inlined, you can avoid
10113 emitting the function by compiling with @option{-fno-implement-inlines}.
10114 If any calls were not inlined, you will get linker errors.
10116 @node Template Instantiation
10117 @section Where's the Template?
10118 @cindex template instantiation
10120 C++ templates are the first language feature to require more
10121 intelligence from the environment than one usually finds on a UNIX
10122 system. Somehow the compiler and linker have to make sure that each
10123 template instance occurs exactly once in the executable if it is needed,
10124 and not at all otherwise. There are two basic approaches to this
10125 problem, which are referred to as the Borland model and the Cfront model.
10128 @item Borland model
10129 Borland C++ solved the template instantiation problem by adding the code
10130 equivalent of common blocks to their linker; the compiler emits template
10131 instances in each translation unit that uses them, and the linker
10132 collapses them together. The advantage of this model is that the linker
10133 only has to consider the object files themselves; there is no external
10134 complexity to worry about. This disadvantage is that compilation time
10135 is increased because the template code is being compiled repeatedly.
10136 Code written for this model tends to include definitions of all
10137 templates in the header file, since they must be seen to be
10141 The AT&T C++ translator, Cfront, solved the template instantiation
10142 problem by creating the notion of a template repository, an
10143 automatically maintained place where template instances are stored. A
10144 more modern version of the repository works as follows: As individual
10145 object files are built, the compiler places any template definitions and
10146 instantiations encountered in the repository. At link time, the link
10147 wrapper adds in the objects in the repository and compiles any needed
10148 instances that were not previously emitted. The advantages of this
10149 model are more optimal compilation speed and the ability to use the
10150 system linker; to implement the Borland model a compiler vendor also
10151 needs to replace the linker. The disadvantages are vastly increased
10152 complexity, and thus potential for error; for some code this can be
10153 just as transparent, but in practice it can been very difficult to build
10154 multiple programs in one directory and one program in multiple
10155 directories. Code written for this model tends to separate definitions
10156 of non-inline member templates into a separate file, which should be
10157 compiled separately.
10160 When used with GNU ld version 2.8 or later on an ELF system such as
10161 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
10162 Borland model. On other systems, G++ implements neither automatic
10165 A future version of G++ will support a hybrid model whereby the compiler
10166 will emit any instantiations for which the template definition is
10167 included in the compile, and store template definitions and
10168 instantiation context information into the object file for the rest.
10169 The link wrapper will extract that information as necessary and invoke
10170 the compiler to produce the remaining instantiations. The linker will
10171 then combine duplicate instantiations.
10173 In the mean time, you have the following options for dealing with
10174 template instantiations:
10179 Compile your template-using code with @option{-frepo}. The compiler will
10180 generate files with the extension @samp{.rpo} listing all of the
10181 template instantiations used in the corresponding object files which
10182 could be instantiated there; the link wrapper, @samp{collect2}, will
10183 then update the @samp{.rpo} files to tell the compiler where to place
10184 those instantiations and rebuild any affected object files. The
10185 link-time overhead is negligible after the first pass, as the compiler
10186 will continue to place the instantiations in the same files.
10188 This is your best option for application code written for the Borland
10189 model, as it will just work. Code written for the Cfront model will
10190 need to be modified so that the template definitions are available at
10191 one or more points of instantiation; usually this is as simple as adding
10192 @code{#include <tmethods.cc>} to the end of each template header.
10194 For library code, if you want the library to provide all of the template
10195 instantiations it needs, just try to link all of its object files
10196 together; the link will fail, but cause the instantiations to be
10197 generated as a side effect. Be warned, however, that this may cause
10198 conflicts if multiple libraries try to provide the same instantiations.
10199 For greater control, use explicit instantiation as described in the next
10203 @opindex fno-implicit-templates
10204 Compile your code with @option{-fno-implicit-templates} to disable the
10205 implicit generation of template instances, and explicitly instantiate
10206 all the ones you use. This approach requires more knowledge of exactly
10207 which instances you need than do the others, but it's less
10208 mysterious and allows greater control. You can scatter the explicit
10209 instantiations throughout your program, perhaps putting them in the
10210 translation units where the instances are used or the translation units
10211 that define the templates themselves; you can put all of the explicit
10212 instantiations you need into one big file; or you can create small files
10219 template class Foo<int>;
10220 template ostream& operator <<
10221 (ostream&, const Foo<int>&);
10224 for each of the instances you need, and create a template instantiation
10225 library from those.
10227 If you are using Cfront-model code, you can probably get away with not
10228 using @option{-fno-implicit-templates} when compiling files that don't
10229 @samp{#include} the member template definitions.
10231 If you use one big file to do the instantiations, you may want to
10232 compile it without @option{-fno-implicit-templates} so you get all of the
10233 instances required by your explicit instantiations (but not by any
10234 other files) without having to specify them as well.
10236 G++ has extended the template instantiation syntax given in the ISO
10237 standard to allow forward declaration of explicit instantiations
10238 (with @code{extern}), instantiation of the compiler support data for a
10239 template class (i.e.@: the vtable) without instantiating any of its
10240 members (with @code{inline}), and instantiation of only the static data
10241 members of a template class, without the support data or member
10242 functions (with (@code{static}):
10245 extern template int max (int, int);
10246 inline template class Foo<int>;
10247 static template class Foo<int>;
10251 Do nothing. Pretend G++ does implement automatic instantiation
10252 management. Code written for the Borland model will work fine, but
10253 each translation unit will contain instances of each of the templates it
10254 uses. In a large program, this can lead to an unacceptable amount of code
10258 @node Bound member functions
10259 @section Extracting the function pointer from a bound pointer to member function
10261 @cindex pointer to member function
10262 @cindex bound pointer to member function
10264 In C++, pointer to member functions (PMFs) are implemented using a wide
10265 pointer of sorts to handle all the possible call mechanisms; the PMF
10266 needs to store information about how to adjust the @samp{this} pointer,
10267 and if the function pointed to is virtual, where to find the vtable, and
10268 where in the vtable to look for the member function. If you are using
10269 PMFs in an inner loop, you should really reconsider that decision. If
10270 that is not an option, you can extract the pointer to the function that
10271 would be called for a given object/PMF pair and call it directly inside
10272 the inner loop, to save a bit of time.
10274 Note that you will still be paying the penalty for the call through a
10275 function pointer; on most modern architectures, such a call defeats the
10276 branch prediction features of the CPU@. This is also true of normal
10277 virtual function calls.
10279 The syntax for this extension is
10283 extern int (A::*fp)();
10284 typedef int (*fptr)(A *);
10286 fptr p = (fptr)(a.*fp);
10289 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
10290 no object is needed to obtain the address of the function. They can be
10291 converted to function pointers directly:
10294 fptr p1 = (fptr)(&A::foo);
10297 @opindex Wno-pmf-conversions
10298 You must specify @option{-Wno-pmf-conversions} to use this extension.
10300 @node C++ Attributes
10301 @section C++-Specific Variable, Function, and Type Attributes
10303 Some attributes only make sense for C++ programs.
10306 @item init_priority (@var{priority})
10307 @cindex init_priority attribute
10310 In Standard C++, objects defined at namespace scope are guaranteed to be
10311 initialized in an order in strict accordance with that of their definitions
10312 @emph{in a given translation unit}. No guarantee is made for initializations
10313 across translation units. However, GNU C++ allows users to control the
10314 order of initialization of objects defined at namespace scope with the
10315 @code{init_priority} attribute by specifying a relative @var{priority},
10316 a constant integral expression currently bounded between 101 and 65535
10317 inclusive. Lower numbers indicate a higher priority.
10319 In the following example, @code{A} would normally be created before
10320 @code{B}, but the @code{init_priority} attribute has reversed that order:
10323 Some_Class A __attribute__ ((init_priority (2000)));
10324 Some_Class B __attribute__ ((init_priority (543)));
10328 Note that the particular values of @var{priority} do not matter; only their
10331 @item java_interface
10332 @cindex java_interface attribute
10334 This type attribute informs C++ that the class is a Java interface. It may
10335 only be applied to classes declared within an @code{extern "Java"} block.
10336 Calls to methods declared in this interface will be dispatched using GCJ's
10337 interface table mechanism, instead of regular virtual table dispatch.
10341 See also @xref{Strong Using}.
10344 @section Strong Using
10346 @strong{Caution:} The semantics of this extension are not fully
10347 defined. Users should refrain from using this extension as its
10348 semantics may change subtly over time. It is possible that this
10349 extension wil be removed in future versions of G++.
10351 A using-directive with @code{__attribute ((strong))} is stronger
10352 than a normal using-directive in two ways:
10356 Templates from the used namespace can be specialized as though they were members of the using namespace.
10359 The using namespace is considered an associated namespace of all
10360 templates in the used namespace for purposes of argument-dependent
10364 This is useful for composing a namespace transparently from
10365 implementation namespaces. For example:
10370 template <class T> struct A @{ @};
10372 using namespace debug __attribute ((__strong__));
10373 template <> struct A<int> @{ @}; // @r{ok to specialize}
10375 template <class T> void f (A<T>);
10380 f (std::A<float>()); // @r{lookup finds} std::f
10385 @node Java Exceptions
10386 @section Java Exceptions
10388 The Java language uses a slightly different exception handling model
10389 from C++. Normally, GNU C++ will automatically detect when you are
10390 writing C++ code that uses Java exceptions, and handle them
10391 appropriately. However, if C++ code only needs to execute destructors
10392 when Java exceptions are thrown through it, GCC will guess incorrectly.
10393 Sample problematic code is:
10396 struct S @{ ~S(); @};
10397 extern void bar(); // @r{is written in Java, and may throw exceptions}
10406 The usual effect of an incorrect guess is a link failure, complaining of
10407 a missing routine called @samp{__gxx_personality_v0}.
10409 You can inform the compiler that Java exceptions are to be used in a
10410 translation unit, irrespective of what it might think, by writing
10411 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
10412 @samp{#pragma} must appear before any functions that throw or catch
10413 exceptions, or run destructors when exceptions are thrown through them.
10415 You cannot mix Java and C++ exceptions in the same translation unit. It
10416 is believed to be safe to throw a C++ exception from one file through
10417 another file compiled for the Java exception model, or vice versa, but
10418 there may be bugs in this area.
10420 @node Deprecated Features
10421 @section Deprecated Features
10423 In the past, the GNU C++ compiler was extended to experiment with new
10424 features, at a time when the C++ language was still evolving. Now that
10425 the C++ standard is complete, some of those features are superseded by
10426 superior alternatives. Using the old features might cause a warning in
10427 some cases that the feature will be dropped in the future. In other
10428 cases, the feature might be gone already.
10430 While the list below is not exhaustive, it documents some of the options
10431 that are now deprecated:
10434 @item -fexternal-templates
10435 @itemx -falt-external-templates
10436 These are two of the many ways for G++ to implement template
10437 instantiation. @xref{Template Instantiation}. The C++ standard clearly
10438 defines how template definitions have to be organized across
10439 implementation units. G++ has an implicit instantiation mechanism that
10440 should work just fine for standard-conforming code.
10442 @item -fstrict-prototype
10443 @itemx -fno-strict-prototype
10444 Previously it was possible to use an empty prototype parameter list to
10445 indicate an unspecified number of parameters (like C), rather than no
10446 parameters, as C++ demands. This feature has been removed, except where
10447 it is required for backwards compatibility @xref{Backwards Compatibility}.
10450 G++ allows a virtual function returning @samp{void *} to be overridden
10451 by one returning a different pointer type. This extension to the
10452 covariant return type rules is now deprecated and will be removed from a
10455 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
10456 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
10457 and will be removed in a future version. Code using these operators
10458 should be modified to use @code{std::min} and @code{std::max} instead.
10460 The named return value extension has been deprecated, and is now
10463 The use of initializer lists with new expressions has been deprecated,
10464 and is now removed from G++.
10466 Floating and complex non-type template parameters have been deprecated,
10467 and are now removed from G++.
10469 The implicit typename extension has been deprecated and is now
10472 The use of default arguments in function pointers, function typedefs and
10473 and other places where they are not permitted by the standard is
10474 deprecated and will be removed from a future version of G++.
10476 G++ allows floating-point literals to appear in integral constant expressions,
10477 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
10478 This extension is deprecated and will be removed from a future version.
10480 G++ allows static data members of const floating-point type to be declared
10481 with an initializer in a class definition. The standard only allows
10482 initializers for static members of const integral types and const
10483 enumeration types so this extension has been deprecated and will be removed
10484 from a future version.
10486 @node Backwards Compatibility
10487 @section Backwards Compatibility
10488 @cindex Backwards Compatibility
10489 @cindex ARM [Annotated C++ Reference Manual]
10491 Now that there is a definitive ISO standard C++, G++ has a specification
10492 to adhere to. The C++ language evolved over time, and features that
10493 used to be acceptable in previous drafts of the standard, such as the ARM
10494 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
10495 compilation of C++ written to such drafts, G++ contains some backwards
10496 compatibilities. @emph{All such backwards compatibility features are
10497 liable to disappear in future versions of G++.} They should be considered
10498 deprecated @xref{Deprecated Features}.
10502 If a variable is declared at for scope, it used to remain in scope until
10503 the end of the scope which contained the for statement (rather than just
10504 within the for scope). G++ retains this, but issues a warning, if such a
10505 variable is accessed outside the for scope.
10507 @item Implicit C language
10508 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
10509 scope to set the language. On such systems, all header files are
10510 implicitly scoped inside a C language scope. Also, an empty prototype
10511 @code{()} will be treated as an unspecified number of arguments, rather
10512 than no arguments, as C++ demands.