1 @c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001,2002,2003,2004
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 * Other Builtins:: Other built-in functions.
74 * Target Builtins:: Built-in functions specific to particular targets.
75 * Target Format Checks:: Format checks specific to particular targets.
76 * Pragmas:: Pragmas accepted by GCC.
77 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
78 * Thread-Local:: Per-thread variables.
82 @section Statements and Declarations in Expressions
83 @cindex statements inside expressions
84 @cindex declarations inside expressions
85 @cindex expressions containing statements
86 @cindex macros, statements in expressions
88 @c the above section title wrapped and causes an underfull hbox.. i
89 @c changed it from "within" to "in". --mew 4feb93
90 A compound statement enclosed in parentheses may appear as an expression
91 in GNU C@. This allows you to use loops, switches, and local variables
94 Recall that a compound statement is a sequence of statements surrounded
95 by braces; in this construct, parentheses go around the braces. For
99 (@{ int y = foo (); int z;
106 is a valid (though slightly more complex than necessary) expression
107 for the absolute value of @code{foo ()}.
109 The last thing in the compound statement should be an expression
110 followed by a semicolon; the value of this subexpression serves as the
111 value of the entire construct. (If you use some other kind of statement
112 last within the braces, the construct has type @code{void}, and thus
113 effectively no value.)
115 This feature is especially useful in making macro definitions ``safe'' (so
116 that they evaluate each operand exactly once). For example, the
117 ``maximum'' function is commonly defined as a macro in standard C as
121 #define max(a,b) ((a) > (b) ? (a) : (b))
125 @cindex side effects, macro argument
126 But this definition computes either @var{a} or @var{b} twice, with bad
127 results if the operand has side effects. In GNU C, if you know the
128 type of the operands (here taken as @code{int}), you can define
129 the macro safely as follows:
132 #define maxint(a,b) \
133 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
136 Embedded statements are not allowed in constant expressions, such as
137 the value of an enumeration constant, the width of a bit-field, or
138 the initial value of a static variable.
140 If you don't know the type of the operand, you can still do this, but you
141 must use @code{typeof} (@pxref{Typeof}).
143 In G++, the result value of a statement expression undergoes array and
144 function pointer decay, and is returned by value to the enclosing
145 expression. For instance, if @code{A} is a class, then
154 will construct a temporary @code{A} object to hold the result of the
155 statement expression, and that will be used to invoke @code{Foo}.
156 Therefore the @code{this} pointer observed by @code{Foo} will not be the
159 Any temporaries created within a statement within a statement expression
160 will be destroyed at the statement's end. This makes statement
161 expressions inside macros slightly different from function calls. In
162 the latter case temporaries introduced during argument evaluation will
163 be destroyed at the end of the statement that includes the function
164 call. In the statement expression case they will be destroyed during
165 the statement expression. For instance,
168 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
169 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
179 will have different places where temporaries are destroyed. For the
180 @code{macro} case, the temporary @code{X} will be destroyed just after
181 the initialization of @code{b}. In the @code{function} case that
182 temporary will be destroyed when the function returns.
184 These considerations mean that it is probably a bad idea to use
185 statement-expressions of this form in header files that are designed to
186 work with C++. (Note that some versions of the GNU C Library contained
187 header files using statement-expression that lead to precisely this
191 @section Locally Declared Labels
193 @cindex macros, local labels
195 GCC allows you to declare @dfn{local labels} in any nested block
196 scope. A local label is just like an ordinary label, but you can
197 only reference it (with a @code{goto} statement, or by taking its
198 address) within the block in which it was declared.
200 A local label declaration looks like this:
203 __label__ @var{label};
210 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
213 Local label declarations must come at the beginning of the block,
214 before any ordinary declarations or statements.
216 The label declaration defines the label @emph{name}, but does not define
217 the label itself. You must do this in the usual way, with
218 @code{@var{label}:}, within the statements of the statement expression.
220 The local label feature is useful for complex macros. If a macro
221 contains nested loops, a @code{goto} can be useful for breaking out of
222 them. However, an ordinary label whose scope is the whole function
223 cannot be used: if the macro can be expanded several times in one
224 function, the label will be multiply defined in that function. A
225 local label avoids this problem. For example:
228 #define SEARCH(value, array, target) \
231 typeof (target) _SEARCH_target = (target); \
232 typeof (*(array)) *_SEARCH_array = (array); \
235 for (i = 0; i < max; i++) \
236 for (j = 0; j < max; j++) \
237 if (_SEARCH_array[i][j] == _SEARCH_target) \
238 @{ (value) = i; goto found; @} \
244 This could also be written using a statement-expression:
247 #define SEARCH(array, target) \
250 typeof (target) _SEARCH_target = (target); \
251 typeof (*(array)) *_SEARCH_array = (array); \
254 for (i = 0; i < max; i++) \
255 for (j = 0; j < max; j++) \
256 if (_SEARCH_array[i][j] == _SEARCH_target) \
257 @{ value = i; goto found; @} \
264 Local label declarations also make the labels they declare visible to
265 nested functions, if there are any. @xref{Nested Functions}, for details.
267 @node Labels as Values
268 @section Labels as Values
269 @cindex labels as values
270 @cindex computed gotos
271 @cindex goto with computed label
272 @cindex address of a label
274 You can get the address of a label defined in the current function
275 (or a containing function) with the unary operator @samp{&&}. The
276 value has type @code{void *}. This value is a constant and can be used
277 wherever a constant of that type is valid. For example:
285 To use these values, you need to be able to jump to one. This is done
286 with the computed goto statement@footnote{The analogous feature in
287 Fortran is called an assigned goto, but that name seems inappropriate in
288 C, where one can do more than simply store label addresses in label
289 variables.}, @code{goto *@var{exp};}. For example,
296 Any expression of type @code{void *} is allowed.
298 One way of using these constants is in initializing a static array that
299 will serve as a jump table:
302 static void *array[] = @{ &&foo, &&bar, &&hack @};
305 Then you can select a label with indexing, like this:
312 Note that this does not check whether the subscript is in bounds---array
313 indexing in C never does that.
315 Such an array of label values serves a purpose much like that of the
316 @code{switch} statement. The @code{switch} statement is cleaner, so
317 use that rather than an array unless the problem does not fit a
318 @code{switch} statement very well.
320 Another use of label values is in an interpreter for threaded code.
321 The labels within the interpreter function can be stored in the
322 threaded code for super-fast dispatching.
324 You may not use this mechanism to jump to code in a different function.
325 If you do that, totally unpredictable things will happen. The best way to
326 avoid this is to store the label address only in automatic variables and
327 never pass it as an argument.
329 An alternate way to write the above example is
332 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
334 goto *(&&foo + array[i]);
338 This is more friendly to code living in shared libraries, as it reduces
339 the number of dynamic relocations that are needed, and by consequence,
340 allows the data to be read-only.
342 @node Nested Functions
343 @section Nested Functions
344 @cindex nested functions
345 @cindex downward funargs
348 A @dfn{nested function} is a function defined inside another function.
349 (Nested functions are not supported for GNU C++.) The nested function's
350 name is local to the block where it is defined. For example, here we
351 define a nested function named @code{square}, and call it twice:
355 foo (double a, double b)
357 double square (double z) @{ return z * z; @}
359 return square (a) + square (b);
364 The nested function can access all the variables of the containing
365 function that are visible at the point of its definition. This is
366 called @dfn{lexical scoping}. For example, here we show a nested
367 function which uses an inherited variable named @code{offset}:
371 bar (int *array, int offset, int size)
373 int access (int *array, int index)
374 @{ return array[index + offset]; @}
377 for (i = 0; i < size; i++)
378 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
383 Nested function definitions are permitted within functions in the places
384 where variable definitions are allowed; that is, in any block, before
385 the first statement in the block.
387 It is possible to call the nested function from outside the scope of its
388 name by storing its address or passing the address to another function:
391 hack (int *array, int size)
393 void store (int index, int value)
394 @{ array[index] = value; @}
396 intermediate (store, size);
400 Here, the function @code{intermediate} receives the address of
401 @code{store} as an argument. If @code{intermediate} calls @code{store},
402 the arguments given to @code{store} are used to store into @code{array}.
403 But this technique works only so long as the containing function
404 (@code{hack}, in this example) does not exit.
406 If you try to call the nested function through its address after the
407 containing function has exited, all hell will break loose. If you try
408 to call it after a containing scope level has exited, and if it refers
409 to some of the variables that are no longer in scope, you may be lucky,
410 but it's not wise to take the risk. If, however, the nested function
411 does not refer to anything that has gone out of scope, you should be
414 GCC implements taking the address of a nested function using a technique
415 called @dfn{trampolines}. A paper describing them is available as
418 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
420 A nested function can jump to a label inherited from a containing
421 function, provided the label was explicitly declared in the containing
422 function (@pxref{Local Labels}). Such a jump returns instantly to the
423 containing function, exiting the nested function which did the
424 @code{goto} and any intermediate functions as well. Here is an example:
428 bar (int *array, int offset, int size)
431 int access (int *array, int index)
435 return array[index + offset];
439 for (i = 0; i < size; i++)
440 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
444 /* @r{Control comes here from @code{access}
445 if it detects an error.} */
452 A nested function always has internal linkage. Declaring one with
453 @code{extern} is erroneous. If you need to declare the nested function
454 before its definition, use @code{auto} (which is otherwise meaningless
455 for function declarations).
458 bar (int *array, int offset, int size)
461 auto int access (int *, int);
463 int access (int *array, int index)
467 return array[index + offset];
473 @node Constructing Calls
474 @section Constructing Function Calls
475 @cindex constructing calls
476 @cindex forwarding calls
478 Using the built-in functions described below, you can record
479 the arguments a function received, and call another function
480 with the same arguments, without knowing the number or types
483 You can also record the return value of that function call,
484 and later return that value, without knowing what data type
485 the function tried to return (as long as your caller expects
488 However, these built-in functions may interact badly with some
489 sophisticated features or other extensions of the language. It
490 is, therefore, not recommended to use them outside very simple
491 functions acting as mere forwarders for their arguments.
493 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
494 This built-in function returns a pointer to data
495 describing how to perform a call with the same arguments as were passed
496 to the current function.
498 The function saves the arg pointer register, structure value address,
499 and all registers that might be used to pass arguments to a function
500 into a block of memory allocated on the stack. Then it returns the
501 address of that block.
504 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
505 This built-in function invokes @var{function}
506 with a copy of the parameters described by @var{arguments}
509 The value of @var{arguments} should be the value returned by
510 @code{__builtin_apply_args}. The argument @var{size} specifies the size
511 of the stack argument data, in bytes.
513 This function returns a pointer to data describing
514 how to return whatever value was returned by @var{function}. The data
515 is saved in a block of memory allocated on the stack.
517 It is not always simple to compute the proper value for @var{size}. The
518 value is used by @code{__builtin_apply} to compute the amount of data
519 that should be pushed on the stack and copied from the incoming argument
523 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
524 This built-in function returns the value described by @var{result} from
525 the containing function. You should specify, for @var{result}, a value
526 returned by @code{__builtin_apply}.
530 @section Referring to a Type with @code{typeof}
533 @cindex macros, types of arguments
535 Another way to refer to the type of an expression is with @code{typeof}.
536 The syntax of using of this keyword looks like @code{sizeof}, but the
537 construct acts semantically like a type name defined with @code{typedef}.
539 There are two ways of writing the argument to @code{typeof}: with an
540 expression or with a type. Here is an example with an expression:
547 This assumes that @code{x} is an array of pointers to functions;
548 the type described is that of the values of the functions.
550 Here is an example with a typename as the argument:
557 Here the type described is that of pointers to @code{int}.
559 If you are writing a header file that must work when included in ISO C
560 programs, write @code{__typeof__} instead of @code{typeof}.
561 @xref{Alternate Keywords}.
563 A @code{typeof}-construct can be used anywhere a typedef name could be
564 used. For example, you can use it in a declaration, in a cast, or inside
565 of @code{sizeof} or @code{typeof}.
567 @code{typeof} is often useful in conjunction with the
568 statements-within-expressions feature. Here is how the two together can
569 be used to define a safe ``maximum'' macro that operates on any
570 arithmetic type and evaluates each of its arguments exactly once:
574 (@{ typeof (a) _a = (a); \
575 typeof (b) _b = (b); \
576 _a > _b ? _a : _b; @})
579 @cindex underscores in variables in macros
580 @cindex @samp{_} in variables in macros
581 @cindex local variables in macros
582 @cindex variables, local, in macros
583 @cindex macros, local variables in
585 The reason for using names that start with underscores for the local
586 variables is to avoid conflicts with variable names that occur within the
587 expressions that are substituted for @code{a} and @code{b}. Eventually we
588 hope to design a new form of declaration syntax that allows you to declare
589 variables whose scopes start only after their initializers; this will be a
590 more reliable way to prevent such conflicts.
593 Some more examples of the use of @code{typeof}:
597 This declares @code{y} with the type of what @code{x} points to.
604 This declares @code{y} as an array of such values.
611 This declares @code{y} as an array of pointers to characters:
614 typeof (typeof (char *)[4]) y;
618 It is equivalent to the following traditional C declaration:
624 To see the meaning of the declaration using @code{typeof}, and why it
625 might be a useful way to write, rewrite it with these macros:
628 #define pointer(T) typeof(T *)
629 #define array(T, N) typeof(T [N])
633 Now the declaration can be rewritten this way:
636 array (pointer (char), 4) y;
640 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
641 pointers to @code{char}.
644 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
645 a more limited extension which permitted one to write
648 typedef @var{T} = @var{expr};
652 with the effect of declaring @var{T} to have the type of the expression
653 @var{expr}. This extension does not work with GCC 3 (versions between
654 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
655 relies on it should be rewritten to use @code{typeof}:
658 typedef typeof(@var{expr}) @var{T};
662 This will work with all versions of GCC@.
665 @section Conditionals with Omitted Operands
666 @cindex conditional expressions, extensions
667 @cindex omitted middle-operands
668 @cindex middle-operands, omitted
669 @cindex extensions, @code{?:}
670 @cindex @code{?:} extensions
672 The middle operand in a conditional expression may be omitted. Then
673 if the first operand is nonzero, its value is the value of the conditional
676 Therefore, the expression
683 has the value of @code{x} if that is nonzero; otherwise, the value of
686 This example is perfectly equivalent to
692 @cindex side effect in ?:
693 @cindex ?: side effect
695 In this simple case, the ability to omit the middle operand is not
696 especially useful. When it becomes useful is when the first operand does,
697 or may (if it is a macro argument), contain a side effect. Then repeating
698 the operand in the middle would perform the side effect twice. Omitting
699 the middle operand uses the value already computed without the undesirable
700 effects of recomputing it.
703 @section Double-Word Integers
704 @cindex @code{long long} data types
705 @cindex double-word arithmetic
706 @cindex multiprecision arithmetic
707 @cindex @code{LL} integer suffix
708 @cindex @code{ULL} integer suffix
710 ISO C99 supports data types for integers that are at least 64 bits wide,
711 and as an extension GCC supports them in C89 mode and in C++.
712 Simply write @code{long long int} for a signed integer, or
713 @code{unsigned long long int} for an unsigned integer. To make an
714 integer constant of type @code{long long int}, add the suffix @samp{LL}
715 to the integer. To make an integer constant of type @code{unsigned long
716 long int}, add the suffix @samp{ULL} to the integer.
718 You can use these types in arithmetic like any other integer types.
719 Addition, subtraction, and bitwise boolean operations on these types
720 are open-coded on all types of machines. Multiplication is open-coded
721 if the machine supports fullword-to-doubleword a widening multiply
722 instruction. Division and shifts are open-coded only on machines that
723 provide special support. The operations that are not open-coded use
724 special library routines that come with GCC@.
726 There may be pitfalls when you use @code{long long} types for function
727 arguments, unless you declare function prototypes. If a function
728 expects type @code{int} for its argument, and you pass a value of type
729 @code{long long int}, confusion will result because the caller and the
730 subroutine will disagree about the number of bytes for the argument.
731 Likewise, if the function expects @code{long long int} and you pass
732 @code{int}. The best way to avoid such problems is to use prototypes.
735 @section Complex Numbers
736 @cindex complex numbers
737 @cindex @code{_Complex} keyword
738 @cindex @code{__complex__} keyword
740 ISO C99 supports complex floating data types, and as an extension GCC
741 supports them in C89 mode and in C++, and supports complex integer data
742 types which are not part of ISO C99. You can declare complex types
743 using the keyword @code{_Complex}. As an extension, the older GNU
744 keyword @code{__complex__} is also supported.
746 For example, @samp{_Complex double x;} declares @code{x} as a
747 variable whose real part and imaginary part are both of type
748 @code{double}. @samp{_Complex short int y;} declares @code{y} to
749 have real and imaginary parts of type @code{short int}; this is not
750 likely to be useful, but it shows that the set of complex types is
753 To write a constant with a complex data type, use the suffix @samp{i} or
754 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
755 has type @code{_Complex float} and @code{3i} has type
756 @code{_Complex int}. Such a constant always has a pure imaginary
757 value, but you can form any complex value you like by adding one to a
758 real constant. This is a GNU extension; if you have an ISO C99
759 conforming C library (such as GNU libc), and want to construct complex
760 constants of floating type, you should include @code{<complex.h>} and
761 use the macros @code{I} or @code{_Complex_I} instead.
763 @cindex @code{__real__} keyword
764 @cindex @code{__imag__} keyword
765 To extract the real part of a complex-valued expression @var{exp}, write
766 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
767 extract the imaginary part. This is a GNU extension; for values of
768 floating type, you should use the ISO C99 functions @code{crealf},
769 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
770 @code{cimagl}, declared in @code{<complex.h>} and also provided as
771 built-in functions by GCC@.
773 @cindex complex conjugation
774 The operator @samp{~} performs complex conjugation when used on a value
775 with a complex type. This is a GNU extension; for values of
776 floating type, you should use the ISO C99 functions @code{conjf},
777 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
778 provided as built-in functions by GCC@.
780 GCC can allocate complex automatic variables in a noncontiguous
781 fashion; it's even possible for the real part to be in a register while
782 the imaginary part is on the stack (or vice-versa). Only the DWARF2
783 debug info format can represent this, so use of DWARF2 is recommended.
784 If you are using the stabs debug info format, GCC describes a noncontiguous
785 complex variable as if it were two separate variables of noncomplex type.
786 If the variable's actual name is @code{foo}, the two fictitious
787 variables are named @code{foo$real} and @code{foo$imag}. You can
788 examine and set these two fictitious variables with your debugger.
794 ISO C99 supports floating-point numbers written not only in the usual
795 decimal notation, such as @code{1.55e1}, but also numbers such as
796 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
797 supports this in C89 mode (except in some cases when strictly
798 conforming) and in C++. In that format the
799 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
800 mandatory. The exponent is a decimal number that indicates the power of
801 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
808 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
809 is the same as @code{1.55e1}.
811 Unlike for floating-point numbers in the decimal notation the exponent
812 is always required in the hexadecimal notation. Otherwise the compiler
813 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
814 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
815 extension for floating-point constants of type @code{float}.
818 @section Arrays of Length Zero
819 @cindex arrays of length zero
820 @cindex zero-length arrays
821 @cindex length-zero arrays
822 @cindex flexible array members
824 Zero-length arrays are allowed in GNU C@. They are very useful as the
825 last element of a structure which is really a header for a variable-length
834 struct line *thisline = (struct line *)
835 malloc (sizeof (struct line) + this_length);
836 thisline->length = this_length;
839 In ISO C90, you would have to give @code{contents} a length of 1, which
840 means either you waste space or complicate the argument to @code{malloc}.
842 In ISO C99, you would use a @dfn{flexible array member}, which is
843 slightly different in syntax and semantics:
847 Flexible array members are written as @code{contents[]} without
851 Flexible array members have incomplete type, and so the @code{sizeof}
852 operator may not be applied. As a quirk of the original implementation
853 of zero-length arrays, @code{sizeof} evaluates to zero.
856 Flexible array members may only appear as the last member of a
857 @code{struct} that is otherwise non-empty.
860 A structure containing a flexible array member, or a union containing
861 such a structure (possibly recursively), may not be a member of a
862 structure or an element of an array. (However, these uses are
863 permitted by GCC as extensions.)
866 GCC versions before 3.0 allowed zero-length arrays to be statically
867 initialized, as if they were flexible arrays. In addition to those
868 cases that were useful, it also allowed initializations in situations
869 that would corrupt later data. Non-empty initialization of zero-length
870 arrays is now treated like any case where there are more initializer
871 elements than the array holds, in that a suitable warning about "excess
872 elements in array" is given, and the excess elements (all of them, in
873 this case) are ignored.
875 Instead GCC allows static initialization of flexible array members.
876 This is equivalent to defining a new structure containing the original
877 structure followed by an array of sufficient size to contain the data.
878 I.e.@: in the following, @code{f1} is constructed as if it were declared
884 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
887 struct f1 f1; int data[3];
888 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
892 The convenience of this extension is that @code{f1} has the desired
893 type, eliminating the need to consistently refer to @code{f2.f1}.
895 This has symmetry with normal static arrays, in that an array of
896 unknown size is also written with @code{[]}.
898 Of course, this extension only makes sense if the extra data comes at
899 the end of a top-level object, as otherwise we would be overwriting
900 data at subsequent offsets. To avoid undue complication and confusion
901 with initialization of deeply nested arrays, we simply disallow any
902 non-empty initialization except when the structure is the top-level
906 struct foo @{ int x; int y[]; @};
907 struct bar @{ struct foo z; @};
909 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
910 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
911 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
912 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
915 @node Empty Structures
916 @section Structures With No Members
917 @cindex empty structures
918 @cindex zero-size structures
920 GCC permits a C structure to have no members:
927 The structure will have size zero. In C++, empty structures are part
928 of the language. G++ treats empty structures as if they had a single
929 member of type @code{char}.
931 @node Variable Length
932 @section Arrays of Variable Length
933 @cindex variable-length arrays
934 @cindex arrays of variable length
937 Variable-length automatic arrays are allowed in ISO C99, and as an
938 extension GCC accepts them in C89 mode and in C++. (However, GCC's
939 implementation of variable-length arrays does not yet conform in detail
940 to the ISO C99 standard.) These arrays are
941 declared like any other automatic arrays, but with a length that is not
942 a constant expression. The storage is allocated at the point of
943 declaration and deallocated when the brace-level is exited. For
948 concat_fopen (char *s1, char *s2, char *mode)
950 char str[strlen (s1) + strlen (s2) + 1];
953 return fopen (str, mode);
957 @cindex scope of a variable length array
958 @cindex variable-length array scope
959 @cindex deallocating variable length arrays
960 Jumping or breaking out of the scope of the array name deallocates the
961 storage. Jumping into the scope is not allowed; you get an error
964 @cindex @code{alloca} vs variable-length arrays
965 You can use the function @code{alloca} to get an effect much like
966 variable-length arrays. The function @code{alloca} is available in
967 many other C implementations (but not in all). On the other hand,
968 variable-length arrays are more elegant.
970 There are other differences between these two methods. Space allocated
971 with @code{alloca} exists until the containing @emph{function} returns.
972 The space for a variable-length array is deallocated as soon as the array
973 name's scope ends. (If you use both variable-length arrays and
974 @code{alloca} in the same function, deallocation of a variable-length array
975 will also deallocate anything more recently allocated with @code{alloca}.)
977 You can also use variable-length arrays as arguments to functions:
981 tester (int len, char data[len][len])
987 The length of an array is computed once when the storage is allocated
988 and is remembered for the scope of the array in case you access it with
991 If you want to pass the array first and the length afterward, you can
992 use a forward declaration in the parameter list---another GNU extension.
996 tester (int len; char data[len][len], int len)
1002 @cindex parameter forward declaration
1003 The @samp{int len} before the semicolon is a @dfn{parameter forward
1004 declaration}, and it serves the purpose of making the name @code{len}
1005 known when the declaration of @code{data} is parsed.
1007 You can write any number of such parameter forward declarations in the
1008 parameter list. They can be separated by commas or semicolons, but the
1009 last one must end with a semicolon, which is followed by the ``real''
1010 parameter declarations. Each forward declaration must match a ``real''
1011 declaration in parameter name and data type. ISO C99 does not support
1012 parameter forward declarations.
1014 @node Variadic Macros
1015 @section Macros with a Variable Number of Arguments.
1016 @cindex variable number of arguments
1017 @cindex macro with variable arguments
1018 @cindex rest argument (in macro)
1019 @cindex variadic macros
1021 In the ISO C standard of 1999, a macro can be declared to accept a
1022 variable number of arguments much as a function can. The syntax for
1023 defining the macro is similar to that of a function. Here is an
1027 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1030 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1031 such a macro, it represents the zero or more tokens until the closing
1032 parenthesis that ends the invocation, including any commas. This set of
1033 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1034 wherever it appears. See the CPP manual for more information.
1036 GCC has long supported variadic macros, and used a different syntax that
1037 allowed you to give a name to the variable arguments just like any other
1038 argument. Here is an example:
1041 #define debug(format, args...) fprintf (stderr, format, args)
1044 This is in all ways equivalent to the ISO C example above, but arguably
1045 more readable and descriptive.
1047 GNU CPP has two further variadic macro extensions, and permits them to
1048 be used with either of the above forms of macro definition.
1050 In standard C, you are not allowed to leave the variable argument out
1051 entirely; but you are allowed to pass an empty argument. For example,
1052 this invocation is invalid in ISO C, because there is no comma after
1059 GNU CPP permits you to completely omit the variable arguments in this
1060 way. In the above examples, the compiler would complain, though since
1061 the expansion of the macro still has the extra comma after the format
1064 To help solve this problem, CPP behaves specially for variable arguments
1065 used with the token paste operator, @samp{##}. If instead you write
1068 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1071 and if the variable arguments are omitted or empty, the @samp{##}
1072 operator causes the preprocessor to remove the comma before it. If you
1073 do provide some variable arguments in your macro invocation, GNU CPP
1074 does not complain about the paste operation and instead places the
1075 variable arguments after the comma. Just like any other pasted macro
1076 argument, these arguments are not macro expanded.
1078 @node Escaped Newlines
1079 @section Slightly Looser Rules for Escaped Newlines
1080 @cindex escaped newlines
1081 @cindex newlines (escaped)
1083 Recently, the preprocessor has relaxed its treatment of escaped
1084 newlines. Previously, the newline had to immediately follow a
1085 backslash. The current implementation allows whitespace in the form
1086 of spaces, horizontal and vertical tabs, and form feeds between the
1087 backslash and the subsequent newline. The preprocessor issues a
1088 warning, but treats it as a valid escaped newline and combines the two
1089 lines to form a single logical line. This works within comments and
1090 tokens, as well as between tokens. Comments are @emph{not} treated as
1091 whitespace for the purposes of this relaxation, since they have not
1092 yet been replaced with spaces.
1095 @section Non-Lvalue Arrays May Have Subscripts
1096 @cindex subscripting
1097 @cindex arrays, non-lvalue
1099 @cindex subscripting and function values
1100 In ISO C99, arrays that are not lvalues still decay to pointers, and
1101 may be subscripted, although they may not be modified or used after
1102 the next sequence point and the unary @samp{&} operator may not be
1103 applied to them. As an extension, GCC allows such arrays to be
1104 subscripted in C89 mode, though otherwise they do not decay to
1105 pointers outside C99 mode. For example,
1106 this is valid in GNU C though not valid in C89:
1110 struct foo @{int a[4];@};
1116 return f().a[index];
1122 @section Arithmetic on @code{void}- and Function-Pointers
1123 @cindex void pointers, arithmetic
1124 @cindex void, size of pointer to
1125 @cindex function pointers, arithmetic
1126 @cindex function, size of pointer to
1128 In GNU C, addition and subtraction operations are supported on pointers to
1129 @code{void} and on pointers to functions. This is done by treating the
1130 size of a @code{void} or of a function as 1.
1132 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1133 and on function types, and returns 1.
1135 @opindex Wpointer-arith
1136 The option @option{-Wpointer-arith} requests a warning if these extensions
1140 @section Non-Constant Initializers
1141 @cindex initializers, non-constant
1142 @cindex non-constant initializers
1144 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1145 automatic variable are not required to be constant expressions in GNU C@.
1146 Here is an example of an initializer with run-time varying elements:
1149 foo (float f, float g)
1151 float beat_freqs[2] = @{ f-g, f+g @};
1156 @node Compound Literals
1157 @section Compound Literals
1158 @cindex constructor expressions
1159 @cindex initializations in expressions
1160 @cindex structures, constructor expression
1161 @cindex expressions, constructor
1162 @cindex compound literals
1163 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1165 ISO C99 supports compound literals. A compound literal looks like
1166 a cast containing an initializer. Its value is an object of the
1167 type specified in the cast, containing the elements specified in
1168 the initializer; it is an lvalue. As an extension, GCC supports
1169 compound literals in C89 mode and in C++.
1171 Usually, the specified type is a structure. Assume that
1172 @code{struct foo} and @code{structure} are declared as shown:
1175 struct foo @{int a; char b[2];@} structure;
1179 Here is an example of constructing a @code{struct foo} with a compound literal:
1182 structure = ((struct foo) @{x + y, 'a', 0@});
1186 This is equivalent to writing the following:
1190 struct foo temp = @{x + y, 'a', 0@};
1195 You can also construct an array. If all the elements of the compound literal
1196 are (made up of) simple constant expressions, suitable for use in
1197 initializers of objects of static storage duration, then the compound
1198 literal can be coerced to a pointer to its first element and used in
1199 such an initializer, as shown here:
1202 char **foo = (char *[]) @{ "x", "y", "z" @};
1205 Compound literals for scalar types and union types are is
1206 also allowed, but then the compound literal is equivalent
1209 As a GNU extension, GCC allows initialization of objects with static storage
1210 duration by compound literals (which is not possible in ISO C99, because
1211 the initializer is not a constant).
1212 It is handled as if the object was initialized only with the bracket
1213 enclosed list if compound literal's and object types match.
1214 The initializer list of the compound literal must be constant.
1215 If the object being initialized has array type of unknown size, the size is
1216 determined by compound literal size.
1219 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1220 static int y[] = (int []) @{1, 2, 3@};
1221 static int z[] = (int [3]) @{1@};
1225 The above lines are equivalent to the following:
1227 static struct foo x = @{1, 'a', 'b'@};
1228 static int y[] = @{1, 2, 3@};
1229 static int z[] = @{1, 0, 0@};
1232 @node Designated Inits
1233 @section Designated Initializers
1234 @cindex initializers with labeled elements
1235 @cindex labeled elements in initializers
1236 @cindex case labels in initializers
1237 @cindex designated initializers
1239 Standard C89 requires the elements of an initializer to appear in a fixed
1240 order, the same as the order of the elements in the array or structure
1243 In ISO C99 you can give the elements in any order, specifying the array
1244 indices or structure field names they apply to, and GNU C allows this as
1245 an extension in C89 mode as well. This extension is not
1246 implemented in GNU C++.
1248 To specify an array index, write
1249 @samp{[@var{index}] =} before the element value. For example,
1252 int a[6] = @{ [4] = 29, [2] = 15 @};
1259 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1263 The index values must be constant expressions, even if the array being
1264 initialized is automatic.
1266 An alternative syntax for this which has been obsolete since GCC 2.5 but
1267 GCC still accepts is to write @samp{[@var{index}]} before the element
1268 value, with no @samp{=}.
1270 To initialize a range of elements to the same value, write
1271 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1272 extension. For example,
1275 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1279 If the value in it has side-effects, the side-effects will happen only once,
1280 not for each initialized field by the range initializer.
1283 Note that the length of the array is the highest value specified
1286 In a structure initializer, specify the name of a field to initialize
1287 with @samp{.@var{fieldname} =} before the element value. For example,
1288 given the following structure,
1291 struct point @{ int x, y; @};
1295 the following initialization
1298 struct point p = @{ .y = yvalue, .x = xvalue @};
1305 struct point p = @{ xvalue, yvalue @};
1308 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1309 @samp{@var{fieldname}:}, as shown here:
1312 struct point p = @{ y: yvalue, x: xvalue @};
1316 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1317 @dfn{designator}. You can also use a designator (or the obsolete colon
1318 syntax) when initializing a union, to specify which element of the union
1319 should be used. For example,
1322 union foo @{ int i; double d; @};
1324 union foo f = @{ .d = 4 @};
1328 will convert 4 to a @code{double} to store it in the union using
1329 the second element. By contrast, casting 4 to type @code{union foo}
1330 would store it into the union as the integer @code{i}, since it is
1331 an integer. (@xref{Cast to Union}.)
1333 You can combine this technique of naming elements with ordinary C
1334 initialization of successive elements. Each initializer element that
1335 does not have a designator applies to the next consecutive element of the
1336 array or structure. For example,
1339 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1346 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1349 Labeling the elements of an array initializer is especially useful
1350 when the indices are characters or belong to an @code{enum} type.
1355 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1356 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1359 @cindex designator lists
1360 You can also write a series of @samp{.@var{fieldname}} and
1361 @samp{[@var{index}]} designators before an @samp{=} to specify a
1362 nested subobject to initialize; the list is taken relative to the
1363 subobject corresponding to the closest surrounding brace pair. For
1364 example, with the @samp{struct point} declaration above:
1367 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1371 If the same field is initialized multiple times, it will have value from
1372 the last initialization. If any such overridden initialization has
1373 side-effect, it is unspecified whether the side-effect happens or not.
1374 Currently, GCC will discard them and issue a warning.
1377 @section Case Ranges
1379 @cindex ranges in case statements
1381 You can specify a range of consecutive values in a single @code{case} label,
1385 case @var{low} ... @var{high}:
1389 This has the same effect as the proper number of individual @code{case}
1390 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1392 This feature is especially useful for ranges of ASCII character codes:
1398 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1399 it may be parsed wrong when you use it with integer values. For example,
1414 @section Cast to a Union Type
1415 @cindex cast to a union
1416 @cindex union, casting to a
1418 A cast to union type is similar to other casts, except that the type
1419 specified is a union type. You can specify the type either with
1420 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1421 a constructor though, not a cast, and hence does not yield an lvalue like
1422 normal casts. (@xref{Compound Literals}.)
1424 The types that may be cast to the union type are those of the members
1425 of the union. Thus, given the following union and variables:
1428 union foo @{ int i; double d; @};
1434 both @code{x} and @code{y} can be cast to type @code{union foo}.
1436 Using the cast as the right-hand side of an assignment to a variable of
1437 union type is equivalent to storing in a member of the union:
1442 u = (union foo) x @equiv{} u.i = x
1443 u = (union foo) y @equiv{} u.d = y
1446 You can also use the union cast as a function argument:
1449 void hack (union foo);
1451 hack ((union foo) x);
1454 @node Mixed Declarations
1455 @section Mixed Declarations and Code
1456 @cindex mixed declarations and code
1457 @cindex declarations, mixed with code
1458 @cindex code, mixed with declarations
1460 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1461 within compound statements. As an extension, GCC also allows this in
1462 C89 mode. For example, you could do:
1471 Each identifier is visible from where it is declared until the end of
1472 the enclosing block.
1474 @node Function Attributes
1475 @section Declaring Attributes of Functions
1476 @cindex function attributes
1477 @cindex declaring attributes of functions
1478 @cindex functions that never return
1479 @cindex functions that have no side effects
1480 @cindex functions in arbitrary sections
1481 @cindex functions that behave like malloc
1482 @cindex @code{volatile} applied to function
1483 @cindex @code{const} applied to function
1484 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1485 @cindex functions with non-null pointer arguments
1486 @cindex functions that are passed arguments in registers on the 386
1487 @cindex functions that pop the argument stack on the 386
1488 @cindex functions that do not pop the argument stack on the 386
1490 In GNU C, you declare certain things about functions called in your program
1491 which help the compiler optimize function calls and check your code more
1494 The keyword @code{__attribute__} allows you to specify special
1495 attributes when making a declaration. This keyword is followed by an
1496 attribute specification inside double parentheses. The following
1497 attributes are currently defined for functions on all targets:
1498 @code{noreturn}, @code{noinline}, @code{always_inline},
1499 @code{pure}, @code{const}, @code{nothrow},
1500 @code{format}, @code{format_arg}, @code{no_instrument_function},
1501 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1502 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1503 @code{alias}, @code{warn_unused_result} and @code{nonnull}. Several other
1504 attributes are defined for functions on particular target systems. Other
1505 attributes, including @code{section} are supported for variables declarations
1506 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1508 You may also specify attributes with @samp{__} preceding and following
1509 each keyword. This allows you to use them in header files without
1510 being concerned about a possible macro of the same name. For example,
1511 you may use @code{__noreturn__} instead of @code{noreturn}.
1513 @xref{Attribute Syntax}, for details of the exact syntax for using
1517 @c Keep this table alphabetized by attribute name. Treat _ as space.
1519 @item alias ("@var{target}")
1520 @cindex @code{alias} attribute
1521 The @code{alias} attribute causes the declaration to be emitted as an
1522 alias for another symbol, which must be specified. For instance,
1525 void __f () @{ /* @r{Do something.} */; @}
1526 void f () __attribute__ ((weak, alias ("__f")));
1529 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
1530 mangled name for the target must be used.
1532 Not all target machines support this attribute.
1535 @cindex @code{always_inline} function attribute
1536 Generally, functions are not inlined unless optimization is specified.
1537 For functions declared inline, this attribute inlines the function even
1538 if no optimization level was specified.
1541 @cindex functions that do pop the argument stack on the 386
1543 On the Intel 386, the @code{cdecl} attribute causes the compiler to
1544 assume that the calling function will pop off the stack space used to
1545 pass arguments. This is
1546 useful to override the effects of the @option{-mrtd} switch.
1549 @cindex @code{const} function attribute
1550 Many functions do not examine any values except their arguments, and
1551 have no effects except the return value. Basically this is just slightly
1552 more strict class than the @code{pure} attribute above, since function is not
1553 allowed to read global memory.
1555 @cindex pointer arguments
1556 Note that a function that has pointer arguments and examines the data
1557 pointed to must @emph{not} be declared @code{const}. Likewise, a
1558 function that calls a non-@code{const} function usually must not be
1559 @code{const}. It does not make sense for a @code{const} function to
1562 The attribute @code{const} is not implemented in GCC versions earlier
1563 than 2.5. An alternative way to declare that a function has no side
1564 effects, which works in the current version and in some older versions,
1568 typedef int intfn ();
1570 extern const intfn square;
1573 This approach does not work in GNU C++ from 2.6.0 on, since the language
1574 specifies that the @samp{const} must be attached to the return value.
1578 @cindex @code{constructor} function attribute
1579 @cindex @code{destructor} function attribute
1580 The @code{constructor} attribute causes the function to be called
1581 automatically before execution enters @code{main ()}. Similarly, the
1582 @code{destructor} attribute causes the function to be called
1583 automatically after @code{main ()} has completed or @code{exit ()} has
1584 been called. Functions with these attributes are useful for
1585 initializing data that will be used implicitly during the execution of
1588 These attributes are not currently implemented for Objective-C@.
1591 @cindex @code{deprecated} attribute.
1592 The @code{deprecated} attribute results in a warning if the function
1593 is used anywhere in the source file. This is useful when identifying
1594 functions that are expected to be removed in a future version of a
1595 program. The warning also includes the location of the declaration
1596 of the deprecated function, to enable users to easily find further
1597 information about why the function is deprecated, or what they should
1598 do instead. Note that the warnings only occurs for uses:
1601 int old_fn () __attribute__ ((deprecated));
1603 int (*fn_ptr)() = old_fn;
1606 results in a warning on line 3 but not line 2.
1608 The @code{deprecated} attribute can also be used for variables and
1609 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
1612 @cindex @code{__declspec(dllexport)}
1613 On Microsoft Windows targets and Symbian targets the @code{dllexport}
1614 attribute causes the compiler to provide a global pointer to a pointer
1615 in a dll, so that it can be referenced with the @code{dllimport}
1616 attribute. The pointer name is formed by combining @code{_imp__} and
1617 the function or variable name.
1619 Currently, the @code{dllexport}attribute is ignored for inlined
1620 functions, but export can be forced by using the
1621 @option{-fkeep-inline-functions} flag. The attribute is also ignored for
1624 When applied to C++ classes. the attribute marks defined non-inlined
1625 member functions and static data members as exports. Static consts
1626 initialized in-class are not marked unless they are also defined
1629 On cygwin, mingw, arm-pe and sh-symbianelf targets,
1630 @code{__declspec(dllexport)} is recognized as a synonym for
1631 @code{__attribute__ ((dllexport))} for compatibility with other
1632 Microsoft Windows and Symbian compilers.
1634 For Microsoft Windows targets there are alternative methods for
1635 including the symbol in the dll's export table such as using a
1636 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
1637 the @option{--export-all} linker flag.
1640 @cindex @code{__declspec(dllimport)}
1641 On Microsoft Windows and Symbian targets, the @code{dllimport}
1642 attribute causes the compiler to reference a function or variable via
1643 a global pointer to a pointer that is set up by the Microsoft Windows
1644 dll library. The pointer name is formed by combining @code{_imp__} and
1645 the function or variable name. The attribute implies @code{extern}
1648 Currently, the attribute is ignored for inlined functions. If the
1649 attribute is applied to a symbol @emph{definition}, an error is reported.
1650 If a symbol previously declared @code{dllimport} is later defined, the
1651 attribute is ignored in subsequent references, and a warning is emitted.
1652 The attribute is also overridden by a subsequent declaration as
1655 When applied to C++ classes, the attribute marks non-inlined
1656 member functions and static data members as imports. However, the
1657 attribute is ignored for virtual methods to allow creation of vtables
1660 For Symbian targets the @code{dllimport} attribute also has another
1661 affect - it can cause the vtable and run-time type information for a
1662 class to be exported. This happens when the class has a dllimport'ed
1663 constructor or a non-inline, non-pure virtual function and, for either
1664 of those two conditions, the class also has a inline constructor or
1665 destructor and has a key function that is defined in the current
1668 On cygwin, mingw, arm-pe sh-symbianelf targets,
1669 @code{__declspec(dllimport)} is recognized as a synonym for
1670 @code{__attribute__ ((dllimport))} for compatibility with other
1671 Microsoft Windows and Symbian compilers.
1673 For Microsoft Windows based targets the use of the @code{dllimport}
1674 attribute on functions is not necessary, but provides a small
1675 performance benefit by eliminating a thunk in the dll. The use of the
1676 @code{dllimport} attribute on imported variables was required on older
1677 versions of GNU ld, but can now be avoided by passing the
1678 @option{--enable-auto-import} switch to ld. As with functions, using
1679 the attribute for a variable eliminates a thunk in the dll.
1681 One drawback to using this attribute is that a pointer to a function or
1682 variable marked as dllimport cannot be used as a constant address. The
1683 attribute can be disabled for functions by setting the
1684 @option{-mnop-fun-dllimport} flag.
1687 @cindex eight bit data on the H8/300, H8/300H, and H8S
1688 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1689 variable should be placed into the eight bit data section.
1690 The compiler will generate more efficient code for certain operations
1691 on data in the eight bit data area. Note the eight bit data area is limited to
1694 You must use GAS and GLD from GNU binutils version 2.7 or later for
1695 this attribute to work correctly.
1698 @cindex functions which handle memory bank switching
1699 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
1700 use a calling convention that takes care of switching memory banks when
1701 entering and leaving a function. This calling convention is also the
1702 default when using the @option{-mlong-calls} option.
1704 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
1705 to call and return from a function.
1707 On 68HC11 the compiler will generate a sequence of instructions
1708 to invoke a board-specific routine to switch the memory bank and call the
1709 real function. The board-specific routine simulates a @code{call}.
1710 At the end of a function, it will jump to a board-specific routine
1711 instead of using @code{rts}. The board-specific return routine simulates
1715 @cindex functions that pop the argument stack on the 386
1716 On the Intel 386, the @code{fastcall} attribute causes the compiler to
1717 pass the first two arguments in the registers ECX and EDX. Subsequent
1718 arguments are passed on the stack. The called function will pop the
1719 arguments off the stack. If the number of arguments is variable all
1720 arguments are pushed on the stack.
1722 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
1723 @cindex @code{format} function attribute
1725 The @code{format} attribute specifies that a function takes @code{printf},
1726 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
1727 should be type-checked against a format string. For example, the
1732 my_printf (void *my_object, const char *my_format, ...)
1733 __attribute__ ((format (printf, 2, 3)));
1737 causes the compiler to check the arguments in calls to @code{my_printf}
1738 for consistency with the @code{printf} style format string argument
1741 The parameter @var{archetype} determines how the format string is
1742 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
1743 or @code{strfmon}. (You can also use @code{__printf__},
1744 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
1745 parameter @var{string-index} specifies which argument is the format
1746 string argument (starting from 1), while @var{first-to-check} is the
1747 number of the first argument to check against the format string. For
1748 functions where the arguments are not available to be checked (such as
1749 @code{vprintf}), specify the third parameter as zero. In this case the
1750 compiler only checks the format string for consistency. For
1751 @code{strftime} formats, the third parameter is required to be zero.
1752 Since non-static C++ methods have an implicit @code{this} argument, the
1753 arguments of such methods should be counted from two, not one, when
1754 giving values for @var{string-index} and @var{first-to-check}.
1756 In the example above, the format string (@code{my_format}) is the second
1757 argument of the function @code{my_print}, and the arguments to check
1758 start with the third argument, so the correct parameters for the format
1759 attribute are 2 and 3.
1761 @opindex ffreestanding
1762 The @code{format} attribute allows you to identify your own functions
1763 which take format strings as arguments, so that GCC can check the
1764 calls to these functions for errors. The compiler always (unless
1765 @option{-ffreestanding} is used) checks formats
1766 for the standard library functions @code{printf}, @code{fprintf},
1767 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
1768 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
1769 warnings are requested (using @option{-Wformat}), so there is no need to
1770 modify the header file @file{stdio.h}. In C99 mode, the functions
1771 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
1772 @code{vsscanf} are also checked. Except in strictly conforming C
1773 standard modes, the X/Open function @code{strfmon} is also checked as
1774 are @code{printf_unlocked} and @code{fprintf_unlocked}.
1775 @xref{C Dialect Options,,Options Controlling C Dialect}.
1777 The target may provide additional types of format checks.
1778 @xref{Target Format Checks,,Format Checks Specific to Particular
1781 @item format_arg (@var{string-index})
1782 @cindex @code{format_arg} function attribute
1783 @opindex Wformat-nonliteral
1784 The @code{format_arg} attribute specifies that a function takes a format
1785 string for a @code{printf}, @code{scanf}, @code{strftime} or
1786 @code{strfmon} style function and modifies it (for example, to translate
1787 it into another language), so the result can be passed to a
1788 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
1789 function (with the remaining arguments to the format function the same
1790 as they would have been for the unmodified string). For example, the
1795 my_dgettext (char *my_domain, const char *my_format)
1796 __attribute__ ((format_arg (2)));
1800 causes the compiler to check the arguments in calls to a @code{printf},
1801 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
1802 format string argument is a call to the @code{my_dgettext} function, for
1803 consistency with the format string argument @code{my_format}. If the
1804 @code{format_arg} attribute had not been specified, all the compiler
1805 could tell in such calls to format functions would be that the format
1806 string argument is not constant; this would generate a warning when
1807 @option{-Wformat-nonliteral} is used, but the calls could not be checked
1808 without the attribute.
1810 The parameter @var{string-index} specifies which argument is the format
1811 string argument (starting from one). Since non-static C++ methods have
1812 an implicit @code{this} argument, the arguments of such methods should
1813 be counted from two.
1815 The @code{format-arg} attribute allows you to identify your own
1816 functions which modify format strings, so that GCC can check the
1817 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
1818 type function whose operands are a call to one of your own function.
1819 The compiler always treats @code{gettext}, @code{dgettext}, and
1820 @code{dcgettext} in this manner except when strict ISO C support is
1821 requested by @option{-ansi} or an appropriate @option{-std} option, or
1822 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
1823 Controlling C Dialect}.
1825 @item function_vector
1826 @cindex calling functions through the function vector on the H8/300 processors
1827 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
1828 function should be called through the function vector. Calling a
1829 function through the function vector will reduce code size, however;
1830 the function vector has a limited size (maximum 128 entries on the H8/300
1831 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
1833 You must use GAS and GLD from GNU binutils version 2.7 or later for
1834 this attribute to work correctly.
1837 @cindex interrupt handler functions
1838 Use this attribute on the ARM, AVR, C4x, M32R/D and Xstormy16 ports to indicate
1839 that the specified function is an interrupt handler. The compiler will
1840 generate function entry and exit sequences suitable for use in an
1841 interrupt handler when this attribute is present.
1843 Note, interrupt handlers for the m68k, H8/300, H8/300H, H8S, and SH processors
1844 can be specified via the @code{interrupt_handler} attribute.
1846 Note, on the AVR, interrupts will be enabled inside the function.
1848 Note, for the ARM, you can specify the kind of interrupt to be handled by
1849 adding an optional parameter to the interrupt attribute like this:
1852 void f () __attribute__ ((interrupt ("IRQ")));
1855 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
1857 @item interrupt_handler
1858 @cindex interrupt handler functions on the m68k, H8/300 and SH processors
1859 Use this attribute on the m68k, H8/300, H8/300H, H8S, and SH to indicate that
1860 the specified function is an interrupt handler. The compiler will generate
1861 function entry and exit sequences suitable for use in an interrupt
1862 handler when this attribute is present.
1864 @item long_call/short_call
1865 @cindex indirect calls on ARM
1866 This attribute specifies how a particular function is called on
1867 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
1868 command line switch and @code{#pragma long_calls} settings. The
1869 @code{long_call} attribute causes the compiler to always call the
1870 function by first loading its address into a register and then using the
1871 contents of that register. The @code{short_call} attribute always places
1872 the offset to the function from the call site into the @samp{BL}
1873 instruction directly.
1875 @item longcall/shortcall
1876 @cindex functions called via pointer on the RS/6000 and PowerPC
1877 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
1878 compiler to always call this function via a pointer, just as it would if
1879 the @option{-mlongcall} option had been specified. The @code{shortcall}
1880 attribute causes the compiler not to do this. These attributes override
1881 both the @option{-mlongcall} switch and the @code{#pragma longcall}
1884 @xref{RS/6000 and PowerPC Options}, for more information on whether long
1885 calls are necessary.
1888 @cindex @code{malloc} attribute
1889 The @code{malloc} attribute is used to tell the compiler that a function
1890 may be treated as if any non-@code{NULL} pointer it returns cannot
1891 alias any other pointer valid when the function returns.
1892 This will often improve optimization.
1893 Standard functions with this property include @code{malloc} and
1894 @code{calloc}. @code{realloc}-like functions have this property as
1895 long as the old pointer is never referred to (including comparing it
1896 to the new pointer) after the function returns a non-@code{NULL}
1899 @item model (@var{model-name})
1900 @cindex function addressability on the M32R/D
1901 @cindex variable addressability on the IA-64
1903 On the M32R/D, use this attribute to set the addressability of an
1904 object, and of the code generated for a function. The identifier
1905 @var{model-name} is one of @code{small}, @code{medium}, or
1906 @code{large}, representing each of the code models.
1908 Small model objects live in the lower 16MB of memory (so that their
1909 addresses can be loaded with the @code{ld24} instruction), and are
1910 callable with the @code{bl} instruction.
1912 Medium model objects may live anywhere in the 32-bit address space (the
1913 compiler will generate @code{seth/add3} instructions to load their addresses),
1914 and are callable with the @code{bl} instruction.
1916 Large model objects may live anywhere in the 32-bit address space (the
1917 compiler will generate @code{seth/add3} instructions to load their addresses),
1918 and may not be reachable with the @code{bl} instruction (the compiler will
1919 generate the much slower @code{seth/add3/jl} instruction sequence).
1921 On IA-64, use this attribute to set the addressability of an object.
1922 At present, the only supported identifier for @var{model-name} is
1923 @code{small}, indicating addressability via ``small'' (22-bit)
1924 addresses (so that their addresses can be loaded with the @code{addl}
1925 instruction). Caveat: such addressing is by definition not position
1926 independent and hence this attribute must not be used for objects
1927 defined by shared libraries.
1930 @cindex function without a prologue/epilogue code
1931 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate that the
1932 specified function does not need prologue/epilogue sequences generated by
1933 the compiler. It is up to the programmer to provide these sequences.
1936 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
1937 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
1938 use the normal calling convention based on @code{jsr} and @code{rts}.
1939 This attribute can be used to cancel the effect of the @option{-mlong-calls}
1942 @item no_instrument_function
1943 @cindex @code{no_instrument_function} function attribute
1944 @opindex finstrument-functions
1945 If @option{-finstrument-functions} is given, profiling function calls will
1946 be generated at entry and exit of most user-compiled functions.
1947 Functions with this attribute will not be so instrumented.
1950 @cindex @code{noinline} function attribute
1951 This function attribute prevents a function from being considered for
1954 @item nonnull (@var{arg-index}, @dots{})
1955 @cindex @code{nonnull} function attribute
1956 The @code{nonnull} attribute specifies that some function parameters should
1957 be non-null pointers. For instance, the declaration:
1961 my_memcpy (void *dest, const void *src, size_t len)
1962 __attribute__((nonnull (1, 2)));
1966 causes the compiler to check that, in calls to @code{my_memcpy},
1967 arguments @var{dest} and @var{src} are non-null. If the compiler
1968 determines that a null pointer is passed in an argument slot marked
1969 as non-null, and the @option{-Wnonnull} option is enabled, a warning
1970 is issued. The compiler may also choose to make optimizations based
1971 on the knowledge that certain function arguments will not be null.
1973 If no argument index list is given to the @code{nonnull} attribute,
1974 all pointer arguments are marked as non-null. To illustrate, the
1975 following declaration is equivalent to the previous example:
1979 my_memcpy (void *dest, const void *src, size_t len)
1980 __attribute__((nonnull));
1984 @cindex @code{noreturn} function attribute
1985 A few standard library functions, such as @code{abort} and @code{exit},
1986 cannot return. GCC knows this automatically. Some programs define
1987 their own functions that never return. You can declare them
1988 @code{noreturn} to tell the compiler this fact. For example,
1992 void fatal () __attribute__ ((noreturn));
1995 fatal (/* @r{@dots{}} */)
1997 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2003 The @code{noreturn} keyword tells the compiler to assume that
2004 @code{fatal} cannot return. It can then optimize without regard to what
2005 would happen if @code{fatal} ever did return. This makes slightly
2006 better code. More importantly, it helps avoid spurious warnings of
2007 uninitialized variables.
2009 The @code{noreturn} keyword does not affect the exceptional path when that
2010 applies: a @code{noreturn}-marked function may still return to the caller
2011 by throwing an exception.
2013 Do not assume that registers saved by the calling function are
2014 restored before calling the @code{noreturn} function.
2016 It does not make sense for a @code{noreturn} function to have a return
2017 type other than @code{void}.
2019 The attribute @code{noreturn} is not implemented in GCC versions
2020 earlier than 2.5. An alternative way to declare that a function does
2021 not return, which works in the current version and in some older
2022 versions, is as follows:
2025 typedef void voidfn ();
2027 volatile voidfn fatal;
2031 @cindex @code{nothrow} function attribute
2032 The @code{nothrow} attribute is used to inform the compiler that a
2033 function cannot throw an exception. For example, most functions in
2034 the standard C library can be guaranteed not to throw an exception
2035 with the notable exceptions of @code{qsort} and @code{bsearch} that
2036 take function pointer arguments. The @code{nothrow} attribute is not
2037 implemented in GCC versions earlier than 3.2.
2040 @cindex @code{pure} function attribute
2041 Many functions have no effects except the return value and their
2042 return value depends only on the parameters and/or global variables.
2043 Such a function can be subject
2044 to common subexpression elimination and loop optimization just as an
2045 arithmetic operator would be. These functions should be declared
2046 with the attribute @code{pure}. For example,
2049 int square (int) __attribute__ ((pure));
2053 says that the hypothetical function @code{square} is safe to call
2054 fewer times than the program says.
2056 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2057 Interesting non-pure functions are functions with infinite loops or those
2058 depending on volatile memory or other system resource, that may change between
2059 two consecutive calls (such as @code{feof} in a multithreading environment).
2061 The attribute @code{pure} is not implemented in GCC versions earlier
2064 @item regparm (@var{number})
2065 @cindex @code{regparm} attribute
2066 @cindex functions that are passed arguments in registers on the 386
2067 On the Intel 386, the @code{regparm} attribute causes the compiler to
2068 pass up to @var{number} integer arguments in registers EAX,
2069 EDX, and ECX instead of on the stack. Functions that take a
2070 variable number of arguments will continue to be passed all of their
2071 arguments on the stack.
2073 Beware that on some ELF systems this attribute is unsuitable for
2074 global functions in shared libraries with lazy binding (which is the
2075 default). Lazy binding will send the first call via resolving code in
2076 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2077 per the standard calling conventions. Solaris 8 is affected by this.
2078 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2079 safe since the loaders there save all registers. (Lazy binding can be
2080 disabled with the linker or the loader if desired, to avoid the
2084 @cindex save all registers on the H8/300, H8/300H, and H8S
2085 Use this attribute on the H8/300, H8/300H, and H8S to indicate that
2086 all registers except the stack pointer should be saved in the prologue
2087 regardless of whether they are used or not.
2089 @item section ("@var{section-name}")
2090 @cindex @code{section} function attribute
2091 Normally, the compiler places the code it generates in the @code{text} section.
2092 Sometimes, however, you need additional sections, or you need certain
2093 particular functions to appear in special sections. The @code{section}
2094 attribute specifies that a function lives in a particular section.
2095 For example, the declaration:
2098 extern void foobar (void) __attribute__ ((section ("bar")));
2102 puts the function @code{foobar} in the @code{bar} section.
2104 Some file formats do not support arbitrary sections so the @code{section}
2105 attribute is not available on all platforms.
2106 If you need to map the entire contents of a module to a particular
2107 section, consider using the facilities of the linker instead.
2110 See long_call/short_call.
2113 See longcall/shortcall.
2116 @cindex signal handler functions on the AVR processors
2117 Use this attribute on the AVR to indicate that the specified
2118 function is a signal handler. The compiler will generate function
2119 entry and exit sequences suitable for use in a signal handler when this
2120 attribute is present. Interrupts will be disabled inside the function.
2123 Use this attribute on the SH to indicate an @code{interrupt_handler}
2124 function should switch to an alternate stack. It expects a string
2125 argument that names a global variable holding the address of the
2130 void f () __attribute__ ((interrupt_handler,
2131 sp_switch ("alt_stack")));
2135 @cindex functions that pop the argument stack on the 386
2136 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2137 assume that the called function will pop off the stack space used to
2138 pass arguments, unless it takes a variable number of arguments.
2141 @cindex tiny data section on the H8/300H and H8S
2142 Use this attribute on the H8/300H and H8S to indicate that the specified
2143 variable should be placed into the tiny data section.
2144 The compiler will generate more efficient code for loads and stores
2145 on data in the tiny data section. Note the tiny data area is limited to
2146 slightly under 32kbytes of data.
2149 Use this attribute on the SH for an @code{interrupt_handler} to return using
2150 @code{trapa} instead of @code{rte}. This attribute expects an integer
2151 argument specifying the trap number to be used.
2154 @cindex @code{unused} attribute.
2155 This attribute, attached to a function, means that the function is meant
2156 to be possibly unused. GCC will not produce a warning for this
2160 @cindex @code{used} attribute.
2161 This attribute, attached to a function, means that code must be emitted
2162 for the function even if it appears that the function is not referenced.
2163 This is useful, for example, when the function is referenced only in
2166 @item visibility ("@var{visibility_type}")
2167 @cindex @code{visibility} attribute
2168 The @code{visibility} attribute on ELF targets causes the declaration
2169 to be emitted with default, hidden, protected or internal visibility.
2172 void __attribute__ ((visibility ("protected")))
2173 f () @{ /* @r{Do something.} */; @}
2174 int i __attribute__ ((visibility ("hidden")));
2177 See the ELF gABI for complete details, but the short story is:
2180 @c keep this list of visibilies in alphabetical order.
2183 Default visibility is the normal case for ELF. This value is
2184 available for the visibility attribute to override other options
2185 that may change the assumed visibility of symbols.
2188 Hidden visibility indicates that the symbol will not be placed into
2189 the dynamic symbol table, so no other @dfn{module} (executable or
2190 shared library) can reference it directly.
2193 Internal visibility is like hidden visibility, but with additional
2194 processor specific semantics. Unless otherwise specified by the psABI,
2195 GCC defines internal visibility to mean that the function is @emph{never}
2196 called from another module. Note that hidden symbols, while they cannot
2197 be referenced directly by other modules, can be referenced indirectly via
2198 function pointers. By indicating that a symbol cannot be called from
2199 outside the module, GCC may for instance omit the load of a PIC register
2200 since it is known that the calling function loaded the correct value.
2203 Protected visibility indicates that the symbol will be placed in the
2204 dynamic symbol table, but that references within the defining module
2205 will bind to the local symbol. That is, the symbol cannot be overridden
2210 Not all ELF targets support this attribute.
2212 @item warn_unused_result
2213 @cindex @code{warn_unused_result} attribute
2214 The @code{warn_unused_result} attribute causes a warning to be emitted
2215 if a caller of the function with this attribute does not use its
2216 return value. This is useful for functions where not checking
2217 the result is either a security problem or always a bug, such as
2221 int fn () __attribute__ ((warn_unused_result));
2224 if (fn () < 0) return -1;
2230 results in warning on line 5.
2233 @cindex @code{weak} attribute
2234 The @code{weak} attribute causes the declaration to be emitted as a weak
2235 symbol rather than a global. This is primarily useful in defining
2236 library functions which can be overridden in user code, though it can
2237 also be used with non-function declarations. Weak symbols are supported
2238 for ELF targets, and also for a.out targets when using the GNU assembler
2243 You can specify multiple attributes in a declaration by separating them
2244 by commas within the double parentheses or by immediately following an
2245 attribute declaration with another attribute declaration.
2247 @cindex @code{#pragma}, reason for not using
2248 @cindex pragma, reason for not using
2249 Some people object to the @code{__attribute__} feature, suggesting that
2250 ISO C's @code{#pragma} should be used instead. At the time
2251 @code{__attribute__} was designed, there were two reasons for not doing
2256 It is impossible to generate @code{#pragma} commands from a macro.
2259 There is no telling what the same @code{#pragma} might mean in another
2263 These two reasons applied to almost any application that might have been
2264 proposed for @code{#pragma}. It was basically a mistake to use
2265 @code{#pragma} for @emph{anything}.
2267 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2268 to be generated from macros. In addition, a @code{#pragma GCC}
2269 namespace is now in use for GCC-specific pragmas. However, it has been
2270 found convenient to use @code{__attribute__} to achieve a natural
2271 attachment of attributes to their corresponding declarations, whereas
2272 @code{#pragma GCC} is of use for constructs that do not naturally form
2273 part of the grammar. @xref{Other Directives,,Miscellaneous
2274 Preprocessing Directives, cpp, The GNU C Preprocessor}.
2276 @node Attribute Syntax
2277 @section Attribute Syntax
2278 @cindex attribute syntax
2280 This section describes the syntax with which @code{__attribute__} may be
2281 used, and the constructs to which attribute specifiers bind, for the C
2282 language. Some details may vary for C++ and Objective-C@. Because of
2283 infelicities in the grammar for attributes, some forms described here
2284 may not be successfully parsed in all cases.
2286 There are some problems with the semantics of attributes in C++. For
2287 example, there are no manglings for attributes, although they may affect
2288 code generation, so problems may arise when attributed types are used in
2289 conjunction with templates or overloading. Similarly, @code{typeid}
2290 does not distinguish between types with different attributes. Support
2291 for attributes in C++ may be restricted in future to attributes on
2292 declarations only, but not on nested declarators.
2294 @xref{Function Attributes}, for details of the semantics of attributes
2295 applying to functions. @xref{Variable Attributes}, for details of the
2296 semantics of attributes applying to variables. @xref{Type Attributes},
2297 for details of the semantics of attributes applying to structure, union
2298 and enumerated types.
2300 An @dfn{attribute specifier} is of the form
2301 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2302 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2303 each attribute is one of the following:
2307 Empty. Empty attributes are ignored.
2310 A word (which may be an identifier such as @code{unused}, or a reserved
2311 word such as @code{const}).
2314 A word, followed by, in parentheses, parameters for the attribute.
2315 These parameters take one of the following forms:
2319 An identifier. For example, @code{mode} attributes use this form.
2322 An identifier followed by a comma and a non-empty comma-separated list
2323 of expressions. For example, @code{format} attributes use this form.
2326 A possibly empty comma-separated list of expressions. For example,
2327 @code{format_arg} attributes use this form with the list being a single
2328 integer constant expression, and @code{alias} attributes use this form
2329 with the list being a single string constant.
2333 An @dfn{attribute specifier list} is a sequence of one or more attribute
2334 specifiers, not separated by any other tokens.
2336 In GNU C, an attribute specifier list may appear after the colon following a
2337 label, other than a @code{case} or @code{default} label. The only
2338 attribute it makes sense to use after a label is @code{unused}. This
2339 feature is intended for code generated by programs which contains labels
2340 that may be unused but which is compiled with @option{-Wall}. It would
2341 not normally be appropriate to use in it human-written code, though it
2342 could be useful in cases where the code that jumps to the label is
2343 contained within an @code{#ifdef} conditional. GNU C++ does not permit
2344 such placement of attribute lists, as it is permissible for a
2345 declaration, which could begin with an attribute list, to be labelled in
2346 C++. Declarations cannot be labelled in C90 or C99, so the ambiguity
2347 does not arise there.
2349 An attribute specifier list may appear as part of a @code{struct},
2350 @code{union} or @code{enum} specifier. It may go either immediately
2351 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2352 the closing brace. It is ignored if the content of the structure, union
2353 or enumerated type is not defined in the specifier in which the
2354 attribute specifier list is used---that is, in usages such as
2355 @code{struct __attribute__((foo)) bar} with no following opening brace.
2356 Where attribute specifiers follow the closing brace, they are considered
2357 to relate to the structure, union or enumerated type defined, not to any
2358 enclosing declaration the type specifier appears in, and the type
2359 defined is not complete until after the attribute specifiers.
2360 @c Otherwise, there would be the following problems: a shift/reduce
2361 @c conflict between attributes binding the struct/union/enum and
2362 @c binding to the list of specifiers/qualifiers; and "aligned"
2363 @c attributes could use sizeof for the structure, but the size could be
2364 @c changed later by "packed" attributes.
2366 Otherwise, an attribute specifier appears as part of a declaration,
2367 counting declarations of unnamed parameters and type names, and relates
2368 to that declaration (which may be nested in another declaration, for
2369 example in the case of a parameter declaration), or to a particular declarator
2370 within a declaration. Where an
2371 attribute specifier is applied to a parameter declared as a function or
2372 an array, it should apply to the function or array rather than the
2373 pointer to which the parameter is implicitly converted, but this is not
2374 yet correctly implemented.
2376 Any list of specifiers and qualifiers at the start of a declaration may
2377 contain attribute specifiers, whether or not such a list may in that
2378 context contain storage class specifiers. (Some attributes, however,
2379 are essentially in the nature of storage class specifiers, and only make
2380 sense where storage class specifiers may be used; for example,
2381 @code{section}.) There is one necessary limitation to this syntax: the
2382 first old-style parameter declaration in a function definition cannot
2383 begin with an attribute specifier, because such an attribute applies to
2384 the function instead by syntax described below (which, however, is not
2385 yet implemented in this case). In some other cases, attribute
2386 specifiers are permitted by this grammar but not yet supported by the
2387 compiler. All attribute specifiers in this place relate to the
2388 declaration as a whole. In the obsolescent usage where a type of
2389 @code{int} is implied by the absence of type specifiers, such a list of
2390 specifiers and qualifiers may be an attribute specifier list with no
2391 other specifiers or qualifiers.
2393 An attribute specifier list may appear immediately before a declarator
2394 (other than the first) in a comma-separated list of declarators in a
2395 declaration of more than one identifier using a single list of
2396 specifiers and qualifiers. Such attribute specifiers apply
2397 only to the identifier before whose declarator they appear. For
2401 __attribute__((noreturn)) void d0 (void),
2402 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2407 the @code{noreturn} attribute applies to all the functions
2408 declared; the @code{format} attribute only applies to @code{d1}.
2410 An attribute specifier list may appear immediately before the comma,
2411 @code{=} or semicolon terminating the declaration of an identifier other
2412 than a function definition. At present, such attribute specifiers apply
2413 to the declared object or function, but in future they may attach to the
2414 outermost adjacent declarator. In simple cases there is no difference,
2415 but, for example, in
2418 void (****f)(void) __attribute__((noreturn));
2422 at present the @code{noreturn} attribute applies to @code{f}, which
2423 causes a warning since @code{f} is not a function, but in future it may
2424 apply to the function @code{****f}. The precise semantics of what
2425 attributes in such cases will apply to are not yet specified. Where an
2426 assembler name for an object or function is specified (@pxref{Asm
2427 Labels}), at present the attribute must follow the @code{asm}
2428 specification; in future, attributes before the @code{asm} specification
2429 may apply to the adjacent declarator, and those after it to the declared
2432 An attribute specifier list may, in future, be permitted to appear after
2433 the declarator in a function definition (before any old-style parameter
2434 declarations or the function body).
2436 Attribute specifiers may be mixed with type qualifiers appearing inside
2437 the @code{[]} of a parameter array declarator, in the C99 construct by
2438 which such qualifiers are applied to the pointer to which the array is
2439 implicitly converted. Such attribute specifiers apply to the pointer,
2440 not to the array, but at present this is not implemented and they are
2443 An attribute specifier list may appear at the start of a nested
2444 declarator. At present, there are some limitations in this usage: the
2445 attributes correctly apply to the declarator, but for most individual
2446 attributes the semantics this implies are not implemented.
2447 When attribute specifiers follow the @code{*} of a pointer
2448 declarator, they may be mixed with any type qualifiers present.
2449 The following describes the formal semantics of this syntax. It will make the
2450 most sense if you are familiar with the formal specification of
2451 declarators in the ISO C standard.
2453 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2454 D1}, where @code{T} contains declaration specifiers that specify a type
2455 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2456 contains an identifier @var{ident}. The type specified for @var{ident}
2457 for derived declarators whose type does not include an attribute
2458 specifier is as in the ISO C standard.
2460 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2461 and the declaration @code{T D} specifies the type
2462 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2463 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2464 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2466 If @code{D1} has the form @code{*
2467 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2468 declaration @code{T D} specifies the type
2469 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2470 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2471 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2477 void (__attribute__((noreturn)) ****f) (void);
2481 specifies the type ``pointer to pointer to pointer to pointer to
2482 non-returning function returning @code{void}''. As another example,
2485 char *__attribute__((aligned(8))) *f;
2489 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2490 Note again that this does not work with most attributes; for example,
2491 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2492 is not yet supported.
2494 For compatibility with existing code written for compiler versions that
2495 did not implement attributes on nested declarators, some laxity is
2496 allowed in the placing of attributes. If an attribute that only applies
2497 to types is applied to a declaration, it will be treated as applying to
2498 the type of that declaration. If an attribute that only applies to
2499 declarations is applied to the type of a declaration, it will be treated
2500 as applying to that declaration; and, for compatibility with code
2501 placing the attributes immediately before the identifier declared, such
2502 an attribute applied to a function return type will be treated as
2503 applying to the function type, and such an attribute applied to an array
2504 element type will be treated as applying to the array type. If an
2505 attribute that only applies to function types is applied to a
2506 pointer-to-function type, it will be treated as applying to the pointer
2507 target type; if such an attribute is applied to a function return type
2508 that is not a pointer-to-function type, it will be treated as applying
2509 to the function type.
2511 @node Function Prototypes
2512 @section Prototypes and Old-Style Function Definitions
2513 @cindex function prototype declarations
2514 @cindex old-style function definitions
2515 @cindex promotion of formal parameters
2517 GNU C extends ISO C to allow a function prototype to override a later
2518 old-style non-prototype definition. Consider the following example:
2521 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2528 /* @r{Prototype function declaration.} */
2529 int isroot P((uid_t));
2531 /* @r{Old-style function definition.} */
2533 isroot (x) /* ??? lossage here ??? */
2540 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2541 not allow this example, because subword arguments in old-style
2542 non-prototype definitions are promoted. Therefore in this example the
2543 function definition's argument is really an @code{int}, which does not
2544 match the prototype argument type of @code{short}.
2546 This restriction of ISO C makes it hard to write code that is portable
2547 to traditional C compilers, because the programmer does not know
2548 whether the @code{uid_t} type is @code{short}, @code{int}, or
2549 @code{long}. Therefore, in cases like these GNU C allows a prototype
2550 to override a later old-style definition. More precisely, in GNU C, a
2551 function prototype argument type overrides the argument type specified
2552 by a later old-style definition if the former type is the same as the
2553 latter type before promotion. Thus in GNU C the above example is
2554 equivalent to the following:
2567 GNU C++ does not support old-style function definitions, so this
2568 extension is irrelevant.
2571 @section C++ Style Comments
2573 @cindex C++ comments
2574 @cindex comments, C++ style
2576 In GNU C, you may use C++ style comments, which start with @samp{//} and
2577 continue until the end of the line. Many other C implementations allow
2578 such comments, and they are included in the 1999 C standard. However,
2579 C++ style comments are not recognized if you specify an @option{-std}
2580 option specifying a version of ISO C before C99, or @option{-ansi}
2581 (equivalent to @option{-std=c89}).
2584 @section Dollar Signs in Identifier Names
2586 @cindex dollar signs in identifier names
2587 @cindex identifier names, dollar signs in
2589 In GNU C, you may normally use dollar signs in identifier names.
2590 This is because many traditional C implementations allow such identifiers.
2591 However, dollar signs in identifiers are not supported on a few target
2592 machines, typically because the target assembler does not allow them.
2594 @node Character Escapes
2595 @section The Character @key{ESC} in Constants
2597 You can use the sequence @samp{\e} in a string or character constant to
2598 stand for the ASCII character @key{ESC}.
2601 @section Inquiring on Alignment of Types or Variables
2603 @cindex type alignment
2604 @cindex variable alignment
2606 The keyword @code{__alignof__} allows you to inquire about how an object
2607 is aligned, or the minimum alignment usually required by a type. Its
2608 syntax is just like @code{sizeof}.
2610 For example, if the target machine requires a @code{double} value to be
2611 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2612 This is true on many RISC machines. On more traditional machine
2613 designs, @code{__alignof__ (double)} is 4 or even 2.
2615 Some machines never actually require alignment; they allow reference to any
2616 data type even at an odd address. For these machines, @code{__alignof__}
2617 reports the @emph{recommended} alignment of a type.
2619 If the operand of @code{__alignof__} is an lvalue rather than a type,
2620 its value is the required alignment for its type, taking into account
2621 any minimum alignment specified with GCC's @code{__attribute__}
2622 extension (@pxref{Variable Attributes}). For example, after this
2626 struct foo @{ int x; char y; @} foo1;
2630 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
2631 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
2633 It is an error to ask for the alignment of an incomplete type.
2635 @node Variable Attributes
2636 @section Specifying Attributes of Variables
2637 @cindex attribute of variables
2638 @cindex variable attributes
2640 The keyword @code{__attribute__} allows you to specify special
2641 attributes of variables or structure fields. This keyword is followed
2642 by an attribute specification inside double parentheses. Some
2643 attributes are currently defined generically for variables.
2644 Other attributes are defined for variables on particular target
2645 systems. Other attributes are available for functions
2646 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
2647 Other front ends might define more attributes
2648 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2650 You may also specify attributes with @samp{__} preceding and following
2651 each keyword. This allows you to use them in header files without
2652 being concerned about a possible macro of the same name. For example,
2653 you may use @code{__aligned__} instead of @code{aligned}.
2655 @xref{Attribute Syntax}, for details of the exact syntax for using
2659 @cindex @code{aligned} attribute
2660 @item aligned (@var{alignment})
2661 This attribute specifies a minimum alignment for the variable or
2662 structure field, measured in bytes. For example, the declaration:
2665 int x __attribute__ ((aligned (16))) = 0;
2669 causes the compiler to allocate the global variable @code{x} on a
2670 16-byte boundary. On a 68040, this could be used in conjunction with
2671 an @code{asm} expression to access the @code{move16} instruction which
2672 requires 16-byte aligned operands.
2674 You can also specify the alignment of structure fields. For example, to
2675 create a double-word aligned @code{int} pair, you could write:
2678 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2682 This is an alternative to creating a union with a @code{double} member
2683 that forces the union to be double-word aligned.
2685 As in the preceding examples, you can explicitly specify the alignment
2686 (in bytes) that you wish the compiler to use for a given variable or
2687 structure field. Alternatively, you can leave out the alignment factor
2688 and just ask the compiler to align a variable or field to the maximum
2689 useful alignment for the target machine you are compiling for. For
2690 example, you could write:
2693 short array[3] __attribute__ ((aligned));
2696 Whenever you leave out the alignment factor in an @code{aligned} attribute
2697 specification, the compiler automatically sets the alignment for the declared
2698 variable or field to the largest alignment which is ever used for any data
2699 type on the target machine you are compiling for. Doing this can often make
2700 copy operations more efficient, because the compiler can use whatever
2701 instructions copy the biggest chunks of memory when performing copies to
2702 or from the variables or fields that you have aligned this way.
2704 The @code{aligned} attribute can only increase the alignment; but you
2705 can decrease it by specifying @code{packed} as well. See below.
2707 Note that the effectiveness of @code{aligned} attributes may be limited
2708 by inherent limitations in your linker. On many systems, the linker is
2709 only able to arrange for variables to be aligned up to a certain maximum
2710 alignment. (For some linkers, the maximum supported alignment may
2711 be very very small.) If your linker is only able to align variables
2712 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2713 in an @code{__attribute__} will still only provide you with 8 byte
2714 alignment. See your linker documentation for further information.
2716 @item cleanup (@var{cleanup_function})
2717 @cindex @code{cleanup} attribute
2718 The @code{cleanup} attribute runs a function when the variable goes
2719 out of scope. This attribute can only be applied to auto function
2720 scope variables; it may not be applied to parameters or variables
2721 with static storage duration. The function must take one parameter,
2722 a pointer to a type compatible with the variable. The return value
2723 of the function (if any) is ignored.
2725 If @option{-fexceptions} is enabled, then @var{cleanup_function}
2726 will be run during the stack unwinding that happens during the
2727 processing of the exception. Note that the @code{cleanup} attribute
2728 does not allow the exception to be caught, only to perform an action.
2729 It is undefined what happens if @var{cleanup_function} does not
2734 @cindex @code{common} attribute
2735 @cindex @code{nocommon} attribute
2738 The @code{common} attribute requests GCC to place a variable in
2739 ``common'' storage. The @code{nocommon} attribute requests the
2740 opposite -- to allocate space for it directly.
2742 These attributes override the default chosen by the
2743 @option{-fno-common} and @option{-fcommon} flags respectively.
2746 @cindex @code{deprecated} attribute
2747 The @code{deprecated} attribute results in a warning if the variable
2748 is used anywhere in the source file. This is useful when identifying
2749 variables that are expected to be removed in a future version of a
2750 program. The warning also includes the location of the declaration
2751 of the deprecated variable, to enable users to easily find further
2752 information about why the variable is deprecated, or what they should
2753 do instead. Note that the warning only occurs for uses:
2756 extern int old_var __attribute__ ((deprecated));
2758 int new_fn () @{ return old_var; @}
2761 results in a warning on line 3 but not line 2.
2763 The @code{deprecated} attribute can also be used for functions and
2764 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
2766 @item mode (@var{mode})
2767 @cindex @code{mode} attribute
2768 This attribute specifies the data type for the declaration---whichever
2769 type corresponds to the mode @var{mode}. This in effect lets you
2770 request an integer or floating point type according to its width.
2772 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2773 indicate the mode corresponding to a one-byte integer, @samp{word} or
2774 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2775 or @samp{__pointer__} for the mode used to represent pointers.
2778 @cindex @code{packed} attribute
2779 The @code{packed} attribute specifies that a variable or structure field
2780 should have the smallest possible alignment---one byte for a variable,
2781 and one bit for a field, unless you specify a larger value with the
2782 @code{aligned} attribute.
2784 Here is a structure in which the field @code{x} is packed, so that it
2785 immediately follows @code{a}:
2791 int x[2] __attribute__ ((packed));
2795 @item section ("@var{section-name}")
2796 @cindex @code{section} variable attribute
2797 Normally, the compiler places the objects it generates in sections like
2798 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
2799 or you need certain particular variables to appear in special sections,
2800 for example to map to special hardware. The @code{section}
2801 attribute specifies that a variable (or function) lives in a particular
2802 section. For example, this small program uses several specific section names:
2805 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
2806 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
2807 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
2808 int init_data __attribute__ ((section ("INITDATA"))) = 0;
2812 /* Initialize stack pointer */
2813 init_sp (stack + sizeof (stack));
2815 /* Initialize initialized data */
2816 memcpy (&init_data, &data, &edata - &data);
2818 /* Turn on the serial ports */
2825 Use the @code{section} attribute with an @emph{initialized} definition
2826 of a @emph{global} variable, as shown in the example. GCC issues
2827 a warning and otherwise ignores the @code{section} attribute in
2828 uninitialized variable declarations.
2830 You may only use the @code{section} attribute with a fully initialized
2831 global definition because of the way linkers work. The linker requires
2832 each object be defined once, with the exception that uninitialized
2833 variables tentatively go in the @code{common} (or @code{bss}) section
2834 and can be multiply ``defined''. You can force a variable to be
2835 initialized with the @option{-fno-common} flag or the @code{nocommon}
2838 Some file formats do not support arbitrary sections so the @code{section}
2839 attribute is not available on all platforms.
2840 If you need to map the entire contents of a module to a particular
2841 section, consider using the facilities of the linker instead.
2844 @cindex @code{shared} variable attribute
2845 On Microsoft Windows, in addition to putting variable definitions in a named
2846 section, the section can also be shared among all running copies of an
2847 executable or DLL@. For example, this small program defines shared data
2848 by putting it in a named section @code{shared} and marking the section
2852 int foo __attribute__((section ("shared"), shared)) = 0;
2857 /* Read and write foo. All running
2858 copies see the same value. */
2864 You may only use the @code{shared} attribute along with @code{section}
2865 attribute with a fully initialized global definition because of the way
2866 linkers work. See @code{section} attribute for more information.
2868 The @code{shared} attribute is only available on Microsoft Windows@.
2870 @item tls_model ("@var{tls_model}")
2871 @cindex @code{tls_model} attribute
2872 The @code{tls_model} attribute sets thread-local storage model
2873 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
2874 overriding @code{-ftls-model=} command line switch on a per-variable
2876 The @var{tls_model} argument should be one of @code{global-dynamic},
2877 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
2879 Not all targets support this attribute.
2881 @item transparent_union
2882 This attribute, attached to a function parameter which is a union, means
2883 that the corresponding argument may have the type of any union member,
2884 but the argument is passed as if its type were that of the first union
2885 member. For more details see @xref{Type Attributes}. You can also use
2886 this attribute on a @code{typedef} for a union data type; then it
2887 applies to all function parameters with that type.
2890 This attribute, attached to a variable, means that the variable is meant
2891 to be possibly unused. GCC will not produce a warning for this
2894 @item vector_size (@var{bytes})
2895 This attribute specifies the vector size for the variable, measured in
2896 bytes. For example, the declaration:
2899 int foo __attribute__ ((vector_size (16)));
2903 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
2904 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
2905 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
2907 This attribute is only applicable to integral and float scalars,
2908 although arrays, pointers, and function return values are allowed in
2909 conjunction with this construct.
2911 Aggregates with this attribute are invalid, even if they are of the same
2912 size as a corresponding scalar. For example, the declaration:
2915 struct S @{ int a; @};
2916 struct S __attribute__ ((vector_size (16))) foo;
2920 is invalid even if the size of the structure is the same as the size of
2924 The @code{weak} attribute is described in @xref{Function Attributes}.
2927 The @code{dllimport} attribute is described in @xref{Function Attributes}.
2930 The @code{dllexport} attribute is described in @xref{Function Attributes}.
2934 @subsection M32R/D Variable Attributes
2936 One attribute is currently defined for the M32R/D.
2939 @item model (@var{model-name})
2940 @cindex variable addressability on the M32R/D
2941 Use this attribute on the M32R/D to set the addressability of an object.
2942 The identifier @var{model-name} is one of @code{small}, @code{medium},
2943 or @code{large}, representing each of the code models.
2945 Small model objects live in the lower 16MB of memory (so that their
2946 addresses can be loaded with the @code{ld24} instruction).
2948 Medium and large model objects may live anywhere in the 32-bit address space
2949 (the compiler will generate @code{seth/add3} instructions to load their
2953 @subsection i386 Variable Attributes
2955 Two attributes are currently defined for i386 configurations:
2956 @code{ms_struct} and @code{gcc_struct}
2961 @cindex @code{ms_struct} attribute
2962 @cindex @code{gcc_struct} attribute
2964 If @code{packed} is used on a structure, or if bit-fields are used
2965 it may be that the Microsoft ABI packs them differently
2966 than GCC would normally pack them. Particularly when moving packed
2967 data between functions compiled with GCC and the native Microsoft compiler
2968 (either via function call or as data in a file), it may be necessary to access
2971 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
2972 compilers to match the native Microsoft compiler.
2975 @node Type Attributes
2976 @section Specifying Attributes of Types
2977 @cindex attribute of types
2978 @cindex type attributes
2980 The keyword @code{__attribute__} allows you to specify special
2981 attributes of @code{struct} and @code{union} types when you define such
2982 types. This keyword is followed by an attribute specification inside
2983 double parentheses. Six attributes are currently defined for types:
2984 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
2985 @code{deprecated} and @code{may_alias}. Other attributes are defined for
2986 functions (@pxref{Function Attributes}) and for variables
2987 (@pxref{Variable Attributes}).
2989 You may also specify any one of these attributes with @samp{__}
2990 preceding and following its keyword. This allows you to use these
2991 attributes in header files without being concerned about a possible
2992 macro of the same name. For example, you may use @code{__aligned__}
2993 instead of @code{aligned}.
2995 You may specify the @code{aligned} and @code{transparent_union}
2996 attributes either in a @code{typedef} declaration or just past the
2997 closing curly brace of a complete enum, struct or union type
2998 @emph{definition} and the @code{packed} attribute only past the closing
2999 brace of a definition.
3001 You may also specify attributes between the enum, struct or union
3002 tag and the name of the type rather than after the closing brace.
3004 @xref{Attribute Syntax}, for details of the exact syntax for using
3008 @cindex @code{aligned} attribute
3009 @item aligned (@var{alignment})
3010 This attribute specifies a minimum alignment (in bytes) for variables
3011 of the specified type. For example, the declarations:
3014 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3015 typedef int more_aligned_int __attribute__ ((aligned (8)));
3019 force the compiler to insure (as far as it can) that each variable whose
3020 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3021 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
3022 variables of type @code{struct S} aligned to 8-byte boundaries allows
3023 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3024 store) instructions when copying one variable of type @code{struct S} to
3025 another, thus improving run-time efficiency.
3027 Note that the alignment of any given @code{struct} or @code{union} type
3028 is required by the ISO C standard to be at least a perfect multiple of
3029 the lowest common multiple of the alignments of all of the members of
3030 the @code{struct} or @code{union} in question. This means that you @emph{can}
3031 effectively adjust the alignment of a @code{struct} or @code{union}
3032 type by attaching an @code{aligned} attribute to any one of the members
3033 of such a type, but the notation illustrated in the example above is a
3034 more obvious, intuitive, and readable way to request the compiler to
3035 adjust the alignment of an entire @code{struct} or @code{union} type.
3037 As in the preceding example, you can explicitly specify the alignment
3038 (in bytes) that you wish the compiler to use for a given @code{struct}
3039 or @code{union} type. Alternatively, you can leave out the alignment factor
3040 and just ask the compiler to align a type to the maximum
3041 useful alignment for the target machine you are compiling for. For
3042 example, you could write:
3045 struct S @{ short f[3]; @} __attribute__ ((aligned));
3048 Whenever you leave out the alignment factor in an @code{aligned}
3049 attribute specification, the compiler automatically sets the alignment
3050 for the type to the largest alignment which is ever used for any data
3051 type on the target machine you are compiling for. Doing this can often
3052 make copy operations more efficient, because the compiler can use
3053 whatever instructions copy the biggest chunks of memory when performing
3054 copies to or from the variables which have types that you have aligned
3057 In the example above, if the size of each @code{short} is 2 bytes, then
3058 the size of the entire @code{struct S} type is 6 bytes. The smallest
3059 power of two which is greater than or equal to that is 8, so the
3060 compiler sets the alignment for the entire @code{struct S} type to 8
3063 Note that although you can ask the compiler to select a time-efficient
3064 alignment for a given type and then declare only individual stand-alone
3065 objects of that type, the compiler's ability to select a time-efficient
3066 alignment is primarily useful only when you plan to create arrays of
3067 variables having the relevant (efficiently aligned) type. If you
3068 declare or use arrays of variables of an efficiently-aligned type, then
3069 it is likely that your program will also be doing pointer arithmetic (or
3070 subscripting, which amounts to the same thing) on pointers to the
3071 relevant type, and the code that the compiler generates for these
3072 pointer arithmetic operations will often be more efficient for
3073 efficiently-aligned types than for other types.
3075 The @code{aligned} attribute can only increase the alignment; but you
3076 can decrease it by specifying @code{packed} as well. See below.
3078 Note that the effectiveness of @code{aligned} attributes may be limited
3079 by inherent limitations in your linker. On many systems, the linker is
3080 only able to arrange for variables to be aligned up to a certain maximum
3081 alignment. (For some linkers, the maximum supported alignment may
3082 be very very small.) If your linker is only able to align variables
3083 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3084 in an @code{__attribute__} will still only provide you with 8 byte
3085 alignment. See your linker documentation for further information.
3088 This attribute, attached to @code{struct} or @code{union} type
3089 definition, specifies that each member of the structure or union is
3090 placed to minimize the memory required. When attached to an @code{enum}
3091 definition, it indicates that the smallest integral type should be used.
3093 @opindex fshort-enums
3094 Specifying this attribute for @code{struct} and @code{union} types is
3095 equivalent to specifying the @code{packed} attribute on each of the
3096 structure or union members. Specifying the @option{-fshort-enums}
3097 flag on the line is equivalent to specifying the @code{packed}
3098 attribute on all @code{enum} definitions.
3100 In the following example @code{struct my_packed_struct}'s members are
3101 packed closely together, but the internal layout of its @code{s} member
3102 is not packed -- to do that, @code{struct my_unpacked_struct} would need to
3106 struct my_unpacked_struct
3112 struct my_packed_struct __attribute__ ((__packed__))
3116 struct my_unpacked_struct s;
3120 You may only specify this attribute on the definition of a @code{enum},
3121 @code{struct} or @code{union}, not on a @code{typedef} which does not
3122 also define the enumerated type, structure or union.
3124 @item transparent_union
3125 This attribute, attached to a @code{union} type definition, indicates
3126 that any function parameter having that union type causes calls to that
3127 function to be treated in a special way.
3129 First, the argument corresponding to a transparent union type can be of
3130 any type in the union; no cast is required. Also, if the union contains
3131 a pointer type, the corresponding argument can be a null pointer
3132 constant or a void pointer expression; and if the union contains a void
3133 pointer type, the corresponding argument can be any pointer expression.
3134 If the union member type is a pointer, qualifiers like @code{const} on
3135 the referenced type must be respected, just as with normal pointer
3138 Second, the argument is passed to the function using the calling
3139 conventions of the first member of the transparent union, not the calling
3140 conventions of the union itself. All members of the union must have the
3141 same machine representation; this is necessary for this argument passing
3144 Transparent unions are designed for library functions that have multiple
3145 interfaces for compatibility reasons. For example, suppose the
3146 @code{wait} function must accept either a value of type @code{int *} to
3147 comply with Posix, or a value of type @code{union wait *} to comply with
3148 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3149 @code{wait} would accept both kinds of arguments, but it would also
3150 accept any other pointer type and this would make argument type checking
3151 less useful. Instead, @code{<sys/wait.h>} might define the interface
3159 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3161 pid_t wait (wait_status_ptr_t);
3164 This interface allows either @code{int *} or @code{union wait *}
3165 arguments to be passed, using the @code{int *} calling convention.
3166 The program can call @code{wait} with arguments of either type:
3169 int w1 () @{ int w; return wait (&w); @}
3170 int w2 () @{ union wait w; return wait (&w); @}
3173 With this interface, @code{wait}'s implementation might look like this:
3176 pid_t wait (wait_status_ptr_t p)
3178 return waitpid (-1, p.__ip, 0);
3183 When attached to a type (including a @code{union} or a @code{struct}),
3184 this attribute means that variables of that type are meant to appear
3185 possibly unused. GCC will not produce a warning for any variables of
3186 that type, even if the variable appears to do nothing. This is often
3187 the case with lock or thread classes, which are usually defined and then
3188 not referenced, but contain constructors and destructors that have
3189 nontrivial bookkeeping functions.
3192 The @code{deprecated} attribute results in a warning if the type
3193 is used anywhere in the source file. This is useful when identifying
3194 types that are expected to be removed in a future version of a program.
3195 If possible, the warning also includes the location of the declaration
3196 of the deprecated type, to enable users to easily find further
3197 information about why the type is deprecated, or what they should do
3198 instead. Note that the warnings only occur for uses and then only
3199 if the type is being applied to an identifier that itself is not being
3200 declared as deprecated.
3203 typedef int T1 __attribute__ ((deprecated));
3207 typedef T1 T3 __attribute__ ((deprecated));
3208 T3 z __attribute__ ((deprecated));
3211 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3212 warning is issued for line 4 because T2 is not explicitly
3213 deprecated. Line 5 has no warning because T3 is explicitly
3214 deprecated. Similarly for line 6.
3216 The @code{deprecated} attribute can also be used for functions and
3217 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3220 Accesses to objects with types with this attribute are not subjected to
3221 type-based alias analysis, but are instead assumed to be able to alias
3222 any other type of objects, just like the @code{char} type. See
3223 @option{-fstrict-aliasing} for more information on aliasing issues.
3228 typedef short __attribute__((__may_alias__)) short_a;
3234 short_a *b = (short_a *) &a;
3238 if (a == 0x12345678)
3245 If you replaced @code{short_a} with @code{short} in the variable
3246 declaration, the above program would abort when compiled with
3247 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3248 above in recent GCC versions.
3250 @subsection i386 Type Attributes
3252 Two attributes are currently defined for i386 configurations:
3253 @code{ms_struct} and @code{gcc_struct}
3257 @cindex @code{ms_struct}
3258 @cindex @code{gcc_struct}
3260 If @code{packed} is used on a structure, or if bit-fields are used
3261 it may be that the Microsoft ABI packs them differently
3262 than GCC would normally pack them. Particularly when moving packed
3263 data between functions compiled with GCC and the native Microsoft compiler
3264 (either via function call or as data in a file), it may be necessary to access
3267 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3268 compilers to match the native Microsoft compiler.
3271 To specify multiple attributes, separate them by commas within the
3272 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3276 @section An Inline Function is As Fast As a Macro
3277 @cindex inline functions
3278 @cindex integrating function code
3280 @cindex macros, inline alternative
3282 By declaring a function @code{inline}, you can direct GCC to
3283 integrate that function's code into the code for its callers. This
3284 makes execution faster by eliminating the function-call overhead; in
3285 addition, if any of the actual argument values are constant, their known
3286 values may permit simplifications at compile time so that not all of the
3287 inline function's code needs to be included. The effect on code size is
3288 less predictable; object code may be larger or smaller with function
3289 inlining, depending on the particular case. Inlining of functions is an
3290 optimization and it really ``works'' only in optimizing compilation. If
3291 you don't use @option{-O}, no function is really inline.
3293 Inline functions are included in the ISO C99 standard, but there are
3294 currently substantial differences between what GCC implements and what
3295 the ISO C99 standard requires.
3297 To declare a function inline, use the @code{inline} keyword in its
3298 declaration, like this:
3308 (If you are writing a header file to be included in ISO C programs, write
3309 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3310 You can also make all ``simple enough'' functions inline with the option
3311 @option{-finline-functions}.
3314 Note that certain usages in a function definition can make it unsuitable
3315 for inline substitution. Among these usages are: use of varargs, use of
3316 alloca, use of variable sized data types (@pxref{Variable Length}),
3317 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3318 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3319 will warn when a function marked @code{inline} could not be substituted,
3320 and will give the reason for the failure.
3322 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3323 does not affect the linkage of the function.
3325 @cindex automatic @code{inline} for C++ member fns
3326 @cindex @code{inline} automatic for C++ member fns
3327 @cindex member fns, automatically @code{inline}
3328 @cindex C++ member fns, automatically @code{inline}
3329 @opindex fno-default-inline
3330 GCC automatically inlines member functions defined within the class
3331 body of C++ programs even if they are not explicitly declared
3332 @code{inline}. (You can override this with @option{-fno-default-inline};
3333 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3335 @cindex inline functions, omission of
3336 @opindex fkeep-inline-functions
3337 When a function is both inline and @code{static}, if all calls to the
3338 function are integrated into the caller, and the function's address is
3339 never used, then the function's own assembler code is never referenced.
3340 In this case, GCC does not actually output assembler code for the
3341 function, unless you specify the option @option{-fkeep-inline-functions}.
3342 Some calls cannot be integrated for various reasons (in particular,
3343 calls that precede the function's definition cannot be integrated, and
3344 neither can recursive calls within the definition). If there is a
3345 nonintegrated call, then the function is compiled to assembler code as
3346 usual. The function must also be compiled as usual if the program
3347 refers to its address, because that can't be inlined.
3349 @cindex non-static inline function
3350 When an inline function is not @code{static}, then the compiler must assume
3351 that there may be calls from other source files; since a global symbol can
3352 be defined only once in any program, the function must not be defined in
3353 the other source files, so the calls therein cannot be integrated.
3354 Therefore, a non-@code{static} inline function is always compiled on its
3355 own in the usual fashion.
3357 If you specify both @code{inline} and @code{extern} in the function
3358 definition, then the definition is used only for inlining. In no case
3359 is the function compiled on its own, not even if you refer to its
3360 address explicitly. Such an address becomes an external reference, as
3361 if you had only declared the function, and had not defined it.
3363 This combination of @code{inline} and @code{extern} has almost the
3364 effect of a macro. The way to use it is to put a function definition in
3365 a header file with these keywords, and put another copy of the
3366 definition (lacking @code{inline} and @code{extern}) in a library file.
3367 The definition in the header file will cause most calls to the function
3368 to be inlined. If any uses of the function remain, they will refer to
3369 the single copy in the library.
3371 Since GCC eventually will implement ISO C99 semantics for
3372 inline functions, it is best to use @code{static inline} only
3373 to guarantee compatibility. (The
3374 existing semantics will remain available when @option{-std=gnu89} is
3375 specified, but eventually the default will be @option{-std=gnu99} and
3376 that will implement the C99 semantics, though it does not do so yet.)
3378 GCC does not inline any functions when not optimizing unless you specify
3379 the @samp{always_inline} attribute for the function, like this:
3383 inline void foo (const char) __attribute__((always_inline));
3387 @section Assembler Instructions with C Expression Operands
3388 @cindex extended @code{asm}
3389 @cindex @code{asm} expressions
3390 @cindex assembler instructions
3393 In an assembler instruction using @code{asm}, you can specify the
3394 operands of the instruction using C expressions. This means you need not
3395 guess which registers or memory locations will contain the data you want
3398 You must specify an assembler instruction template much like what
3399 appears in a machine description, plus an operand constraint string for
3402 For example, here is how to use the 68881's @code{fsinx} instruction:
3405 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3409 Here @code{angle} is the C expression for the input operand while
3410 @code{result} is that of the output operand. Each has @samp{"f"} as its
3411 operand constraint, saying that a floating point register is required.
3412 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3413 output operands' constraints must use @samp{=}. The constraints use the
3414 same language used in the machine description (@pxref{Constraints}).
3416 Each operand is described by an operand-constraint string followed by
3417 the C expression in parentheses. A colon separates the assembler
3418 template from the first output operand and another separates the last
3419 output operand from the first input, if any. Commas separate the
3420 operands within each group. The total number of operands is currently
3421 limited to 30; this limitation may be lifted in some future version of
3424 If there are no output operands but there are input operands, you must
3425 place two consecutive colons surrounding the place where the output
3428 As of GCC version 3.1, it is also possible to specify input and output
3429 operands using symbolic names which can be referenced within the
3430 assembler code. These names are specified inside square brackets
3431 preceding the constraint string, and can be referenced inside the
3432 assembler code using @code{%[@var{name}]} instead of a percentage sign
3433 followed by the operand number. Using named operands the above example
3437 asm ("fsinx %[angle],%[output]"
3438 : [output] "=f" (result)
3439 : [angle] "f" (angle));
3443 Note that the symbolic operand names have no relation whatsoever to
3444 other C identifiers. You may use any name you like, even those of
3445 existing C symbols, but you must ensure that no two operands within the same
3446 assembler construct use the same symbolic name.
3448 Output operand expressions must be lvalues; the compiler can check this.
3449 The input operands need not be lvalues. The compiler cannot check
3450 whether the operands have data types that are reasonable for the
3451 instruction being executed. It does not parse the assembler instruction
3452 template and does not know what it means or even whether it is valid
3453 assembler input. The extended @code{asm} feature is most often used for
3454 machine instructions the compiler itself does not know exist. If
3455 the output expression cannot be directly addressed (for example, it is a
3456 bit-field), your constraint must allow a register. In that case, GCC
3457 will use the register as the output of the @code{asm}, and then store
3458 that register into the output.
3460 The ordinary output operands must be write-only; GCC will assume that
3461 the values in these operands before the instruction are dead and need
3462 not be generated. Extended asm supports input-output or read-write
3463 operands. Use the constraint character @samp{+} to indicate such an
3464 operand and list it with the output operands. You should only use
3465 read-write operands when the constraints for the operand (or the
3466 operand in which only some of the bits are to be changed) allow a
3469 You may, as an alternative, logically split its function into two
3470 separate operands, one input operand and one write-only output
3471 operand. The connection between them is expressed by constraints
3472 which say they need to be in the same location when the instruction
3473 executes. You can use the same C expression for both operands, or
3474 different expressions. For example, here we write the (fictitious)
3475 @samp{combine} instruction with @code{bar} as its read-only source
3476 operand and @code{foo} as its read-write destination:
3479 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3483 The constraint @samp{"0"} for operand 1 says that it must occupy the
3484 same location as operand 0. A number in constraint is allowed only in
3485 an input operand and it must refer to an output operand.
3487 Only a number in the constraint can guarantee that one operand will be in
3488 the same place as another. The mere fact that @code{foo} is the value
3489 of both operands is not enough to guarantee that they will be in the
3490 same place in the generated assembler code. The following would not
3494 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3497 Various optimizations or reloading could cause operands 0 and 1 to be in
3498 different registers; GCC knows no reason not to do so. For example, the
3499 compiler might find a copy of the value of @code{foo} in one register and
3500 use it for operand 1, but generate the output operand 0 in a different
3501 register (copying it afterward to @code{foo}'s own address). Of course,
3502 since the register for operand 1 is not even mentioned in the assembler
3503 code, the result will not work, but GCC can't tell that.
3505 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3506 the operand number for a matching constraint. For example:
3509 asm ("cmoveq %1,%2,%[result]"
3510 : [result] "=r"(result)
3511 : "r" (test), "r"(new), "[result]"(old));
3514 Some instructions clobber specific hard registers. To describe this,
3515 write a third colon after the input operands, followed by the names of
3516 the clobbered hard registers (given as strings). Here is a realistic
3517 example for the VAX:
3520 asm volatile ("movc3 %0,%1,%2"
3522 : "g" (from), "g" (to), "g" (count)
3523 : "r0", "r1", "r2", "r3", "r4", "r5");
3526 You may not write a clobber description in a way that overlaps with an
3527 input or output operand. For example, you may not have an operand
3528 describing a register class with one member if you mention that register
3529 in the clobber list. Variables declared to live in specific registers
3530 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
3531 have no part mentioned in the clobber description.
3532 There is no way for you to specify that an input
3533 operand is modified without also specifying it as an output
3534 operand. Note that if all the output operands you specify are for this
3535 purpose (and hence unused), you will then also need to specify
3536 @code{volatile} for the @code{asm} construct, as described below, to
3537 prevent GCC from deleting the @code{asm} statement as unused.
3539 If you refer to a particular hardware register from the assembler code,
3540 you will probably have to list the register after the third colon to
3541 tell the compiler the register's value is modified. In some assemblers,
3542 the register names begin with @samp{%}; to produce one @samp{%} in the
3543 assembler code, you must write @samp{%%} in the input.
3545 If your assembler instruction can alter the condition code register, add
3546 @samp{cc} to the list of clobbered registers. GCC on some machines
3547 represents the condition codes as a specific hardware register;
3548 @samp{cc} serves to name this register. On other machines, the
3549 condition code is handled differently, and specifying @samp{cc} has no
3550 effect. But it is valid no matter what the machine.
3552 If your assembler instructions access memory in an unpredictable
3553 fashion, add @samp{memory} to the list of clobbered registers. This
3554 will cause GCC to not keep memory values cached in registers across the
3555 assembler instruction and not optimize stores or loads to that memory.
3556 You will also want to add the @code{volatile} keyword if the memory
3557 affected is not listed in the inputs or outputs of the @code{asm}, as
3558 the @samp{memory} clobber does not count as a side-effect of the
3559 @code{asm}. If you know how large the accessed memory is, you can add
3560 it as input or output but if this is not known, you should add
3561 @samp{memory}. As an example, if you access ten bytes of a string, you
3562 can use a memory input like:
3565 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
3568 Note that in the following example the memory input is necessary,
3569 otherwise GCC might optimize the store to @code{x} away:
3576 asm ("magic stuff accessing an 'int' pointed to by '%1'"
3577 "=&d" (r) : "a" (y), "m" (*y));
3582 You can put multiple assembler instructions together in a single
3583 @code{asm} template, separated by the characters normally used in assembly
3584 code for the system. A combination that works in most places is a newline
3585 to break the line, plus a tab character to move to the instruction field
3586 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3587 assembler allows semicolons as a line-breaking character. Note that some
3588 assembler dialects use semicolons to start a comment.
3589 The input operands are guaranteed not to use any of the clobbered
3590 registers, and neither will the output operands' addresses, so you can
3591 read and write the clobbered registers as many times as you like. Here
3592 is an example of multiple instructions in a template; it assumes the
3593 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3596 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3598 : "g" (from), "g" (to)
3602 Unless an output operand has the @samp{&} constraint modifier, GCC
3603 may allocate it in the same register as an unrelated input operand, on
3604 the assumption the inputs are consumed before the outputs are produced.
3605 This assumption may be false if the assembler code actually consists of
3606 more than one instruction. In such a case, use @samp{&} for each output
3607 operand that may not overlap an input. @xref{Modifiers}.
3609 If you want to test the condition code produced by an assembler
3610 instruction, you must include a branch and a label in the @code{asm}
3611 construct, as follows:
3614 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3620 This assumes your assembler supports local labels, as the GNU assembler
3621 and most Unix assemblers do.
3623 Speaking of labels, jumps from one @code{asm} to another are not
3624 supported. The compiler's optimizers do not know about these jumps, and
3625 therefore they cannot take account of them when deciding how to
3628 @cindex macros containing @code{asm}
3629 Usually the most convenient way to use these @code{asm} instructions is to
3630 encapsulate them in macros that look like functions. For example,
3634 (@{ double __value, __arg = (x); \
3635 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3640 Here the variable @code{__arg} is used to make sure that the instruction
3641 operates on a proper @code{double} value, and to accept only those
3642 arguments @code{x} which can convert automatically to a @code{double}.
3644 Another way to make sure the instruction operates on the correct data
3645 type is to use a cast in the @code{asm}. This is different from using a
3646 variable @code{__arg} in that it converts more different types. For
3647 example, if the desired type were @code{int}, casting the argument to
3648 @code{int} would accept a pointer with no complaint, while assigning the
3649 argument to an @code{int} variable named @code{__arg} would warn about
3650 using a pointer unless the caller explicitly casts it.
3652 If an @code{asm} has output operands, GCC assumes for optimization
3653 purposes the instruction has no side effects except to change the output
3654 operands. This does not mean instructions with a side effect cannot be
3655 used, but you must be careful, because the compiler may eliminate them
3656 if the output operands aren't used, or move them out of loops, or
3657 replace two with one if they constitute a common subexpression. Also,
3658 if your instruction does have a side effect on a variable that otherwise
3659 appears not to change, the old value of the variable may be reused later
3660 if it happens to be found in a register.
3662 You can prevent an @code{asm} instruction from being deleted, moved
3663 significantly, or combined, by writing the keyword @code{volatile} after
3664 the @code{asm}. For example:
3667 #define get_and_set_priority(new) \
3669 asm volatile ("get_and_set_priority %0, %1" \
3670 : "=g" (__old) : "g" (new)); \
3675 If you write an @code{asm} instruction with no outputs, GCC will know
3676 the instruction has side-effects and will not delete the instruction or
3677 move it outside of loops.
3679 The @code{volatile} keyword indicates that the instruction has
3680 important side-effects. GCC will not delete a volatile @code{asm} if
3681 it is reachable. (The instruction can still be deleted if GCC can
3682 prove that control-flow will never reach the location of the
3683 instruction.) In addition, GCC will not reschedule instructions
3684 across a volatile @code{asm} instruction. For example:
3687 *(volatile int *)addr = foo;
3688 asm volatile ("eieio" : : );
3692 Assume @code{addr} contains the address of a memory mapped device
3693 register. The PowerPC @code{eieio} instruction (Enforce In-order
3694 Execution of I/O) tells the CPU to make sure that the store to that
3695 device register happens before it issues any other I/O@.
3697 Note that even a volatile @code{asm} instruction can be moved in ways
3698 that appear insignificant to the compiler, such as across jump
3699 instructions. You can't expect a sequence of volatile @code{asm}
3700 instructions to remain perfectly consecutive. If you want consecutive
3701 output, use a single @code{asm}. Also, GCC will perform some
3702 optimizations across a volatile @code{asm} instruction; GCC does not
3703 ``forget everything'' when it encounters a volatile @code{asm}
3704 instruction the way some other compilers do.
3706 An @code{asm} instruction without any operands or clobbers (an ``old
3707 style'' @code{asm}) will be treated identically to a volatile
3708 @code{asm} instruction.
3710 It is a natural idea to look for a way to give access to the condition
3711 code left by the assembler instruction. However, when we attempted to
3712 implement this, we found no way to make it work reliably. The problem
3713 is that output operands might need reloading, which would result in
3714 additional following ``store'' instructions. On most machines, these
3715 instructions would alter the condition code before there was time to
3716 test it. This problem doesn't arise for ordinary ``test'' and
3717 ``compare'' instructions because they don't have any output operands.
3719 For reasons similar to those described above, it is not possible to give
3720 an assembler instruction access to the condition code left by previous
3723 If you are writing a header file that should be includable in ISO C
3724 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3727 @subsection Size of an @code{asm}
3729 Some targets require that GCC track the size of each instruction used in
3730 order to generate correct code. Because the final length of an
3731 @code{asm} is only known by the assembler, GCC must make an estimate as
3732 to how big it will be. The estimate is formed by counting the number of
3733 statements in the pattern of the @code{asm} and multiplying that by the
3734 length of the longest instruction on that processor. Statements in the
3735 @code{asm} are identified by newline characters and whatever statement
3736 separator characters are supported by the assembler; on most processors
3737 this is the `@code{;}' character.
3739 Normally, GCC's estimate is perfectly adequate to ensure that correct
3740 code is generated, but it is possible to confuse the compiler if you use
3741 pseudo instructions or assembler macros that expand into multiple real
3742 instructions or if you use assembler directives that expand to more
3743 space in the object file than would be needed for a single instruction.
3744 If this happens then the assembler will produce a diagnostic saying that
3745 a label is unreachable.
3747 @subsection i386 floating point asm operands
3749 There are several rules on the usage of stack-like regs in
3750 asm_operands insns. These rules apply only to the operands that are
3755 Given a set of input regs that die in an asm_operands, it is
3756 necessary to know which are implicitly popped by the asm, and
3757 which must be explicitly popped by gcc.
3759 An input reg that is implicitly popped by the asm must be
3760 explicitly clobbered, unless it is constrained to match an
3764 For any input reg that is implicitly popped by an asm, it is
3765 necessary to know how to adjust the stack to compensate for the pop.
3766 If any non-popped input is closer to the top of the reg-stack than
3767 the implicitly popped reg, it would not be possible to know what the
3768 stack looked like---it's not clear how the rest of the stack ``slides
3771 All implicitly popped input regs must be closer to the top of
3772 the reg-stack than any input that is not implicitly popped.
3774 It is possible that if an input dies in an insn, reload might
3775 use the input reg for an output reload. Consider this example:
3778 asm ("foo" : "=t" (a) : "f" (b));
3781 This asm says that input B is not popped by the asm, and that
3782 the asm pushes a result onto the reg-stack, i.e., the stack is one
3783 deeper after the asm than it was before. But, it is possible that
3784 reload will think that it can use the same reg for both the input and
3785 the output, if input B dies in this insn.
3787 If any input operand uses the @code{f} constraint, all output reg
3788 constraints must use the @code{&} earlyclobber.
3790 The asm above would be written as
3793 asm ("foo" : "=&t" (a) : "f" (b));
3797 Some operands need to be in particular places on the stack. All
3798 output operands fall in this category---there is no other way to
3799 know which regs the outputs appear in unless the user indicates
3800 this in the constraints.
3802 Output operands must specifically indicate which reg an output
3803 appears in after an asm. @code{=f} is not allowed: the operand
3804 constraints must select a class with a single reg.
3807 Output operands may not be ``inserted'' between existing stack regs.
3808 Since no 387 opcode uses a read/write operand, all output operands
3809 are dead before the asm_operands, and are pushed by the asm_operands.
3810 It makes no sense to push anywhere but the top of the reg-stack.
3812 Output operands must start at the top of the reg-stack: output
3813 operands may not ``skip'' a reg.
3816 Some asm statements may need extra stack space for internal
3817 calculations. This can be guaranteed by clobbering stack registers
3818 unrelated to the inputs and outputs.
3822 Here are a couple of reasonable asms to want to write. This asm
3823 takes one input, which is internally popped, and produces two outputs.
3826 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
3829 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
3830 and replaces them with one output. The user must code the @code{st(1)}
3831 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
3834 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
3840 @section Controlling Names Used in Assembler Code
3841 @cindex assembler names for identifiers
3842 @cindex names used in assembler code
3843 @cindex identifiers, names in assembler code
3845 You can specify the name to be used in the assembler code for a C
3846 function or variable by writing the @code{asm} (or @code{__asm__})
3847 keyword after the declarator as follows:
3850 int foo asm ("myfoo") = 2;
3854 This specifies that the name to be used for the variable @code{foo} in
3855 the assembler code should be @samp{myfoo} rather than the usual
3858 On systems where an underscore is normally prepended to the name of a C
3859 function or variable, this feature allows you to define names for the
3860 linker that do not start with an underscore.
3862 It does not make sense to use this feature with a non-static local
3863 variable since such variables do not have assembler names. If you are
3864 trying to put the variable in a particular register, see @ref{Explicit
3865 Reg Vars}. GCC presently accepts such code with a warning, but will
3866 probably be changed to issue an error, rather than a warning, in the
3869 You cannot use @code{asm} in this way in a function @emph{definition}; but
3870 you can get the same effect by writing a declaration for the function
3871 before its definition and putting @code{asm} there, like this:
3874 extern func () asm ("FUNC");
3881 It is up to you to make sure that the assembler names you choose do not
3882 conflict with any other assembler symbols. Also, you must not use a
3883 register name; that would produce completely invalid assembler code. GCC
3884 does not as yet have the ability to store static variables in registers.
3885 Perhaps that will be added.
3887 @node Explicit Reg Vars
3888 @section Variables in Specified Registers
3889 @cindex explicit register variables
3890 @cindex variables in specified registers
3891 @cindex specified registers
3892 @cindex registers, global allocation
3894 GNU C allows you to put a few global variables into specified hardware
3895 registers. You can also specify the register in which an ordinary
3896 register variable should be allocated.
3900 Global register variables reserve registers throughout the program.
3901 This may be useful in programs such as programming language
3902 interpreters which have a couple of global variables that are accessed
3906 Local register variables in specific registers do not reserve the
3907 registers. The compiler's data flow analysis is capable of determining
3908 where the specified registers contain live values, and where they are
3909 available for other uses. Stores into local register variables may be deleted
3910 when they appear to be dead according to dataflow analysis. References
3911 to local register variables may be deleted or moved or simplified.
3913 These local variables are sometimes convenient for use with the extended
3914 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
3915 output of the assembler instruction directly into a particular register.
3916 (This will work provided the register you specify fits the constraints
3917 specified for that operand in the @code{asm}.)
3925 @node Global Reg Vars
3926 @subsection Defining Global Register Variables
3927 @cindex global register variables
3928 @cindex registers, global variables in
3930 You can define a global register variable in GNU C like this:
3933 register int *foo asm ("a5");
3937 Here @code{a5} is the name of the register which should be used. Choose a
3938 register which is normally saved and restored by function calls on your
3939 machine, so that library routines will not clobber it.
3941 Naturally the register name is cpu-dependent, so you would need to
3942 conditionalize your program according to cpu type. The register
3943 @code{a5} would be a good choice on a 68000 for a variable of pointer
3944 type. On machines with register windows, be sure to choose a ``global''
3945 register that is not affected magically by the function call mechanism.
3947 In addition, operating systems on one type of cpu may differ in how they
3948 name the registers; then you would need additional conditionals. For
3949 example, some 68000 operating systems call this register @code{%a5}.
3951 Eventually there may be a way of asking the compiler to choose a register
3952 automatically, but first we need to figure out how it should choose and
3953 how to enable you to guide the choice. No solution is evident.
3955 Defining a global register variable in a certain register reserves that
3956 register entirely for this use, at least within the current compilation.
3957 The register will not be allocated for any other purpose in the functions
3958 in the current compilation. The register will not be saved and restored by
3959 these functions. Stores into this register are never deleted even if they
3960 would appear to be dead, but references may be deleted or moved or
3963 It is not safe to access the global register variables from signal
3964 handlers, or from more than one thread of control, because the system
3965 library routines may temporarily use the register for other things (unless
3966 you recompile them specially for the task at hand).
3968 @cindex @code{qsort}, and global register variables
3969 It is not safe for one function that uses a global register variable to
3970 call another such function @code{foo} by way of a third function
3971 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
3972 different source file in which the variable wasn't declared). This is
3973 because @code{lose} might save the register and put some other value there.
3974 For example, you can't expect a global register variable to be available in
3975 the comparison-function that you pass to @code{qsort}, since @code{qsort}
3976 might have put something else in that register. (If you are prepared to
3977 recompile @code{qsort} with the same global register variable, you can
3978 solve this problem.)
3980 If you want to recompile @code{qsort} or other source files which do not
3981 actually use your global register variable, so that they will not use that
3982 register for any other purpose, then it suffices to specify the compiler
3983 option @option{-ffixed-@var{reg}}. You need not actually add a global
3984 register declaration to their source code.
3986 A function which can alter the value of a global register variable cannot
3987 safely be called from a function compiled without this variable, because it
3988 could clobber the value the caller expects to find there on return.
3989 Therefore, the function which is the entry point into the part of the
3990 program that uses the global register variable must explicitly save and
3991 restore the value which belongs to its caller.
3993 @cindex register variable after @code{longjmp}
3994 @cindex global register after @code{longjmp}
3995 @cindex value after @code{longjmp}
3998 On most machines, @code{longjmp} will restore to each global register
3999 variable the value it had at the time of the @code{setjmp}. On some
4000 machines, however, @code{longjmp} will not change the value of global
4001 register variables. To be portable, the function that called @code{setjmp}
4002 should make other arrangements to save the values of the global register
4003 variables, and to restore them in a @code{longjmp}. This way, the same
4004 thing will happen regardless of what @code{longjmp} does.
4006 All global register variable declarations must precede all function
4007 definitions. If such a declaration could appear after function
4008 definitions, the declaration would be too late to prevent the register from
4009 being used for other purposes in the preceding functions.
4011 Global register variables may not have initial values, because an
4012 executable file has no means to supply initial contents for a register.
4014 On the SPARC, there are reports that g3 @dots{} g7 are suitable
4015 registers, but certain library functions, such as @code{getwd}, as well
4016 as the subroutines for division and remainder, modify g3 and g4. g1 and
4017 g2 are local temporaries.
4019 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4020 Of course, it will not do to use more than a few of those.
4022 @node Local Reg Vars
4023 @subsection Specifying Registers for Local Variables
4024 @cindex local variables, specifying registers
4025 @cindex specifying registers for local variables
4026 @cindex registers for local variables
4028 You can define a local register variable with a specified register
4032 register int *foo asm ("a5");
4036 Here @code{a5} is the name of the register which should be used. Note
4037 that this is the same syntax used for defining global register
4038 variables, but for a local variable it would appear within a function.
4040 Naturally the register name is cpu-dependent, but this is not a
4041 problem, since specific registers are most often useful with explicit
4042 assembler instructions (@pxref{Extended Asm}). Both of these things
4043 generally require that you conditionalize your program according to
4046 In addition, operating systems on one type of cpu may differ in how they
4047 name the registers; then you would need additional conditionals. For
4048 example, some 68000 operating systems call this register @code{%a5}.
4050 Defining such a register variable does not reserve the register; it
4051 remains available for other uses in places where flow control determines
4052 the variable's value is not live.
4054 This option does not guarantee that GCC will generate code that has
4055 this variable in the register you specify at all times. You may not
4056 code an explicit reference to this register in an @code{asm} statement
4057 and assume it will always refer to this variable.
4059 Stores into local register variables may be deleted when they appear to be dead
4060 according to dataflow analysis. References to local register variables may
4061 be deleted or moved or simplified.
4063 @node Alternate Keywords
4064 @section Alternate Keywords
4065 @cindex alternate keywords
4066 @cindex keywords, alternate
4068 @option{-ansi} and the various @option{-std} options disable certain
4069 keywords. This causes trouble when you want to use GNU C extensions, or
4070 a general-purpose header file that should be usable by all programs,
4071 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4072 @code{inline} are not available in programs compiled with
4073 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4074 program compiled with @option{-std=c99}). The ISO C99 keyword
4075 @code{restrict} is only available when @option{-std=gnu99} (which will
4076 eventually be the default) or @option{-std=c99} (or the equivalent
4077 @option{-std=iso9899:1999}) is used.
4079 The way to solve these problems is to put @samp{__} at the beginning and
4080 end of each problematical keyword. For example, use @code{__asm__}
4081 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4083 Other C compilers won't accept these alternative keywords; if you want to
4084 compile with another compiler, you can define the alternate keywords as
4085 macros to replace them with the customary keywords. It looks like this:
4093 @findex __extension__
4095 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4097 prevent such warnings within one expression by writing
4098 @code{__extension__} before the expression. @code{__extension__} has no
4099 effect aside from this.
4101 @node Incomplete Enums
4102 @section Incomplete @code{enum} Types
4104 You can define an @code{enum} tag without specifying its possible values.
4105 This results in an incomplete type, much like what you get if you write
4106 @code{struct foo} without describing the elements. A later declaration
4107 which does specify the possible values completes the type.
4109 You can't allocate variables or storage using the type while it is
4110 incomplete. However, you can work with pointers to that type.
4112 This extension may not be very useful, but it makes the handling of
4113 @code{enum} more consistent with the way @code{struct} and @code{union}
4116 This extension is not supported by GNU C++.
4118 @node Function Names
4119 @section Function Names as Strings
4120 @cindex @code{__func__} identifier
4121 @cindex @code{__FUNCTION__} identifier
4122 @cindex @code{__PRETTY_FUNCTION__} identifier
4124 GCC provides three magic variables which hold the name of the current
4125 function, as a string. The first of these is @code{__func__}, which
4126 is part of the C99 standard:
4129 The identifier @code{__func__} is implicitly declared by the translator
4130 as if, immediately following the opening brace of each function
4131 definition, the declaration
4134 static const char __func__[] = "function-name";
4137 appeared, where function-name is the name of the lexically-enclosing
4138 function. This name is the unadorned name of the function.
4141 @code{__FUNCTION__} is another name for @code{__func__}. Older
4142 versions of GCC recognize only this name. However, it is not
4143 standardized. For maximum portability, we recommend you use
4144 @code{__func__}, but provide a fallback definition with the
4148 #if __STDC_VERSION__ < 199901L
4150 # define __func__ __FUNCTION__
4152 # define __func__ "<unknown>"
4157 In C, @code{__PRETTY_FUNCTION__} is yet another name for
4158 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
4159 the type signature of the function as well as its bare name. For
4160 example, this program:
4164 extern int printf (char *, ...);
4171 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4172 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4190 __PRETTY_FUNCTION__ = void a::sub(int)
4193 These identifiers are not preprocessor macros. In GCC 3.3 and
4194 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
4195 were treated as string literals; they could be used to initialize
4196 @code{char} arrays, and they could be concatenated with other string
4197 literals. GCC 3.4 and later treat them as variables, like
4198 @code{__func__}. In C++, @code{__FUNCTION__} and
4199 @code{__PRETTY_FUNCTION__} have always been variables.
4201 @node Return Address
4202 @section Getting the Return or Frame Address of a Function
4204 These functions may be used to get information about the callers of a
4207 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4208 This function returns the return address of the current function, or of
4209 one of its callers. The @var{level} argument is number of frames to
4210 scan up the call stack. A value of @code{0} yields the return address
4211 of the current function, a value of @code{1} yields the return address
4212 of the caller of the current function, and so forth. When inlining
4213 the expected behavior is that the function will return the address of
4214 the function that will be returned to. To work around this behavior use
4215 the @code{noinline} function attribute.
4217 The @var{level} argument must be a constant integer.
4219 On some machines it may be impossible to determine the return address of
4220 any function other than the current one; in such cases, or when the top
4221 of the stack has been reached, this function will return @code{0} or a
4222 random value. In addition, @code{__builtin_frame_address} may be used
4223 to determine if the top of the stack has been reached.
4225 This function should only be used with a nonzero argument for debugging
4229 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4230 This function is similar to @code{__builtin_return_address}, but it
4231 returns the address of the function frame rather than the return address
4232 of the function. Calling @code{__builtin_frame_address} with a value of
4233 @code{0} yields the frame address of the current function, a value of
4234 @code{1} yields the frame address of the caller of the current function,
4237 The frame is the area on the stack which holds local variables and saved
4238 registers. The frame address is normally the address of the first word
4239 pushed on to the stack by the function. However, the exact definition
4240 depends upon the processor and the calling convention. If the processor
4241 has a dedicated frame pointer register, and the function has a frame,
4242 then @code{__builtin_frame_address} will return the value of the frame
4245 On some machines it may be impossible to determine the frame address of
4246 any function other than the current one; in such cases, or when the top
4247 of the stack has been reached, this function will return @code{0} if
4248 the first frame pointer is properly initialized by the startup code.
4250 This function should only be used with a nonzero argument for debugging
4254 @node Vector Extensions
4255 @section Using vector instructions through built-in functions
4257 On some targets, the instruction set contains SIMD vector instructions that
4258 operate on multiple values contained in one large register at the same time.
4259 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4262 The first step in using these extensions is to provide the necessary data
4263 types. This should be done using an appropriate @code{typedef}:
4266 typedef int v4si __attribute__ ((vector_size (16)));
4269 The @code{int} type specifies the base type, while the attribute specifies
4270 the vector size for the variable, measured in bytes. For example, the
4271 declaration above causes the compiler to set the mode for the @code{v4si}
4272 type to be 16 bytes wide and divided into @code{int} sized units. For
4273 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
4274 corresponding mode of @code{foo} will be @acronym{V4SI}.
4276 The @code{vector_size} attribute is only applicable to integral and
4277 float scalars, although arrays, pointers, and function return values
4278 are allowed in conjunction with this construct.
4280 All the basic integer types can be used as base types, both as signed
4281 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
4282 @code{long long}. In addition, @code{float} and @code{double} can be
4283 used to build floating-point vector types.
4285 Specifying a combination that is not valid for the current architecture
4286 will cause GCC to synthesize the instructions using a narrower mode.
4287 For example, if you specify a variable of type @code{V4SI} and your
4288 architecture does not allow for this specific SIMD type, GCC will
4289 produce code that uses 4 @code{SIs}.
4291 The types defined in this manner can be used with a subset of normal C
4292 operations. Currently, GCC will allow using the following operators
4293 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~}@.
4295 The operations behave like C++ @code{valarrays}. Addition is defined as
4296 the addition of the corresponding elements of the operands. For
4297 example, in the code below, each of the 4 elements in @var{a} will be
4298 added to the corresponding 4 elements in @var{b} and the resulting
4299 vector will be stored in @var{c}.
4302 typedef int v4si __attribute__ ((vector_size (16)));
4309 Subtraction, multiplication, division, and the logical operations
4310 operate in a similar manner. Likewise, the result of using the unary
4311 minus or complement operators on a vector type is a vector whose
4312 elements are the negative or complemented values of the corresponding
4313 elements in the operand.
4315 You can declare variables and use them in function calls and returns, as
4316 well as in assignments and some casts. You can specify a vector type as
4317 a return type for a function. Vector types can also be used as function
4318 arguments. It is possible to cast from one vector type to another,
4319 provided they are of the same size (in fact, you can also cast vectors
4320 to and from other datatypes of the same size).
4322 You cannot operate between vectors of different lengths or different
4323 signedness without a cast.
4325 A port that supports hardware vector operations, usually provides a set
4326 of built-in functions that can be used to operate on vectors. For
4327 example, a function to add two vectors and multiply the result by a
4328 third could look like this:
4331 v4si f (v4si a, v4si b, v4si c)
4333 v4si tmp = __builtin_addv4si (a, b);
4334 return __builtin_mulv4si (tmp, c);
4341 @findex __builtin_offsetof
4343 GCC implements for both C and C++ a syntactic extension to implement
4344 the @code{offsetof} macro.
4348 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
4350 offsetof_member_designator:
4352 | offsetof_member_designator "." @code{identifier}
4353 | offsetof_member_designator "[" @code{expr} "]"
4356 This extension is sufficient such that
4359 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
4362 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
4363 may be dependent. In either case, @var{member} may consist of a single
4364 identifier, or a sequence of member accesses and array references.
4366 @node Other Builtins
4367 @section Other built-in functions provided by GCC
4368 @cindex built-in functions
4369 @findex __builtin_isgreater
4370 @findex __builtin_isgreaterequal
4371 @findex __builtin_isless
4372 @findex __builtin_islessequal
4373 @findex __builtin_islessgreater
4374 @findex __builtin_isunordered
4529 @findex fprintf_unlocked
4531 @findex fputs_unlocked
4641 @findex printf_unlocked
4670 @findex significandf
4671 @findex significandl
4738 GCC provides a large number of built-in functions other than the ones
4739 mentioned above. Some of these are for internal use in the processing
4740 of exceptions or variable-length argument lists and will not be
4741 documented here because they may change from time to time; we do not
4742 recommend general use of these functions.
4744 The remaining functions are provided for optimization purposes.
4746 @opindex fno-builtin
4747 GCC includes built-in versions of many of the functions in the standard
4748 C library. The versions prefixed with @code{__builtin_} will always be
4749 treated as having the same meaning as the C library function even if you
4750 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
4751 Many of these functions are only optimized in certain cases; if they are
4752 not optimized in a particular case, a call to the library function will
4757 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
4758 @option{-std=c99}), the functions
4759 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
4760 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
4761 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
4762 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked}, @code{fputs_unlocked},
4763 @code{gammaf}, @code{gammal}, @code{gamma}, @code{gettext},
4764 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
4765 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
4766 @code{mempcpy}, @code{pow10f}, @code{pow10l}, @code{pow10},
4767 @code{printf_unlocked}, @code{rindex}, @code{scalbf}, @code{scalbl},
4768 @code{scalb}, @code{signbit}, @code{signbitf}, @code{signbitl},
4769 @code{significandf}, @code{significandl}, @code{significand},
4770 @code{sincosf}, @code{sincosl}, @code{sincos}, @code{stpcpy},
4771 @code{strdup}, @code{strfmon}, @code{toascii}, @code{y0f}, @code{y0l},
4772 @code{y0}, @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
4774 may be handled as built-in functions.
4775 All these functions have corresponding versions
4776 prefixed with @code{__builtin_}, which may be used even in strict C89
4779 The ISO C99 functions
4780 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
4781 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
4782 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
4783 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
4784 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
4785 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
4786 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
4787 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
4788 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
4789 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
4790 @code{cimagl}, @code{cimag}, @code{conjf}, @code{conjl}, @code{conj},
4791 @code{copysignf}, @code{copysignl}, @code{copysign}, @code{cpowf},
4792 @code{cpowl}, @code{cpow}, @code{cprojf}, @code{cprojl}, @code{cproj},
4793 @code{crealf}, @code{creall}, @code{creal}, @code{csinf}, @code{csinhf},
4794 @code{csinhl}, @code{csinh}, @code{csinl}, @code{csin}, @code{csqrtf},
4795 @code{csqrtl}, @code{csqrt}, @code{ctanf}, @code{ctanhf}, @code{ctanhl},
4796 @code{ctanh}, @code{ctanl}, @code{ctan}, @code{erfcf}, @code{erfcl},
4797 @code{erfc}, @code{erff}, @code{erfl}, @code{erf}, @code{exp2f},
4798 @code{exp2l}, @code{exp2}, @code{expm1f}, @code{expm1l}, @code{expm1},
4799 @code{fdimf}, @code{fdiml}, @code{fdim}, @code{fmaf}, @code{fmal},
4800 @code{fmaxf}, @code{fmaxl}, @code{fmax}, @code{fma}, @code{fminf},
4801 @code{fminl}, @code{fmin}, @code{hypotf}, @code{hypotl}, @code{hypot},
4802 @code{ilogbf}, @code{ilogbl}, @code{ilogb}, @code{imaxabs},
4803 @code{isblank}, @code{iswblank}, @code{lgammaf}, @code{lgammal},
4804 @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
4805 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
4806 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
4807 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
4808 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
4809 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
4810 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
4811 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
4812 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
4813 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
4814 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
4815 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
4816 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
4817 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
4818 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
4819 are handled as built-in functions
4820 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
4822 There are also built-in versions of the ISO C99 functions
4823 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
4824 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
4825 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
4826 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
4827 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
4828 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
4829 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
4830 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
4831 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
4832 that are recognized in any mode since ISO C90 reserves these names for
4833 the purpose to which ISO C99 puts them. All these functions have
4834 corresponding versions prefixed with @code{__builtin_}.
4836 The ISO C94 functions
4837 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
4838 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
4839 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
4841 are handled as built-in functions
4842 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
4844 The ISO C90 functions
4845 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
4846 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
4847 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
4848 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
4849 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
4850 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
4851 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
4852 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
4853 @code{malloc}, @code{memcmp}, @code{memcpy}, @code{memset}, @code{modf},
4854 @code{pow}, @code{printf}, @code{putchar}, @code{puts}, @code{scanf},
4855 @code{sinh}, @code{sin}, @code{snprintf}, @code{sprintf}, @code{sqrt},
4856 @code{sscanf}, @code{strcat}, @code{strchr}, @code{strcmp},
4857 @code{strcpy}, @code{strcspn}, @code{strlen}, @code{strncat},
4858 @code{strncmp}, @code{strncpy}, @code{strpbrk}, @code{strrchr},
4859 @code{strspn}, @code{strstr}, @code{tanh}, @code{tan}, @code{vfprintf},
4860 @code{vprintf} and @code{vsprintf}
4861 are all recognized as built-in functions unless
4862 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
4863 is specified for an individual function). All of these functions have
4864 corresponding versions prefixed with @code{__builtin_}.
4866 GCC provides built-in versions of the ISO C99 floating point comparison
4867 macros that avoid raising exceptions for unordered operands. They have
4868 the same names as the standard macros ( @code{isgreater},
4869 @code{isgreaterequal}, @code{isless}, @code{islessequal},
4870 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
4871 prefixed. We intend for a library implementor to be able to simply
4872 @code{#define} each standard macro to its built-in equivalent.
4874 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
4876 You can use the built-in function @code{__builtin_types_compatible_p} to
4877 determine whether two types are the same.
4879 This built-in function returns 1 if the unqualified versions of the
4880 types @var{type1} and @var{type2} (which are types, not expressions) are
4881 compatible, 0 otherwise. The result of this built-in function can be
4882 used in integer constant expressions.
4884 This built-in function ignores top level qualifiers (e.g., @code{const},
4885 @code{volatile}). For example, @code{int} is equivalent to @code{const
4888 The type @code{int[]} and @code{int[5]} are compatible. On the other
4889 hand, @code{int} and @code{char *} are not compatible, even if the size
4890 of their types, on the particular architecture are the same. Also, the
4891 amount of pointer indirection is taken into account when determining
4892 similarity. Consequently, @code{short *} is not similar to
4893 @code{short **}. Furthermore, two types that are typedefed are
4894 considered compatible if their underlying types are compatible.
4896 An @code{enum} type is not considered to be compatible with another
4897 @code{enum} type even if both are compatible with the same integer
4898 type; this is what the C standard specifies.
4899 For example, @code{enum @{foo, bar@}} is not similar to
4900 @code{enum @{hot, dog@}}.
4902 You would typically use this function in code whose execution varies
4903 depending on the arguments' types. For example:
4909 if (__builtin_types_compatible_p (typeof (x), long double)) \
4910 tmp = foo_long_double (tmp); \
4911 else if (__builtin_types_compatible_p (typeof (x), double)) \
4912 tmp = foo_double (tmp); \
4913 else if (__builtin_types_compatible_p (typeof (x), float)) \
4914 tmp = foo_float (tmp); \
4921 @emph{Note:} This construct is only available for C.
4925 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
4927 You can use the built-in function @code{__builtin_choose_expr} to
4928 evaluate code depending on the value of a constant expression. This
4929 built-in function returns @var{exp1} if @var{const_exp}, which is a
4930 constant expression that must be able to be determined at compile time,
4931 is nonzero. Otherwise it returns 0.
4933 This built-in function is analogous to the @samp{? :} operator in C,
4934 except that the expression returned has its type unaltered by promotion
4935 rules. Also, the built-in function does not evaluate the expression
4936 that was not chosen. For example, if @var{const_exp} evaluates to true,
4937 @var{exp2} is not evaluated even if it has side-effects.
4939 This built-in function can return an lvalue if the chosen argument is an
4942 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
4943 type. Similarly, if @var{exp2} is returned, its return type is the same
4950 __builtin_choose_expr ( \
4951 __builtin_types_compatible_p (typeof (x), double), \
4953 __builtin_choose_expr ( \
4954 __builtin_types_compatible_p (typeof (x), float), \
4956 /* @r{The void expression results in a compile-time error} \
4957 @r{when assigning the result to something.} */ \
4961 @emph{Note:} This construct is only available for C. Furthermore, the
4962 unused expression (@var{exp1} or @var{exp2} depending on the value of
4963 @var{const_exp}) may still generate syntax errors. This may change in
4968 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
4969 You can use the built-in function @code{__builtin_constant_p} to
4970 determine if a value is known to be constant at compile-time and hence
4971 that GCC can perform constant-folding on expressions involving that
4972 value. The argument of the function is the value to test. The function
4973 returns the integer 1 if the argument is known to be a compile-time
4974 constant and 0 if it is not known to be a compile-time constant. A
4975 return of 0 does not indicate that the value is @emph{not} a constant,
4976 but merely that GCC cannot prove it is a constant with the specified
4977 value of the @option{-O} option.
4979 You would typically use this function in an embedded application where
4980 memory was a critical resource. If you have some complex calculation,
4981 you may want it to be folded if it involves constants, but need to call
4982 a function if it does not. For example:
4985 #define Scale_Value(X) \
4986 (__builtin_constant_p (X) \
4987 ? ((X) * SCALE + OFFSET) : Scale (X))
4990 You may use this built-in function in either a macro or an inline
4991 function. However, if you use it in an inlined function and pass an
4992 argument of the function as the argument to the built-in, GCC will
4993 never return 1 when you call the inline function with a string constant
4994 or compound literal (@pxref{Compound Literals}) and will not return 1
4995 when you pass a constant numeric value to the inline function unless you
4996 specify the @option{-O} option.
4998 You may also use @code{__builtin_constant_p} in initializers for static
4999 data. For instance, you can write
5002 static const int table[] = @{
5003 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
5009 This is an acceptable initializer even if @var{EXPRESSION} is not a
5010 constant expression. GCC must be more conservative about evaluating the
5011 built-in in this case, because it has no opportunity to perform
5014 Previous versions of GCC did not accept this built-in in data
5015 initializers. The earliest version where it is completely safe is
5019 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
5020 @opindex fprofile-arcs
5021 You may use @code{__builtin_expect} to provide the compiler with
5022 branch prediction information. In general, you should prefer to
5023 use actual profile feedback for this (@option{-fprofile-arcs}), as
5024 programmers are notoriously bad at predicting how their programs
5025 actually perform. However, there are applications in which this
5026 data is hard to collect.
5028 The return value is the value of @var{exp}, which should be an
5029 integral expression. The value of @var{c} must be a compile-time
5030 constant. The semantics of the built-in are that it is expected
5031 that @var{exp} == @var{c}. For example:
5034 if (__builtin_expect (x, 0))
5039 would indicate that we do not expect to call @code{foo}, since
5040 we expect @code{x} to be zero. Since you are limited to integral
5041 expressions for @var{exp}, you should use constructions such as
5044 if (__builtin_expect (ptr != NULL, 1))
5049 when testing pointer or floating-point values.
5052 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
5053 This function is used to minimize cache-miss latency by moving data into
5054 a cache before it is accessed.
5055 You can insert calls to @code{__builtin_prefetch} into code for which
5056 you know addresses of data in memory that is likely to be accessed soon.
5057 If the target supports them, data prefetch instructions will be generated.
5058 If the prefetch is done early enough before the access then the data will
5059 be in the cache by the time it is accessed.
5061 The value of @var{addr} is the address of the memory to prefetch.
5062 There are two optional arguments, @var{rw} and @var{locality}.
5063 The value of @var{rw} is a compile-time constant one or zero; one
5064 means that the prefetch is preparing for a write to the memory address
5065 and zero, the default, means that the prefetch is preparing for a read.
5066 The value @var{locality} must be a compile-time constant integer between
5067 zero and three. A value of zero means that the data has no temporal
5068 locality, so it need not be left in the cache after the access. A value
5069 of three means that the data has a high degree of temporal locality and
5070 should be left in all levels of cache possible. Values of one and two
5071 mean, respectively, a low or moderate degree of temporal locality. The
5075 for (i = 0; i < n; i++)
5078 __builtin_prefetch (&a[i+j], 1, 1);
5079 __builtin_prefetch (&b[i+j], 0, 1);
5084 Data prefetch does not generate faults if @var{addr} is invalid, but
5085 the address expression itself must be valid. For example, a prefetch
5086 of @code{p->next} will not fault if @code{p->next} is not a valid
5087 address, but evaluation will fault if @code{p} is not a valid address.
5089 If the target does not support data prefetch, the address expression
5090 is evaluated if it includes side effects but no other code is generated
5091 and GCC does not issue a warning.
5094 @deftypefn {Built-in Function} double __builtin_huge_val (void)
5095 Returns a positive infinity, if supported by the floating-point format,
5096 else @code{DBL_MAX}. This function is suitable for implementing the
5097 ISO C macro @code{HUGE_VAL}.
5100 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
5101 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
5104 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
5105 Similar to @code{__builtin_huge_val}, except the return
5106 type is @code{long double}.
5109 @deftypefn {Built-in Function} double __builtin_inf (void)
5110 Similar to @code{__builtin_huge_val}, except a warning is generated
5111 if the target floating-point format does not support infinities.
5112 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
5115 @deftypefn {Built-in Function} float __builtin_inff (void)
5116 Similar to @code{__builtin_inf}, except the return type is @code{float}.
5119 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
5120 Similar to @code{__builtin_inf}, except the return
5121 type is @code{long double}.
5124 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
5125 This is an implementation of the ISO C99 function @code{nan}.
5127 Since ISO C99 defines this function in terms of @code{strtod}, which we
5128 do not implement, a description of the parsing is in order. The string
5129 is parsed as by @code{strtol}; that is, the base is recognized by
5130 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
5131 in the significand such that the least significant bit of the number
5132 is at the least significant bit of the significand. The number is
5133 truncated to fit the significand field provided. The significand is
5134 forced to be a quiet NaN.
5136 This function, if given a string literal, is evaluated early enough
5137 that it is considered a compile-time constant.
5140 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
5141 Similar to @code{__builtin_nan}, except the return type is @code{float}.
5144 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
5145 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
5148 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
5149 Similar to @code{__builtin_nan}, except the significand is forced
5150 to be a signaling NaN. The @code{nans} function is proposed by
5151 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
5154 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
5155 Similar to @code{__builtin_nans}, except the return type is @code{float}.
5158 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
5159 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
5162 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
5163 Returns one plus the index of the least significant 1-bit of @var{x}, or
5164 if @var{x} is zero, returns zero.
5167 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
5168 Returns the number of leading 0-bits in @var{x}, starting at the most
5169 significant bit position. If @var{x} is 0, the result is undefined.
5172 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
5173 Returns the number of trailing 0-bits in @var{x}, starting at the least
5174 significant bit position. If @var{x} is 0, the result is undefined.
5177 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
5178 Returns the number of 1-bits in @var{x}.
5181 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
5182 Returns the parity of @var{x}, i.@:e. the number of 1-bits in @var{x}
5186 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
5187 Similar to @code{__builtin_ffs}, except the argument type is
5188 @code{unsigned long}.
5191 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
5192 Similar to @code{__builtin_clz}, except the argument type is
5193 @code{unsigned long}.
5196 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
5197 Similar to @code{__builtin_ctz}, except the argument type is
5198 @code{unsigned long}.
5201 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
5202 Similar to @code{__builtin_popcount}, except the argument type is
5203 @code{unsigned long}.
5206 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
5207 Similar to @code{__builtin_parity}, except the argument type is
5208 @code{unsigned long}.
5211 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
5212 Similar to @code{__builtin_ffs}, except the argument type is
5213 @code{unsigned long long}.
5216 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
5217 Similar to @code{__builtin_clz}, except the argument type is
5218 @code{unsigned long long}.
5221 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
5222 Similar to @code{__builtin_ctz}, except the argument type is
5223 @code{unsigned long long}.
5226 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
5227 Similar to @code{__builtin_popcount}, except the argument type is
5228 @code{unsigned long long}.
5231 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
5232 Similar to @code{__builtin_parity}, except the argument type is
5233 @code{unsigned long long}.
5237 @node Target Builtins
5238 @section Built-in Functions Specific to Particular Target Machines
5240 On some target machines, GCC supports many built-in functions specific
5241 to those machines. Generally these generate calls to specific machine
5242 instructions, but allow the compiler to schedule those calls.
5245 * Alpha Built-in Functions::
5246 * ARM Built-in Functions::
5247 * X86 Built-in Functions::
5248 * PowerPC AltiVec Built-in Functions::
5251 @node Alpha Built-in Functions
5252 @subsection Alpha Built-in Functions
5254 These built-in functions are available for the Alpha family of
5255 processors, depending on the command-line switches used.
5257 The following built-in functions are always available. They
5258 all generate the machine instruction that is part of the name.
5261 long __builtin_alpha_implver (void)
5262 long __builtin_alpha_rpcc (void)
5263 long __builtin_alpha_amask (long)
5264 long __builtin_alpha_cmpbge (long, long)
5265 long __builtin_alpha_extbl (long, long)
5266 long __builtin_alpha_extwl (long, long)
5267 long __builtin_alpha_extll (long, long)
5268 long __builtin_alpha_extql (long, long)
5269 long __builtin_alpha_extwh (long, long)
5270 long __builtin_alpha_extlh (long, long)
5271 long __builtin_alpha_extqh (long, long)
5272 long __builtin_alpha_insbl (long, long)
5273 long __builtin_alpha_inswl (long, long)
5274 long __builtin_alpha_insll (long, long)
5275 long __builtin_alpha_insql (long, long)
5276 long __builtin_alpha_inswh (long, long)
5277 long __builtin_alpha_inslh (long, long)
5278 long __builtin_alpha_insqh (long, long)
5279 long __builtin_alpha_mskbl (long, long)
5280 long __builtin_alpha_mskwl (long, long)
5281 long __builtin_alpha_mskll (long, long)
5282 long __builtin_alpha_mskql (long, long)
5283 long __builtin_alpha_mskwh (long, long)
5284 long __builtin_alpha_msklh (long, long)
5285 long __builtin_alpha_mskqh (long, long)
5286 long __builtin_alpha_umulh (long, long)
5287 long __builtin_alpha_zap (long, long)
5288 long __builtin_alpha_zapnot (long, long)
5291 The following built-in functions are always with @option{-mmax}
5292 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
5293 later. They all generate the machine instruction that is part
5297 long __builtin_alpha_pklb (long)
5298 long __builtin_alpha_pkwb (long)
5299 long __builtin_alpha_unpkbl (long)
5300 long __builtin_alpha_unpkbw (long)
5301 long __builtin_alpha_minub8 (long, long)
5302 long __builtin_alpha_minsb8 (long, long)
5303 long __builtin_alpha_minuw4 (long, long)
5304 long __builtin_alpha_minsw4 (long, long)
5305 long __builtin_alpha_maxub8 (long, long)
5306 long __builtin_alpha_maxsb8 (long, long)
5307 long __builtin_alpha_maxuw4 (long, long)
5308 long __builtin_alpha_maxsw4 (long, long)
5309 long __builtin_alpha_perr (long, long)
5312 The following built-in functions are always with @option{-mcix}
5313 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
5314 later. They all generate the machine instruction that is part
5318 long __builtin_alpha_cttz (long)
5319 long __builtin_alpha_ctlz (long)
5320 long __builtin_alpha_ctpop (long)
5323 The following builtins are available on systems that use the OSF/1
5324 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
5325 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
5326 @code{rdval} and @code{wrval}.
5329 void *__builtin_thread_pointer (void)
5330 void __builtin_set_thread_pointer (void *)
5333 @node ARM Built-in Functions
5334 @subsection ARM Built-in Functions
5336 These built-in functions are available for the ARM family of
5337 processors, when the @option{-mcpu=iwmmxt} switch is used:
5340 typedef int v2si __attribute__ ((vector_size (8)));
5341 typedef short v4hi __attribute__ ((vector_size (8)));
5342 typedef char v8qi __attribute__ ((vector_size (8)));
5344 int __builtin_arm_getwcx (int)
5345 void __builtin_arm_setwcx (int, int)
5346 int __builtin_arm_textrmsb (v8qi, int)
5347 int __builtin_arm_textrmsh (v4hi, int)
5348 int __builtin_arm_textrmsw (v2si, int)
5349 int __builtin_arm_textrmub (v8qi, int)
5350 int __builtin_arm_textrmuh (v4hi, int)
5351 int __builtin_arm_textrmuw (v2si, int)
5352 v8qi __builtin_arm_tinsrb (v8qi, int)
5353 v4hi __builtin_arm_tinsrh (v4hi, int)
5354 v2si __builtin_arm_tinsrw (v2si, int)
5355 long long __builtin_arm_tmia (long long, int, int)
5356 long long __builtin_arm_tmiabb (long long, int, int)
5357 long long __builtin_arm_tmiabt (long long, int, int)
5358 long long __builtin_arm_tmiaph (long long, int, int)
5359 long long __builtin_arm_tmiatb (long long, int, int)
5360 long long __builtin_arm_tmiatt (long long, int, int)
5361 int __builtin_arm_tmovmskb (v8qi)
5362 int __builtin_arm_tmovmskh (v4hi)
5363 int __builtin_arm_tmovmskw (v2si)
5364 long long __builtin_arm_waccb (v8qi)
5365 long long __builtin_arm_wacch (v4hi)
5366 long long __builtin_arm_waccw (v2si)
5367 v8qi __builtin_arm_waddb (v8qi, v8qi)
5368 v8qi __builtin_arm_waddbss (v8qi, v8qi)
5369 v8qi __builtin_arm_waddbus (v8qi, v8qi)
5370 v4hi __builtin_arm_waddh (v4hi, v4hi)
5371 v4hi __builtin_arm_waddhss (v4hi, v4hi)
5372 v4hi __builtin_arm_waddhus (v4hi, v4hi)
5373 v2si __builtin_arm_waddw (v2si, v2si)
5374 v2si __builtin_arm_waddwss (v2si, v2si)
5375 v2si __builtin_arm_waddwus (v2si, v2si)
5376 v8qi __builtin_arm_walign (v8qi, v8qi, int)
5377 long long __builtin_arm_wand(long long, long long)
5378 long long __builtin_arm_wandn (long long, long long)
5379 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
5380 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
5381 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
5382 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
5383 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
5384 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
5385 v2si __builtin_arm_wcmpeqw (v2si, v2si)
5386 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
5387 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
5388 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
5389 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
5390 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
5391 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
5392 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
5393 long long __builtin_arm_wmacsz (v4hi, v4hi)
5394 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
5395 long long __builtin_arm_wmacuz (v4hi, v4hi)
5396 v4hi __builtin_arm_wmadds (v4hi, v4hi)
5397 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
5398 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
5399 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
5400 v2si __builtin_arm_wmaxsw (v2si, v2si)
5401 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
5402 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
5403 v2si __builtin_arm_wmaxuw (v2si, v2si)
5404 v8qi __builtin_arm_wminsb (v8qi, v8qi)
5405 v4hi __builtin_arm_wminsh (v4hi, v4hi)
5406 v2si __builtin_arm_wminsw (v2si, v2si)
5407 v8qi __builtin_arm_wminub (v8qi, v8qi)
5408 v4hi __builtin_arm_wminuh (v4hi, v4hi)
5409 v2si __builtin_arm_wminuw (v2si, v2si)
5410 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
5411 v4hi __builtin_arm_wmulul (v4hi, v4hi)
5412 v4hi __builtin_arm_wmulum (v4hi, v4hi)
5413 long long __builtin_arm_wor (long long, long long)
5414 v2si __builtin_arm_wpackdss (long long, long long)
5415 v2si __builtin_arm_wpackdus (long long, long long)
5416 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
5417 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
5418 v4hi __builtin_arm_wpackwss (v2si, v2si)
5419 v4hi __builtin_arm_wpackwus (v2si, v2si)
5420 long long __builtin_arm_wrord (long long, long long)
5421 long long __builtin_arm_wrordi (long long, int)
5422 v4hi __builtin_arm_wrorh (v4hi, long long)
5423 v4hi __builtin_arm_wrorhi (v4hi, int)
5424 v2si __builtin_arm_wrorw (v2si, long long)
5425 v2si __builtin_arm_wrorwi (v2si, int)
5426 v2si __builtin_arm_wsadb (v8qi, v8qi)
5427 v2si __builtin_arm_wsadbz (v8qi, v8qi)
5428 v2si __builtin_arm_wsadh (v4hi, v4hi)
5429 v2si __builtin_arm_wsadhz (v4hi, v4hi)
5430 v4hi __builtin_arm_wshufh (v4hi, int)
5431 long long __builtin_arm_wslld (long long, long long)
5432 long long __builtin_arm_wslldi (long long, int)
5433 v4hi __builtin_arm_wsllh (v4hi, long long)
5434 v4hi __builtin_arm_wsllhi (v4hi, int)
5435 v2si __builtin_arm_wsllw (v2si, long long)
5436 v2si __builtin_arm_wsllwi (v2si, int)
5437 long long __builtin_arm_wsrad (long long, long long)
5438 long long __builtin_arm_wsradi (long long, int)
5439 v4hi __builtin_arm_wsrah (v4hi, long long)
5440 v4hi __builtin_arm_wsrahi (v4hi, int)
5441 v2si __builtin_arm_wsraw (v2si, long long)
5442 v2si __builtin_arm_wsrawi (v2si, int)
5443 long long __builtin_arm_wsrld (long long, long long)
5444 long long __builtin_arm_wsrldi (long long, int)
5445 v4hi __builtin_arm_wsrlh (v4hi, long long)
5446 v4hi __builtin_arm_wsrlhi (v4hi, int)
5447 v2si __builtin_arm_wsrlw (v2si, long long)
5448 v2si __builtin_arm_wsrlwi (v2si, int)
5449 v8qi __builtin_arm_wsubb (v8qi, v8qi)
5450 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
5451 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
5452 v4hi __builtin_arm_wsubh (v4hi, v4hi)
5453 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
5454 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
5455 v2si __builtin_arm_wsubw (v2si, v2si)
5456 v2si __builtin_arm_wsubwss (v2si, v2si)
5457 v2si __builtin_arm_wsubwus (v2si, v2si)
5458 v4hi __builtin_arm_wunpckehsb (v8qi)
5459 v2si __builtin_arm_wunpckehsh (v4hi)
5460 long long __builtin_arm_wunpckehsw (v2si)
5461 v4hi __builtin_arm_wunpckehub (v8qi)
5462 v2si __builtin_arm_wunpckehuh (v4hi)
5463 long long __builtin_arm_wunpckehuw (v2si)
5464 v4hi __builtin_arm_wunpckelsb (v8qi)
5465 v2si __builtin_arm_wunpckelsh (v4hi)
5466 long long __builtin_arm_wunpckelsw (v2si)
5467 v4hi __builtin_arm_wunpckelub (v8qi)
5468 v2si __builtin_arm_wunpckeluh (v4hi)
5469 long long __builtin_arm_wunpckeluw (v2si)
5470 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
5471 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
5472 v2si __builtin_arm_wunpckihw (v2si, v2si)
5473 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
5474 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
5475 v2si __builtin_arm_wunpckilw (v2si, v2si)
5476 long long __builtin_arm_wxor (long long, long long)
5477 long long __builtin_arm_wzero ()
5480 @node X86 Built-in Functions
5481 @subsection X86 Built-in Functions
5483 These built-in functions are available for the i386 and x86-64 family
5484 of computers, depending on the command-line switches used.
5486 The following machine modes are available for use with MMX built-in functions
5487 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
5488 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
5489 vector of eight 8-bit integers. Some of the built-in functions operate on
5490 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
5492 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
5493 of two 32-bit floating point values.
5495 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
5496 floating point values. Some instructions use a vector of four 32-bit
5497 integers, these use @code{V4SI}. Finally, some instructions operate on an
5498 entire vector register, interpreting it as a 128-bit integer, these use mode
5501 The following built-in functions are made available by @option{-mmmx}.
5502 All of them generate the machine instruction that is part of the name.
5505 v8qi __builtin_ia32_paddb (v8qi, v8qi)
5506 v4hi __builtin_ia32_paddw (v4hi, v4hi)
5507 v2si __builtin_ia32_paddd (v2si, v2si)
5508 v8qi __builtin_ia32_psubb (v8qi, v8qi)
5509 v4hi __builtin_ia32_psubw (v4hi, v4hi)
5510 v2si __builtin_ia32_psubd (v2si, v2si)
5511 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
5512 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
5513 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
5514 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
5515 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
5516 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
5517 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
5518 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
5519 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
5520 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
5521 di __builtin_ia32_pand (di, di)
5522 di __builtin_ia32_pandn (di,di)
5523 di __builtin_ia32_por (di, di)
5524 di __builtin_ia32_pxor (di, di)
5525 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
5526 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
5527 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
5528 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
5529 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
5530 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
5531 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
5532 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
5533 v2si __builtin_ia32_punpckhdq (v2si, v2si)
5534 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
5535 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
5536 v2si __builtin_ia32_punpckldq (v2si, v2si)
5537 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
5538 v4hi __builtin_ia32_packssdw (v2si, v2si)
5539 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
5542 The following built-in functions are made available either with
5543 @option{-msse}, or with a combination of @option{-m3dnow} and
5544 @option{-march=athlon}. All of them generate the machine
5545 instruction that is part of the name.
5548 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
5549 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
5550 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
5551 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
5552 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
5553 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
5554 v8qi __builtin_ia32_pminub (v8qi, v8qi)
5555 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
5556 int __builtin_ia32_pextrw (v4hi, int)
5557 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
5558 int __builtin_ia32_pmovmskb (v8qi)
5559 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
5560 void __builtin_ia32_movntq (di *, di)
5561 void __builtin_ia32_sfence (void)
5564 The following built-in functions are available when @option{-msse} is used.
5565 All of them generate the machine instruction that is part of the name.
5568 int __builtin_ia32_comieq (v4sf, v4sf)
5569 int __builtin_ia32_comineq (v4sf, v4sf)
5570 int __builtin_ia32_comilt (v4sf, v4sf)
5571 int __builtin_ia32_comile (v4sf, v4sf)
5572 int __builtin_ia32_comigt (v4sf, v4sf)
5573 int __builtin_ia32_comige (v4sf, v4sf)
5574 int __builtin_ia32_ucomieq (v4sf, v4sf)
5575 int __builtin_ia32_ucomineq (v4sf, v4sf)
5576 int __builtin_ia32_ucomilt (v4sf, v4sf)
5577 int __builtin_ia32_ucomile (v4sf, v4sf)
5578 int __builtin_ia32_ucomigt (v4sf, v4sf)
5579 int __builtin_ia32_ucomige (v4sf, v4sf)
5580 v4sf __builtin_ia32_addps (v4sf, v4sf)
5581 v4sf __builtin_ia32_subps (v4sf, v4sf)
5582 v4sf __builtin_ia32_mulps (v4sf, v4sf)
5583 v4sf __builtin_ia32_divps (v4sf, v4sf)
5584 v4sf __builtin_ia32_addss (v4sf, v4sf)
5585 v4sf __builtin_ia32_subss (v4sf, v4sf)
5586 v4sf __builtin_ia32_mulss (v4sf, v4sf)
5587 v4sf __builtin_ia32_divss (v4sf, v4sf)
5588 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
5589 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
5590 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
5591 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
5592 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
5593 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
5594 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
5595 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
5596 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
5597 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
5598 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
5599 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
5600 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
5601 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
5602 v4si __builtin_ia32_cmpless (v4sf, v4sf)
5603 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
5604 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
5605 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
5606 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
5607 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
5608 v4sf __builtin_ia32_maxps (v4sf, v4sf)
5609 v4sf __builtin_ia32_maxss (v4sf, v4sf)
5610 v4sf __builtin_ia32_minps (v4sf, v4sf)
5611 v4sf __builtin_ia32_minss (v4sf, v4sf)
5612 v4sf __builtin_ia32_andps (v4sf, v4sf)
5613 v4sf __builtin_ia32_andnps (v4sf, v4sf)
5614 v4sf __builtin_ia32_orps (v4sf, v4sf)
5615 v4sf __builtin_ia32_xorps (v4sf, v4sf)
5616 v4sf __builtin_ia32_movss (v4sf, v4sf)
5617 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
5618 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
5619 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
5620 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
5621 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
5622 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
5623 v2si __builtin_ia32_cvtps2pi (v4sf)
5624 int __builtin_ia32_cvtss2si (v4sf)
5625 v2si __builtin_ia32_cvttps2pi (v4sf)
5626 int __builtin_ia32_cvttss2si (v4sf)
5627 v4sf __builtin_ia32_rcpps (v4sf)
5628 v4sf __builtin_ia32_rsqrtps (v4sf)
5629 v4sf __builtin_ia32_sqrtps (v4sf)
5630 v4sf __builtin_ia32_rcpss (v4sf)
5631 v4sf __builtin_ia32_rsqrtss (v4sf)
5632 v4sf __builtin_ia32_sqrtss (v4sf)
5633 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
5634 void __builtin_ia32_movntps (float *, v4sf)
5635 int __builtin_ia32_movmskps (v4sf)
5638 The following built-in functions are available when @option{-msse} is used.
5641 @item v4sf __builtin_ia32_loadaps (float *)
5642 Generates the @code{movaps} machine instruction as a load from memory.
5643 @item void __builtin_ia32_storeaps (float *, v4sf)
5644 Generates the @code{movaps} machine instruction as a store to memory.
5645 @item v4sf __builtin_ia32_loadups (float *)
5646 Generates the @code{movups} machine instruction as a load from memory.
5647 @item void __builtin_ia32_storeups (float *, v4sf)
5648 Generates the @code{movups} machine instruction as a store to memory.
5649 @item v4sf __builtin_ia32_loadsss (float *)
5650 Generates the @code{movss} machine instruction as a load from memory.
5651 @item void __builtin_ia32_storess (float *, v4sf)
5652 Generates the @code{movss} machine instruction as a store to memory.
5653 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
5654 Generates the @code{movhps} machine instruction as a load from memory.
5655 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
5656 Generates the @code{movlps} machine instruction as a load from memory
5657 @item void __builtin_ia32_storehps (v4sf, v2si *)
5658 Generates the @code{movhps} machine instruction as a store to memory.
5659 @item void __builtin_ia32_storelps (v4sf, v2si *)
5660 Generates the @code{movlps} machine instruction as a store to memory.
5663 The following built-in functions are available when @option{-msse3} is used.
5664 All of them generate the machine instruction that is part of the name.
5667 v2df __builtin_ia32_addsubpd (v2df, v2df)
5668 v2df __builtin_ia32_addsubps (v2df, v2df)
5669 v2df __builtin_ia32_haddpd (v2df, v2df)
5670 v2df __builtin_ia32_haddps (v2df, v2df)
5671 v2df __builtin_ia32_hsubpd (v2df, v2df)
5672 v2df __builtin_ia32_hsubps (v2df, v2df)
5673 v16qi __builtin_ia32_lddqu (char const *)
5674 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
5675 v2df __builtin_ia32_movddup (v2df)
5676 v4sf __builtin_ia32_movshdup (v4sf)
5677 v4sf __builtin_ia32_movsldup (v4sf)
5678 void __builtin_ia32_mwait (unsigned int, unsigned int)
5681 The following built-in functions are available when @option{-msse3} is used.
5684 @item v2df __builtin_ia32_loadddup (double const *)
5685 Generates the @code{movddup} machine instruction as a load from memory.
5688 The following built-in functions are available when @option{-m3dnow} is used.
5689 All of them generate the machine instruction that is part of the name.
5692 void __builtin_ia32_femms (void)
5693 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
5694 v2si __builtin_ia32_pf2id (v2sf)
5695 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
5696 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
5697 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
5698 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
5699 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
5700 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
5701 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
5702 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
5703 v2sf __builtin_ia32_pfrcp (v2sf)
5704 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
5705 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
5706 v2sf __builtin_ia32_pfrsqrt (v2sf)
5707 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
5708 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
5709 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
5710 v2sf __builtin_ia32_pi2fd (v2si)
5711 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
5714 The following built-in functions are available when both @option{-m3dnow}
5715 and @option{-march=athlon} are used. All of them generate the machine
5716 instruction that is part of the name.
5719 v2si __builtin_ia32_pf2iw (v2sf)
5720 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
5721 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
5722 v2sf __builtin_ia32_pi2fw (v2si)
5723 v2sf __builtin_ia32_pswapdsf (v2sf)
5724 v2si __builtin_ia32_pswapdsi (v2si)
5727 @node PowerPC AltiVec Built-in Functions
5728 @subsection PowerPC AltiVec Built-in Functions
5730 These built-in functions are available for the PowerPC family
5731 of computers, depending on the command-line switches used.
5733 The following machine modes are available for use with AltiVec built-in
5734 functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four
5735 32-bit integers, @code{V4SF} for a vector of four 32-bit floating point
5736 numbers, @code{V8HI} for a vector of eight 16-bit integers, and
5737 @code{V16QI} for a vector of sixteen 8-bit integers.
5739 The following functions are made available by including
5740 @code{<altivec.h>} and using @option{-maltivec} and
5741 @option{-mabi=altivec}. The functions implement the functionality
5742 described in Motorola's AltiVec Programming Interface Manual.
5744 There are a few differences from Motorola's documentation and GCC's
5745 implementation. Vector constants are done with curly braces (not
5746 parentheses). Vector initializers require no casts if the vector
5747 constant is of the same type as the variable it is initializing. The
5748 @code{vector bool} type is deprecated and will be discontinued in
5749 further revisions. Use @code{vector signed} instead. If @code{signed}
5750 or @code{unsigned} is omitted, the vector type will default to
5751 @code{signed}. Lastly, all overloaded functions are implemented with macros
5752 for the C implementation. So code the following example will not work:
5755 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
5758 Since vec_add is a macro, the vector constant in the above example will
5759 be treated as four different arguments. Wrap the entire argument in
5760 parentheses for this to work. The C++ implementation does not use
5763 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
5764 Internally, GCC uses built-in functions to achieve the functionality in
5765 the aforementioned header file, but they are not supported and are
5766 subject to change without notice.
5769 vector signed char vec_abs (vector signed char, vector signed char);
5770 vector signed short vec_abs (vector signed short, vector signed short);
5771 vector signed int vec_abs (vector signed int, vector signed int);
5772 vector signed float vec_abs (vector signed float, vector signed float);
5774 vector signed char vec_abss (vector signed char, vector signed char);
5775 vector signed short vec_abss (vector signed short, vector signed short);
5777 vector signed char vec_add (vector signed char, vector signed char);
5778 vector unsigned char vec_add (vector signed char, vector unsigned char);
5780 vector unsigned char vec_add (vector unsigned char, vector signed char);
5782 vector unsigned char vec_add (vector unsigned char,
5783 vector unsigned char);
5784 vector signed short vec_add (vector signed short, vector signed short);
5785 vector unsigned short vec_add (vector signed short,
5786 vector unsigned short);
5787 vector unsigned short vec_add (vector unsigned short,
5788 vector signed short);
5789 vector unsigned short vec_add (vector unsigned short,
5790 vector unsigned short);
5791 vector signed int vec_add (vector signed int, vector signed int);
5792 vector unsigned int vec_add (vector signed int, vector unsigned int);
5793 vector unsigned int vec_add (vector unsigned int, vector signed int);
5794 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
5795 vector float vec_add (vector float, vector float);
5797 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
5799 vector unsigned char vec_adds (vector signed char,
5800 vector unsigned char);
5801 vector unsigned char vec_adds (vector unsigned char,
5802 vector signed char);
5803 vector unsigned char vec_adds (vector unsigned char,
5804 vector unsigned char);
5805 vector signed char vec_adds (vector signed char, vector signed char);
5806 vector unsigned short vec_adds (vector signed short,
5807 vector unsigned short);
5808 vector unsigned short vec_adds (vector unsigned short,
5809 vector signed short);
5810 vector unsigned short vec_adds (vector unsigned short,
5811 vector unsigned short);
5812 vector signed short vec_adds (vector signed short, vector signed short);
5814 vector unsigned int vec_adds (vector signed int, vector unsigned int);
5815 vector unsigned int vec_adds (vector unsigned int, vector signed int);
5816 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
5818 vector signed int vec_adds (vector signed int, vector signed int);
5820 vector float vec_and (vector float, vector float);
5821 vector float vec_and (vector float, vector signed int);
5822 vector float vec_and (vector signed int, vector float);
5823 vector signed int vec_and (vector signed int, vector signed int);
5824 vector unsigned int vec_and (vector signed int, vector unsigned int);
5825 vector unsigned int vec_and (vector unsigned int, vector signed int);
5826 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
5827 vector signed short vec_and (vector signed short, vector signed short);
5828 vector unsigned short vec_and (vector signed short,
5829 vector unsigned short);
5830 vector unsigned short vec_and (vector unsigned short,
5831 vector signed short);
5832 vector unsigned short vec_and (vector unsigned short,
5833 vector unsigned short);
5834 vector signed char vec_and (vector signed char, vector signed char);
5835 vector unsigned char vec_and (vector signed char, vector unsigned char);
5837 vector unsigned char vec_and (vector unsigned char, vector signed char);
5839 vector unsigned char vec_and (vector unsigned char,
5840 vector unsigned char);
5842 vector float vec_andc (vector float, vector float);
5843 vector float vec_andc (vector float, vector signed int);
5844 vector float vec_andc (vector signed int, vector float);
5845 vector signed int vec_andc (vector signed int, vector signed int);
5846 vector unsigned int vec_andc (vector signed int, vector unsigned int);
5847 vector unsigned int vec_andc (vector unsigned int, vector signed int);
5848 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
5850 vector signed short vec_andc (vector signed short, vector signed short);
5852 vector unsigned short vec_andc (vector signed short,
5853 vector unsigned short);
5854 vector unsigned short vec_andc (vector unsigned short,
5855 vector signed short);
5856 vector unsigned short vec_andc (vector unsigned short,
5857 vector unsigned short);
5858 vector signed char vec_andc (vector signed char, vector signed char);
5859 vector unsigned char vec_andc (vector signed char,
5860 vector unsigned char);
5861 vector unsigned char vec_andc (vector unsigned char,
5862 vector signed char);
5863 vector unsigned char vec_andc (vector unsigned char,
5864 vector unsigned char);
5866 vector unsigned char vec_avg (vector unsigned char,
5867 vector unsigned char);
5868 vector signed char vec_avg (vector signed char, vector signed char);
5869 vector unsigned short vec_avg (vector unsigned short,
5870 vector unsigned short);
5871 vector signed short vec_avg (vector signed short, vector signed short);
5872 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
5873 vector signed int vec_avg (vector signed int, vector signed int);
5875 vector float vec_ceil (vector float);
5877 vector signed int vec_cmpb (vector float, vector float);
5879 vector signed char vec_cmpeq (vector signed char, vector signed char);
5880 vector signed char vec_cmpeq (vector unsigned char,
5881 vector unsigned char);
5882 vector signed short vec_cmpeq (vector signed short,
5883 vector signed short);
5884 vector signed short vec_cmpeq (vector unsigned short,
5885 vector unsigned short);
5886 vector signed int vec_cmpeq (vector signed int, vector signed int);
5887 vector signed int vec_cmpeq (vector unsigned int, vector unsigned int);
5888 vector signed int vec_cmpeq (vector float, vector float);
5890 vector signed int vec_cmpge (vector float, vector float);
5892 vector signed char vec_cmpgt (vector unsigned char,
5893 vector unsigned char);
5894 vector signed char vec_cmpgt (vector signed char, vector signed char);
5895 vector signed short vec_cmpgt (vector unsigned short,
5896 vector unsigned short);
5897 vector signed short vec_cmpgt (vector signed short,
5898 vector signed short);
5899 vector signed int vec_cmpgt (vector unsigned int, vector unsigned int);
5900 vector signed int vec_cmpgt (vector signed int, vector signed int);
5901 vector signed int vec_cmpgt (vector float, vector float);
5903 vector signed int vec_cmple (vector float, vector float);
5905 vector signed char vec_cmplt (vector unsigned char,
5906 vector unsigned char);
5907 vector signed char vec_cmplt (vector signed char, vector signed char);
5908 vector signed short vec_cmplt (vector unsigned short,
5909 vector unsigned short);
5910 vector signed short vec_cmplt (vector signed short,
5911 vector signed short);
5912 vector signed int vec_cmplt (vector unsigned int, vector unsigned int);
5913 vector signed int vec_cmplt (vector signed int, vector signed int);
5914 vector signed int vec_cmplt (vector float, vector float);
5916 vector float vec_ctf (vector unsigned int, const char);
5917 vector float vec_ctf (vector signed int, const char);
5919 vector signed int vec_cts (vector float, const char);
5921 vector unsigned int vec_ctu (vector float, const char);
5923 void vec_dss (const char);
5925 void vec_dssall (void);
5927 void vec_dst (void *, int, const char);
5929 void vec_dstst (void *, int, const char);
5931 void vec_dststt (void *, int, const char);
5933 void vec_dstt (void *, int, const char);
5935 vector float vec_expte (vector float, vector float);
5937 vector float vec_floor (vector float, vector float);
5939 vector float vec_ld (int, vector float *);
5940 vector float vec_ld (int, float *):
5941 vector signed int vec_ld (int, int *);
5942 vector signed int vec_ld (int, vector signed int *);
5943 vector unsigned int vec_ld (int, vector unsigned int *);
5944 vector unsigned int vec_ld (int, unsigned int *);
5945 vector signed short vec_ld (int, short *, vector signed short *);
5946 vector unsigned short vec_ld (int, unsigned short *,
5947 vector unsigned short *);
5948 vector signed char vec_ld (int, signed char *);
5949 vector signed char vec_ld (int, vector signed char *);
5950 vector unsigned char vec_ld (int, unsigned char *);
5951 vector unsigned char vec_ld (int, vector unsigned char *);
5953 vector signed char vec_lde (int, signed char *);
5954 vector unsigned char vec_lde (int, unsigned char *);
5955 vector signed short vec_lde (int, short *);
5956 vector unsigned short vec_lde (int, unsigned short *);
5957 vector float vec_lde (int, float *);
5958 vector signed int vec_lde (int, int *);
5959 vector unsigned int vec_lde (int, unsigned int *);
5961 void float vec_ldl (int, float *);
5962 void float vec_ldl (int, vector float *);
5963 void signed int vec_ldl (int, vector signed int *);
5964 void signed int vec_ldl (int, int *);
5965 void unsigned int vec_ldl (int, unsigned int *);
5966 void unsigned int vec_ldl (int, vector unsigned int *);
5967 void signed short vec_ldl (int, vector signed short *);
5968 void signed short vec_ldl (int, short *);
5969 void unsigned short vec_ldl (int, vector unsigned short *);
5970 void unsigned short vec_ldl (int, unsigned short *);
5971 void signed char vec_ldl (int, vector signed char *);
5972 void signed char vec_ldl (int, signed char *);
5973 void unsigned char vec_ldl (int, vector unsigned char *);
5974 void unsigned char vec_ldl (int, unsigned char *);
5976 vector float vec_loge (vector float);
5978 vector unsigned char vec_lvsl (int, void *, int *);
5980 vector unsigned char vec_lvsr (int, void *, int *);
5982 vector float vec_madd (vector float, vector float, vector float);
5984 vector signed short vec_madds (vector signed short, vector signed short,
5985 vector signed short);
5987 vector unsigned char vec_max (vector signed char, vector unsigned char);
5989 vector unsigned char vec_max (vector unsigned char, vector signed char);
5991 vector unsigned char vec_max (vector unsigned char,
5992 vector unsigned char);
5993 vector signed char vec_max (vector signed char, vector signed char);
5994 vector unsigned short vec_max (vector signed short,
5995 vector unsigned short);
5996 vector unsigned short vec_max (vector unsigned short,
5997 vector signed short);
5998 vector unsigned short vec_max (vector unsigned short,
5999 vector unsigned short);
6000 vector signed short vec_max (vector signed short, vector signed short);
6001 vector unsigned int vec_max (vector signed int, vector unsigned int);
6002 vector unsigned int vec_max (vector unsigned int, vector signed int);
6003 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
6004 vector signed int vec_max (vector signed int, vector signed int);
6005 vector float vec_max (vector float, vector float);
6007 vector signed char vec_mergeh (vector signed char, vector signed char);
6008 vector unsigned char vec_mergeh (vector unsigned char,
6009 vector unsigned char);
6010 vector signed short vec_mergeh (vector signed short,
6011 vector signed short);
6012 vector unsigned short vec_mergeh (vector unsigned short,
6013 vector unsigned short);
6014 vector float vec_mergeh (vector float, vector float);
6015 vector signed int vec_mergeh (vector signed int, vector signed int);
6016 vector unsigned int vec_mergeh (vector unsigned int,
6017 vector unsigned int);
6019 vector signed char vec_mergel (vector signed char, vector signed char);
6020 vector unsigned char vec_mergel (vector unsigned char,
6021 vector unsigned char);
6022 vector signed short vec_mergel (vector signed short,
6023 vector signed short);
6024 vector unsigned short vec_mergel (vector unsigned short,
6025 vector unsigned short);
6026 vector float vec_mergel (vector float, vector float);
6027 vector signed int vec_mergel (vector signed int, vector signed int);
6028 vector unsigned int vec_mergel (vector unsigned int,
6029 vector unsigned int);
6031 vector unsigned short vec_mfvscr (void);
6033 vector unsigned char vec_min (vector signed char, vector unsigned char);
6035 vector unsigned char vec_min (vector unsigned char, vector signed char);
6037 vector unsigned char vec_min (vector unsigned char,
6038 vector unsigned char);
6039 vector signed char vec_min (vector signed char, vector signed char);
6040 vector unsigned short vec_min (vector signed short,
6041 vector unsigned short);
6042 vector unsigned short vec_min (vector unsigned short,
6043 vector signed short);
6044 vector unsigned short vec_min (vector unsigned short,
6045 vector unsigned short);
6046 vector signed short vec_min (vector signed short, vector signed short);
6047 vector unsigned int vec_min (vector signed int, vector unsigned int);
6048 vector unsigned int vec_min (vector unsigned int, vector signed int);
6049 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
6050 vector signed int vec_min (vector signed int, vector signed int);
6051 vector float vec_min (vector float, vector float);
6053 vector signed short vec_mladd (vector signed short, vector signed short,
6054 vector signed short);
6055 vector signed short vec_mladd (vector signed short,
6056 vector unsigned short,
6057 vector unsigned short);
6058 vector signed short vec_mladd (vector unsigned short,
6059 vector signed short,
6060 vector signed short);
6061 vector unsigned short vec_mladd (vector unsigned short,
6062 vector unsigned short,
6063 vector unsigned short);
6065 vector signed short vec_mradds (vector signed short,
6066 vector signed short,
6067 vector signed short);
6069 vector unsigned int vec_msum (vector unsigned char,
6070 vector unsigned char,
6071 vector unsigned int);
6072 vector signed int vec_msum (vector signed char, vector unsigned char,
6074 vector unsigned int vec_msum (vector unsigned short,
6075 vector unsigned short,
6076 vector unsigned int);
6077 vector signed int vec_msum (vector signed short, vector signed short,
6080 vector unsigned int vec_msums (vector unsigned short,
6081 vector unsigned short,
6082 vector unsigned int);
6083 vector signed int vec_msums (vector signed short, vector signed short,
6086 void vec_mtvscr (vector signed int);
6087 void vec_mtvscr (vector unsigned int);
6088 void vec_mtvscr (vector signed short);
6089 void vec_mtvscr (vector unsigned short);
6090 void vec_mtvscr (vector signed char);
6091 void vec_mtvscr (vector unsigned char);
6093 vector unsigned short vec_mule (vector unsigned char,
6094 vector unsigned char);
6095 vector signed short vec_mule (vector signed char, vector signed char);
6096 vector unsigned int vec_mule (vector unsigned short,
6097 vector unsigned short);
6098 vector signed int vec_mule (vector signed short, vector signed short);
6100 vector unsigned short vec_mulo (vector unsigned char,
6101 vector unsigned char);
6102 vector signed short vec_mulo (vector signed char, vector signed char);
6103 vector unsigned int vec_mulo (vector unsigned short,
6104 vector unsigned short);
6105 vector signed int vec_mulo (vector signed short, vector signed short);
6107 vector float vec_nmsub (vector float, vector float, vector float);
6109 vector float vec_nor (vector float, vector float);
6110 vector signed int vec_nor (vector signed int, vector signed int);
6111 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
6112 vector signed short vec_nor (vector signed short, vector signed short);
6113 vector unsigned short vec_nor (vector unsigned short,
6114 vector unsigned short);
6115 vector signed char vec_nor (vector signed char, vector signed char);
6116 vector unsigned char vec_nor (vector unsigned char,
6117 vector unsigned char);
6119 vector float vec_or (vector float, vector float);
6120 vector float vec_or (vector float, vector signed int);
6121 vector float vec_or (vector signed int, vector float);
6122 vector signed int vec_or (vector signed int, vector signed int);
6123 vector unsigned int vec_or (vector signed int, vector unsigned int);
6124 vector unsigned int vec_or (vector unsigned int, vector signed int);
6125 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
6126 vector signed short vec_or (vector signed short, vector signed short);
6127 vector unsigned short vec_or (vector signed short,
6128 vector unsigned short);
6129 vector unsigned short vec_or (vector unsigned short,
6130 vector signed short);
6131 vector unsigned short vec_or (vector unsigned short,
6132 vector unsigned short);
6133 vector signed char vec_or (vector signed char, vector signed char);
6134 vector unsigned char vec_or (vector signed char, vector unsigned char);
6135 vector unsigned char vec_or (vector unsigned char, vector signed char);
6136 vector unsigned char vec_or (vector unsigned char,
6137 vector unsigned char);
6139 vector signed char vec_pack (vector signed short, vector signed short);
6140 vector unsigned char vec_pack (vector unsigned short,
6141 vector unsigned short);
6142 vector signed short vec_pack (vector signed int, vector signed int);
6143 vector unsigned short vec_pack (vector unsigned int,
6144 vector unsigned int);
6146 vector signed short vec_packpx (vector unsigned int,
6147 vector unsigned int);
6149 vector unsigned char vec_packs (vector unsigned short,
6150 vector unsigned short);
6151 vector signed char vec_packs (vector signed short, vector signed short);
6153 vector unsigned short vec_packs (vector unsigned int,
6154 vector unsigned int);
6155 vector signed short vec_packs (vector signed int, vector signed int);
6157 vector unsigned char vec_packsu (vector unsigned short,
6158 vector unsigned short);
6159 vector unsigned char vec_packsu (vector signed short,
6160 vector signed short);
6161 vector unsigned short vec_packsu (vector unsigned int,
6162 vector unsigned int);
6163 vector unsigned short vec_packsu (vector signed int, vector signed int);
6165 vector float vec_perm (vector float, vector float,
6166 vector unsigned char);
6167 vector signed int vec_perm (vector signed int, vector signed int,
6168 vector unsigned char);
6169 vector unsigned int vec_perm (vector unsigned int, vector unsigned int,
6170 vector unsigned char);
6171 vector signed short vec_perm (vector signed short, vector signed short,
6172 vector unsigned char);
6173 vector unsigned short vec_perm (vector unsigned short,
6174 vector unsigned short,
6175 vector unsigned char);
6176 vector signed char vec_perm (vector signed char, vector signed char,
6177 vector unsigned char);
6178 vector unsigned char vec_perm (vector unsigned char,
6179 vector unsigned char,
6180 vector unsigned char);
6182 vector float vec_re (vector float);
6184 vector signed char vec_rl (vector signed char, vector unsigned char);
6185 vector unsigned char vec_rl (vector unsigned char,
6186 vector unsigned char);
6187 vector signed short vec_rl (vector signed short, vector unsigned short);
6189 vector unsigned short vec_rl (vector unsigned short,
6190 vector unsigned short);
6191 vector signed int vec_rl (vector signed int, vector unsigned int);
6192 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
6194 vector float vec_round (vector float);
6196 vector float vec_rsqrte (vector float);
6198 vector float vec_sel (vector float, vector float, vector signed int);
6199 vector float vec_sel (vector float, vector float, vector unsigned int);
6200 vector signed int vec_sel (vector signed int, vector signed int,
6202 vector signed int vec_sel (vector signed int, vector signed int,
6203 vector unsigned int);
6204 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
6206 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
6207 vector unsigned int);
6208 vector signed short vec_sel (vector signed short, vector signed short,
6209 vector signed short);
6210 vector signed short vec_sel (vector signed short, vector signed short,
6211 vector unsigned short);
6212 vector unsigned short vec_sel (vector unsigned short,
6213 vector unsigned short,
6214 vector signed short);
6215 vector unsigned short vec_sel (vector unsigned short,
6216 vector unsigned short,
6217 vector unsigned short);
6218 vector signed char vec_sel (vector signed char, vector signed char,
6219 vector signed char);
6220 vector signed char vec_sel (vector signed char, vector signed char,
6221 vector unsigned char);
6222 vector unsigned char vec_sel (vector unsigned char,
6223 vector unsigned char,
6224 vector signed char);
6225 vector unsigned char vec_sel (vector unsigned char,
6226 vector unsigned char,
6227 vector unsigned char);
6229 vector signed char vec_sl (vector signed char, vector unsigned char);
6230 vector unsigned char vec_sl (vector unsigned char,
6231 vector unsigned char);
6232 vector signed short vec_sl (vector signed short, vector unsigned short);
6234 vector unsigned short vec_sl (vector unsigned short,
6235 vector unsigned short);
6236 vector signed int vec_sl (vector signed int, vector unsigned int);
6237 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
6239 vector float vec_sld (vector float, vector float, const char);
6240 vector signed int vec_sld (vector signed int, vector signed int,
6242 vector unsigned int vec_sld (vector unsigned int, vector unsigned int,
6244 vector signed short vec_sld (vector signed short, vector signed short,
6246 vector unsigned short vec_sld (vector unsigned short,
6247 vector unsigned short, const char);
6248 vector signed char vec_sld (vector signed char, vector signed char,
6250 vector unsigned char vec_sld (vector unsigned char,
6251 vector unsigned char,
6254 vector signed int vec_sll (vector signed int, vector unsigned int);
6255 vector signed int vec_sll (vector signed int, vector unsigned short);
6256 vector signed int vec_sll (vector signed int, vector unsigned char);
6257 vector unsigned int vec_sll (vector unsigned int, vector unsigned int);
6258 vector unsigned int vec_sll (vector unsigned int,
6259 vector unsigned short);
6260 vector unsigned int vec_sll (vector unsigned int, vector unsigned char);
6262 vector signed short vec_sll (vector signed short, vector unsigned int);
6263 vector signed short vec_sll (vector signed short,
6264 vector unsigned short);
6265 vector signed short vec_sll (vector signed short, vector unsigned char);
6267 vector unsigned short vec_sll (vector unsigned short,
6268 vector unsigned int);
6269 vector unsigned short vec_sll (vector unsigned short,
6270 vector unsigned short);
6271 vector unsigned short vec_sll (vector unsigned short,
6272 vector unsigned char);
6273 vector signed char vec_sll (vector signed char, vector unsigned int);
6274 vector signed char vec_sll (vector signed char, vector unsigned short);
6275 vector signed char vec_sll (vector signed char, vector unsigned char);
6276 vector unsigned char vec_sll (vector unsigned char,
6277 vector unsigned int);
6278 vector unsigned char vec_sll (vector unsigned char,
6279 vector unsigned short);
6280 vector unsigned char vec_sll (vector unsigned char,
6281 vector unsigned char);
6283 vector float vec_slo (vector float, vector signed char);
6284 vector float vec_slo (vector float, vector unsigned char);
6285 vector signed int vec_slo (vector signed int, vector signed char);
6286 vector signed int vec_slo (vector signed int, vector unsigned char);
6287 vector unsigned int vec_slo (vector unsigned int, vector signed char);
6288 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
6290 vector signed short vec_slo (vector signed short, vector signed char);
6291 vector signed short vec_slo (vector signed short, vector unsigned char);
6293 vector unsigned short vec_slo (vector unsigned short,
6294 vector signed char);
6295 vector unsigned short vec_slo (vector unsigned short,
6296 vector unsigned char);
6297 vector signed char vec_slo (vector signed char, vector signed char);
6298 vector signed char vec_slo (vector signed char, vector unsigned char);
6299 vector unsigned char vec_slo (vector unsigned char, vector signed char);
6301 vector unsigned char vec_slo (vector unsigned char,
6302 vector unsigned char);
6304 vector signed char vec_splat (vector signed char, const char);
6305 vector unsigned char vec_splat (vector unsigned char, const char);
6306 vector signed short vec_splat (vector signed short, const char);
6307 vector unsigned short vec_splat (vector unsigned short, const char);
6308 vector float vec_splat (vector float, const char);
6309 vector signed int vec_splat (vector signed int, const char);
6310 vector unsigned int vec_splat (vector unsigned int, const char);
6312 vector signed char vec_splat_s8 (const char);
6314 vector signed short vec_splat_s16 (const char);
6316 vector signed int vec_splat_s32 (const char);
6318 vector unsigned char vec_splat_u8 (const char);
6320 vector unsigned short vec_splat_u16 (const char);
6322 vector unsigned int vec_splat_u32 (const char);
6324 vector signed char vec_sr (vector signed char, vector unsigned char);
6325 vector unsigned char vec_sr (vector unsigned char,
6326 vector unsigned char);
6327 vector signed short vec_sr (vector signed short, vector unsigned short);
6329 vector unsigned short vec_sr (vector unsigned short,
6330 vector unsigned short);
6331 vector signed int vec_sr (vector signed int, vector unsigned int);
6332 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
6334 vector signed char vec_sra (vector signed char, vector unsigned char);
6335 vector unsigned char vec_sra (vector unsigned char,
6336 vector unsigned char);
6337 vector signed short vec_sra (vector signed short,
6338 vector unsigned short);
6339 vector unsigned short vec_sra (vector unsigned short,
6340 vector unsigned short);
6341 vector signed int vec_sra (vector signed int, vector unsigned int);
6342 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
6344 vector signed int vec_srl (vector signed int, vector unsigned int);
6345 vector signed int vec_srl (vector signed int, vector unsigned short);
6346 vector signed int vec_srl (vector signed int, vector unsigned char);
6347 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
6348 vector unsigned int vec_srl (vector unsigned int,
6349 vector unsigned short);
6350 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
6352 vector signed short vec_srl (vector signed short, vector unsigned int);
6353 vector signed short vec_srl (vector signed short,
6354 vector unsigned short);
6355 vector signed short vec_srl (vector signed short, vector unsigned char);
6357 vector unsigned short vec_srl (vector unsigned short,
6358 vector unsigned int);
6359 vector unsigned short vec_srl (vector unsigned short,
6360 vector unsigned short);
6361 vector unsigned short vec_srl (vector unsigned short,
6362 vector unsigned char);
6363 vector signed char vec_srl (vector signed char, vector unsigned int);
6364 vector signed char vec_srl (vector signed char, vector unsigned short);
6365 vector signed char vec_srl (vector signed char, vector unsigned char);
6366 vector unsigned char vec_srl (vector unsigned char,
6367 vector unsigned int);
6368 vector unsigned char vec_srl (vector unsigned char,
6369 vector unsigned short);
6370 vector unsigned char vec_srl (vector unsigned char,
6371 vector unsigned char);
6373 vector float vec_sro (vector float, vector signed char);
6374 vector float vec_sro (vector float, vector unsigned char);
6375 vector signed int vec_sro (vector signed int, vector signed char);
6376 vector signed int vec_sro (vector signed int, vector unsigned char);
6377 vector unsigned int vec_sro (vector unsigned int, vector signed char);
6378 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
6380 vector signed short vec_sro (vector signed short, vector signed char);
6381 vector signed short vec_sro (vector signed short, vector unsigned char);
6383 vector unsigned short vec_sro (vector unsigned short,
6384 vector signed char);
6385 vector unsigned short vec_sro (vector unsigned short,
6386 vector unsigned char);
6387 vector signed char vec_sro (vector signed char, vector signed char);
6388 vector signed char vec_sro (vector signed char, vector unsigned char);
6389 vector unsigned char vec_sro (vector unsigned char, vector signed char);
6391 vector unsigned char vec_sro (vector unsigned char,
6392 vector unsigned char);
6394 void vec_st (vector float, int, float *);
6395 void vec_st (vector float, int, vector float *);
6396 void vec_st (vector signed int, int, int *);
6397 void vec_st (vector signed int, int, unsigned int *);
6398 void vec_st (vector unsigned int, int, unsigned int *);
6399 void vec_st (vector unsigned int, int, vector unsigned int *);
6400 void vec_st (vector signed short, int, short *);
6401 void vec_st (vector signed short, int, vector unsigned short *);
6402 void vec_st (vector signed short, int, vector signed short *);
6403 void vec_st (vector unsigned short, int, unsigned short *);
6404 void vec_st (vector unsigned short, int, vector unsigned short *);
6405 void vec_st (vector signed char, int, signed char *);
6406 void vec_st (vector signed char, int, unsigned char *);
6407 void vec_st (vector signed char, int, vector signed char *);
6408 void vec_st (vector unsigned char, int, unsigned char *);
6409 void vec_st (vector unsigned char, int, vector unsigned char *);
6411 void vec_ste (vector signed char, int, unsigned char *);
6412 void vec_ste (vector signed char, int, signed char *);
6413 void vec_ste (vector unsigned char, int, unsigned char *);
6414 void vec_ste (vector signed short, int, short *);
6415 void vec_ste (vector signed short, int, unsigned short *);
6416 void vec_ste (vector unsigned short, int, void *);
6417 void vec_ste (vector signed int, int, unsigned int *);
6418 void vec_ste (vector signed int, int, int *);
6419 void vec_ste (vector unsigned int, int, unsigned int *);
6420 void vec_ste (vector float, int, float *);
6422 void vec_stl (vector float, int, vector float *);
6423 void vec_stl (vector float, int, float *);
6424 void vec_stl (vector signed int, int, vector signed int *);
6425 void vec_stl (vector signed int, int, int *);
6426 void vec_stl (vector signed int, int, unsigned int *);
6427 void vec_stl (vector unsigned int, int, vector unsigned int *);
6428 void vec_stl (vector unsigned int, int, unsigned int *);
6429 void vec_stl (vector signed short, int, short *);
6430 void vec_stl (vector signed short, int, unsigned short *);
6431 void vec_stl (vector signed short, int, vector signed short *);
6432 void vec_stl (vector unsigned short, int, unsigned short *);
6433 void vec_stl (vector unsigned short, int, vector signed short *);
6434 void vec_stl (vector signed char, int, signed char *);
6435 void vec_stl (vector signed char, int, unsigned char *);
6436 void vec_stl (vector signed char, int, vector signed char *);
6437 void vec_stl (vector unsigned char, int, unsigned char *);
6438 void vec_stl (vector unsigned char, int, vector unsigned char *);
6440 vector signed char vec_sub (vector signed char, vector signed char);
6441 vector unsigned char vec_sub (vector signed char, vector unsigned char);
6443 vector unsigned char vec_sub (vector unsigned char, vector signed char);
6445 vector unsigned char vec_sub (vector unsigned char,
6446 vector unsigned char);
6447 vector signed short vec_sub (vector signed short, vector signed short);
6448 vector unsigned short vec_sub (vector signed short,
6449 vector unsigned short);
6450 vector unsigned short vec_sub (vector unsigned short,
6451 vector signed short);
6452 vector unsigned short vec_sub (vector unsigned short,
6453 vector unsigned short);
6454 vector signed int vec_sub (vector signed int, vector signed int);
6455 vector unsigned int vec_sub (vector signed int, vector unsigned int);
6456 vector unsigned int vec_sub (vector unsigned int, vector signed int);
6457 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
6458 vector float vec_sub (vector float, vector float);
6460 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
6462 vector unsigned char vec_subs (vector signed char,
6463 vector unsigned char);
6464 vector unsigned char vec_subs (vector unsigned char,
6465 vector signed char);
6466 vector unsigned char vec_subs (vector unsigned char,
6467 vector unsigned char);
6468 vector signed char vec_subs (vector signed char, vector signed char);
6469 vector unsigned short vec_subs (vector signed short,
6470 vector unsigned short);
6471 vector unsigned short vec_subs (vector unsigned short,
6472 vector signed short);
6473 vector unsigned short vec_subs (vector unsigned short,
6474 vector unsigned short);
6475 vector signed short vec_subs (vector signed short, vector signed short);
6477 vector unsigned int vec_subs (vector signed int, vector unsigned int);
6478 vector unsigned int vec_subs (vector unsigned int, vector signed int);
6479 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
6481 vector signed int vec_subs (vector signed int, vector signed int);
6483 vector unsigned int vec_sum4s (vector unsigned char,
6484 vector unsigned int);
6485 vector signed int vec_sum4s (vector signed char, vector signed int);
6486 vector signed int vec_sum4s (vector signed short, vector signed int);
6488 vector signed int vec_sum2s (vector signed int, vector signed int);
6490 vector signed int vec_sums (vector signed int, vector signed int);
6492 vector float vec_trunc (vector float);
6494 vector signed short vec_unpackh (vector signed char);
6495 vector unsigned int vec_unpackh (vector signed short);
6496 vector signed int vec_unpackh (vector signed short);
6498 vector signed short vec_unpackl (vector signed char);
6499 vector unsigned int vec_unpackl (vector signed short);
6500 vector signed int vec_unpackl (vector signed short);
6502 vector float vec_xor (vector float, vector float);
6503 vector float vec_xor (vector float, vector signed int);
6504 vector float vec_xor (vector signed int, vector float);
6505 vector signed int vec_xor (vector signed int, vector signed int);
6506 vector unsigned int vec_xor (vector signed int, vector unsigned int);
6507 vector unsigned int vec_xor (vector unsigned int, vector signed int);
6508 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
6509 vector signed short vec_xor (vector signed short, vector signed short);
6510 vector unsigned short vec_xor (vector signed short,
6511 vector unsigned short);
6512 vector unsigned short vec_xor (vector unsigned short,
6513 vector signed short);
6514 vector unsigned short vec_xor (vector unsigned short,
6515 vector unsigned short);
6516 vector signed char vec_xor (vector signed char, vector signed char);
6517 vector unsigned char vec_xor (vector signed char, vector unsigned char);
6519 vector unsigned char vec_xor (vector unsigned char, vector signed char);
6521 vector unsigned char vec_xor (vector unsigned char,
6522 vector unsigned char);
6524 vector signed int vec_all_eq (vector signed char, vector unsigned char);
6526 vector signed int vec_all_eq (vector signed char, vector signed char);
6527 vector signed int vec_all_eq (vector unsigned char, vector signed char);
6529 vector signed int vec_all_eq (vector unsigned char,
6530 vector unsigned char);
6531 vector signed int vec_all_eq (vector signed short,
6532 vector unsigned short);
6533 vector signed int vec_all_eq (vector signed short, vector signed short);
6535 vector signed int vec_all_eq (vector unsigned short,
6536 vector signed short);
6537 vector signed int vec_all_eq (vector unsigned short,
6538 vector unsigned short);
6539 vector signed int vec_all_eq (vector signed int, vector unsigned int);
6540 vector signed int vec_all_eq (vector signed int, vector signed int);
6541 vector signed int vec_all_eq (vector unsigned int, vector signed int);
6542 vector signed int vec_all_eq (vector unsigned int, vector unsigned int);
6544 vector signed int vec_all_eq (vector float, vector float);
6546 vector signed int vec_all_ge (vector signed char, vector unsigned char);
6548 vector signed int vec_all_ge (vector unsigned char, vector signed char);
6550 vector signed int vec_all_ge (vector unsigned char,
6551 vector unsigned char);
6552 vector signed int vec_all_ge (vector signed char, vector signed char);
6553 vector signed int vec_all_ge (vector signed short,
6554 vector unsigned short);
6555 vector signed int vec_all_ge (vector unsigned short,
6556 vector signed short);
6557 vector signed int vec_all_ge (vector unsigned short,
6558 vector unsigned short);
6559 vector signed int vec_all_ge (vector signed short, vector signed short);
6561 vector signed int vec_all_ge (vector signed int, vector unsigned int);
6562 vector signed int vec_all_ge (vector unsigned int, vector signed int);
6563 vector signed int vec_all_ge (vector unsigned int, vector unsigned int);
6565 vector signed int vec_all_ge (vector signed int, vector signed int);
6566 vector signed int vec_all_ge (vector float, vector float);
6568 vector signed int vec_all_gt (vector signed char, vector unsigned char);
6570 vector signed int vec_all_gt (vector unsigned char, vector signed char);
6572 vector signed int vec_all_gt (vector unsigned char,
6573 vector unsigned char);
6574 vector signed int vec_all_gt (vector signed char, vector signed char);
6575 vector signed int vec_all_gt (vector signed short,
6576 vector unsigned short);
6577 vector signed int vec_all_gt (vector unsigned short,
6578 vector signed short);
6579 vector signed int vec_all_gt (vector unsigned short,
6580 vector unsigned short);
6581 vector signed int vec_all_gt (vector signed short, vector signed short);
6583 vector signed int vec_all_gt (vector signed int, vector unsigned int);
6584 vector signed int vec_all_gt (vector unsigned int, vector signed int);
6585 vector signed int vec_all_gt (vector unsigned int, vector unsigned int);
6587 vector signed int vec_all_gt (vector signed int, vector signed int);
6588 vector signed int vec_all_gt (vector float, vector float);
6590 vector signed int vec_all_in (vector float, vector float);
6592 vector signed int vec_all_le (vector signed char, vector unsigned char);
6594 vector signed int vec_all_le (vector unsigned char, vector signed char);
6596 vector signed int vec_all_le (vector unsigned char,
6597 vector unsigned char);
6598 vector signed int vec_all_le (vector signed char, vector signed char);
6599 vector signed int vec_all_le (vector signed short,
6600 vector unsigned short);
6601 vector signed int vec_all_le (vector unsigned short,
6602 vector signed short);
6603 vector signed int vec_all_le (vector unsigned short,
6604 vector unsigned short);
6605 vector signed int vec_all_le (vector signed short, vector signed short);
6607 vector signed int vec_all_le (vector signed int, vector unsigned int);
6608 vector signed int vec_all_le (vector unsigned int, vector signed int);
6609 vector signed int vec_all_le (vector unsigned int, vector unsigned int);
6611 vector signed int vec_all_le (vector signed int, vector signed int);
6612 vector signed int vec_all_le (vector float, vector float);
6614 vector signed int vec_all_lt (vector signed char, vector unsigned char);
6616 vector signed int vec_all_lt (vector unsigned char, vector signed char);
6618 vector signed int vec_all_lt (vector unsigned char,
6619 vector unsigned char);
6620 vector signed int vec_all_lt (vector signed char, vector signed char);
6621 vector signed int vec_all_lt (vector signed short,
6622 vector unsigned short);
6623 vector signed int vec_all_lt (vector unsigned short,
6624 vector signed short);
6625 vector signed int vec_all_lt (vector unsigned short,
6626 vector unsigned short);
6627 vector signed int vec_all_lt (vector signed short, vector signed short);
6629 vector signed int vec_all_lt (vector signed int, vector unsigned int);
6630 vector signed int vec_all_lt (vector unsigned int, vector signed int);
6631 vector signed int vec_all_lt (vector unsigned int, vector unsigned int);
6633 vector signed int vec_all_lt (vector signed int, vector signed int);
6634 vector signed int vec_all_lt (vector float, vector float);
6636 vector signed int vec_all_nan (vector float);
6638 vector signed int vec_all_ne (vector signed char, vector unsigned char);
6640 vector signed int vec_all_ne (vector signed char, vector signed char);
6641 vector signed int vec_all_ne (vector unsigned char, vector signed char);
6643 vector signed int vec_all_ne (vector unsigned char,
6644 vector unsigned char);
6645 vector signed int vec_all_ne (vector signed short,
6646 vector unsigned short);
6647 vector signed int vec_all_ne (vector signed short, vector signed short);
6649 vector signed int vec_all_ne (vector unsigned short,
6650 vector signed short);
6651 vector signed int vec_all_ne (vector unsigned short,
6652 vector unsigned short);
6653 vector signed int vec_all_ne (vector signed int, vector unsigned int);
6654 vector signed int vec_all_ne (vector signed int, vector signed int);
6655 vector signed int vec_all_ne (vector unsigned int, vector signed int);
6656 vector signed int vec_all_ne (vector unsigned int, vector unsigned int);
6658 vector signed int vec_all_ne (vector float, vector float);
6660 vector signed int vec_all_nge (vector float, vector float);
6662 vector signed int vec_all_ngt (vector float, vector float);
6664 vector signed int vec_all_nle (vector float, vector float);
6666 vector signed int vec_all_nlt (vector float, vector float);
6668 vector signed int vec_all_numeric (vector float);
6670 vector signed int vec_any_eq (vector signed char, vector unsigned char);
6672 vector signed int vec_any_eq (vector signed char, vector signed char);
6673 vector signed int vec_any_eq (vector unsigned char, vector signed char);
6675 vector signed int vec_any_eq (vector unsigned char,
6676 vector unsigned char);
6677 vector signed int vec_any_eq (vector signed short,
6678 vector unsigned short);
6679 vector signed int vec_any_eq (vector signed short, vector signed short);
6681 vector signed int vec_any_eq (vector unsigned short,
6682 vector signed short);
6683 vector signed int vec_any_eq (vector unsigned short,
6684 vector unsigned short);
6685 vector signed int vec_any_eq (vector signed int, vector unsigned int);
6686 vector signed int vec_any_eq (vector signed int, vector signed int);
6687 vector signed int vec_any_eq (vector unsigned int, vector signed int);
6688 vector signed int vec_any_eq (vector unsigned int, vector unsigned int);
6690 vector signed int vec_any_eq (vector float, vector float);
6692 vector signed int vec_any_ge (vector signed char, vector unsigned char);
6694 vector signed int vec_any_ge (vector unsigned char, vector signed char);
6696 vector signed int vec_any_ge (vector unsigned char,
6697 vector unsigned char);
6698 vector signed int vec_any_ge (vector signed char, vector signed char);
6699 vector signed int vec_any_ge (vector signed short,
6700 vector unsigned short);
6701 vector signed int vec_any_ge (vector unsigned short,
6702 vector signed short);
6703 vector signed int vec_any_ge (vector unsigned short,
6704 vector unsigned short);
6705 vector signed int vec_any_ge (vector signed short, vector signed short);
6707 vector signed int vec_any_ge (vector signed int, vector unsigned int);
6708 vector signed int vec_any_ge (vector unsigned int, vector signed int);
6709 vector signed int vec_any_ge (vector unsigned int, vector unsigned int);
6711 vector signed int vec_any_ge (vector signed int, vector signed int);
6712 vector signed int vec_any_ge (vector float, vector float);
6714 vector signed int vec_any_gt (vector signed char, vector unsigned char);
6716 vector signed int vec_any_gt (vector unsigned char, vector signed char);
6718 vector signed int vec_any_gt (vector unsigned char,
6719 vector unsigned char);
6720 vector signed int vec_any_gt (vector signed char, vector signed char);
6721 vector signed int vec_any_gt (vector signed short,
6722 vector unsigned short);
6723 vector signed int vec_any_gt (vector unsigned short,
6724 vector signed short);
6725 vector signed int vec_any_gt (vector unsigned short,
6726 vector unsigned short);
6727 vector signed int vec_any_gt (vector signed short, vector signed short);
6729 vector signed int vec_any_gt (vector signed int, vector unsigned int);
6730 vector signed int vec_any_gt (vector unsigned int, vector signed int);
6731 vector signed int vec_any_gt (vector unsigned int, vector unsigned int);
6733 vector signed int vec_any_gt (vector signed int, vector signed int);
6734 vector signed int vec_any_gt (vector float, vector float);
6736 vector signed int vec_any_le (vector signed char, vector unsigned char);
6738 vector signed int vec_any_le (vector unsigned char, vector signed char);
6740 vector signed int vec_any_le (vector unsigned char,
6741 vector unsigned char);
6742 vector signed int vec_any_le (vector signed char, vector signed char);
6743 vector signed int vec_any_le (vector signed short,
6744 vector unsigned short);
6745 vector signed int vec_any_le (vector unsigned short,
6746 vector signed short);
6747 vector signed int vec_any_le (vector unsigned short,
6748 vector unsigned short);
6749 vector signed int vec_any_le (vector signed short, vector signed short);
6751 vector signed int vec_any_le (vector signed int, vector unsigned int);
6752 vector signed int vec_any_le (vector unsigned int, vector signed int);
6753 vector signed int vec_any_le (vector unsigned int, vector unsigned int);
6755 vector signed int vec_any_le (vector signed int, vector signed int);
6756 vector signed int vec_any_le (vector float, vector float);
6758 vector signed int vec_any_lt (vector signed char, vector unsigned char);
6760 vector signed int vec_any_lt (vector unsigned char, vector signed char);
6762 vector signed int vec_any_lt (vector unsigned char,
6763 vector unsigned char);
6764 vector signed int vec_any_lt (vector signed char, vector signed char);
6765 vector signed int vec_any_lt (vector signed short,
6766 vector unsigned short);
6767 vector signed int vec_any_lt (vector unsigned short,
6768 vector signed short);
6769 vector signed int vec_any_lt (vector unsigned short,
6770 vector unsigned short);
6771 vector signed int vec_any_lt (vector signed short, vector signed short);
6773 vector signed int vec_any_lt (vector signed int, vector unsigned int);
6774 vector signed int vec_any_lt (vector unsigned int, vector signed int);
6775 vector signed int vec_any_lt (vector unsigned int, vector unsigned int);
6777 vector signed int vec_any_lt (vector signed int, vector signed int);
6778 vector signed int vec_any_lt (vector float, vector float);
6780 vector signed int vec_any_nan (vector float);
6782 vector signed int vec_any_ne (vector signed char, vector unsigned char);
6784 vector signed int vec_any_ne (vector signed char, vector signed char);
6785 vector signed int vec_any_ne (vector unsigned char, vector signed char);
6787 vector signed int vec_any_ne (vector unsigned char,
6788 vector unsigned char);
6789 vector signed int vec_any_ne (vector signed short,
6790 vector unsigned short);
6791 vector signed int vec_any_ne (vector signed short, vector signed short);
6793 vector signed int vec_any_ne (vector unsigned short,
6794 vector signed short);
6795 vector signed int vec_any_ne (vector unsigned short,
6796 vector unsigned short);
6797 vector signed int vec_any_ne (vector signed int, vector unsigned int);
6798 vector signed int vec_any_ne (vector signed int, vector signed int);
6799 vector signed int vec_any_ne (vector unsigned int, vector signed int);
6800 vector signed int vec_any_ne (vector unsigned int, vector unsigned int);
6802 vector signed int vec_any_ne (vector float, vector float);
6804 vector signed int vec_any_nge (vector float, vector float);
6806 vector signed int vec_any_ngt (vector float, vector float);
6808 vector signed int vec_any_nle (vector float, vector float);
6810 vector signed int vec_any_nlt (vector float, vector float);
6812 vector signed int vec_any_numeric (vector float);
6814 vector signed int vec_any_out (vector float, vector float);
6817 @node Target Format Checks
6818 @section Format Checks Specific to Particular Target Machines
6820 For some target machines, GCC supports additional options to the
6822 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
6825 * Solaris Format Checks::
6828 @node Solaris Format Checks
6829 @subsection Solaris Format Checks
6831 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
6832 check. @code{cmn_err} accepts a subset of the standard @code{printf}
6833 conversions, and the two-argument @code{%b} conversion for displaying
6834 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
6837 @section Pragmas Accepted by GCC
6841 GCC supports several types of pragmas, primarily in order to compile
6842 code originally written for other compilers. Note that in general
6843 we do not recommend the use of pragmas; @xref{Function Attributes},
6844 for further explanation.
6848 * RS/6000 and PowerPC Pragmas::
6850 * Symbol-Renaming Pragmas::
6854 @subsection ARM Pragmas
6856 The ARM target defines pragmas for controlling the default addition of
6857 @code{long_call} and @code{short_call} attributes to functions.
6858 @xref{Function Attributes}, for information about the effects of these
6863 @cindex pragma, long_calls
6864 Set all subsequent functions to have the @code{long_call} attribute.
6867 @cindex pragma, no_long_calls
6868 Set all subsequent functions to have the @code{short_call} attribute.
6870 @item long_calls_off
6871 @cindex pragma, long_calls_off
6872 Do not affect the @code{long_call} or @code{short_call} attributes of
6873 subsequent functions.
6876 @node RS/6000 and PowerPC Pragmas
6877 @subsection RS/6000 and PowerPC Pragmas
6879 The RS/6000 and PowerPC targets define one pragma for controlling
6880 whether or not the @code{longcall} attribute is added to function
6881 declarations by default. This pragma overrides the @option{-mlongcall}
6882 option, but not the @code{longcall} and @code{shortcall} attributes.
6883 @xref{RS/6000 and PowerPC Options}, for more information about when long
6884 calls are and are not necessary.
6888 @cindex pragma, longcall
6889 Apply the @code{longcall} attribute to all subsequent function
6893 Do not apply the @code{longcall} attribute to subsequent function
6897 @c Describe c4x pragmas here.
6898 @c Describe h8300 pragmas here.
6899 @c Describe sh pragmas here.
6900 @c Describe v850 pragmas here.
6902 @node Darwin Pragmas
6903 @subsection Darwin Pragmas
6905 The following pragmas are available for all architectures running the
6906 Darwin operating system. These are useful for compatibility with other
6910 @item mark @var{tokens}@dots{}
6911 @cindex pragma, mark
6912 This pragma is accepted, but has no effect.
6914 @item options align=@var{alignment}
6915 @cindex pragma, options align
6916 This pragma sets the alignment of fields in structures. The values of
6917 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
6918 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
6919 properly; to restore the previous setting, use @code{reset} for the
6922 @item segment @var{tokens}@dots{}
6923 @cindex pragma, segment
6924 This pragma is accepted, but has no effect.
6926 @item unused (@var{var} [, @var{var}]@dots{})
6927 @cindex pragma, unused
6928 This pragma declares variables to be possibly unused. GCC will not
6929 produce warnings for the listed variables. The effect is similar to
6930 that of the @code{unused} attribute, except that this pragma may appear
6931 anywhere within the variables' scopes.
6934 @node Symbol-Renaming Pragmas
6935 @subsection Symbol-Renaming Pragmas
6937 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
6938 supports two @code{#pragma} directives which change the name used in
6939 assembly for a given declaration. These pragmas are only available on
6940 platforms whose system headers need them. To get this effect on all
6941 platforms supported by GCC, use the asm labels extension (@pxref{Asm
6945 @item redefine_extname @var{oldname} @var{newname}
6946 @cindex pragma, redefine_extname
6948 This pragma gives the C function @var{oldname} the assembly symbol
6949 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
6950 will be defined if this pragma is available (currently only on
6953 @item extern_prefix @var{string}
6954 @cindex pragma, extern_prefix
6956 This pragma causes all subsequent external function and variable
6957 declarations to have @var{string} prepended to their assembly symbols.
6958 This effect may be terminated with another @code{extern_prefix} pragma
6959 whose argument is an empty string. The preprocessor macro
6960 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
6961 available (currently only on Tru64 UNIX).
6964 These pragmas and the asm labels extension interact in a complicated
6965 manner. Here are some corner cases you may want to be aware of.
6968 @item Both pragmas silently apply only to declarations with external
6969 linkage. Asm labels do not have this restriction.
6971 @item In C++, both pragmas silently apply only to declarations with
6972 ``C'' linkage. Again, asm labels do not have this restriction.
6974 @item If any of the three ways of changing the assembly name of a
6975 declaration is applied to a declaration whose assembly name has
6976 already been determined (either by a previous use of one of these
6977 features, or because the compiler needed the assembly name in order to
6978 generate code), and the new name is different, a warning issues and
6979 the name does not change.
6981 @item The @var{oldname} used by @code{#pragma redefine_extname} is
6982 always the C-language name.
6984 @item If @code{#pragma extern_prefix} is in effect, and a declaration
6985 occurs with an asm label attached, the prefix is silently ignored for
6988 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
6989 apply to the same declaration, whichever triggered first wins, and a
6990 warning issues if they contradict each other. (We would like to have
6991 @code{#pragma redefine_extname} always win, for consistency with asm
6992 labels, but if @code{#pragma extern_prefix} triggers first we have no
6993 way of knowing that that happened.)
6996 @node Unnamed Fields
6997 @section Unnamed struct/union fields within structs/unions.
7001 For compatibility with other compilers, GCC allows you to define
7002 a structure or union that contains, as fields, structures and unions
7003 without names. For example:
7016 In this example, the user would be able to access members of the unnamed
7017 union with code like @samp{foo.b}. Note that only unnamed structs and
7018 unions are allowed, you may not have, for example, an unnamed
7021 You must never create such structures that cause ambiguous field definitions.
7022 For example, this structure:
7033 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
7034 Such constructs are not supported and must be avoided. In the future,
7035 such constructs may be detected and treated as compilation errors.
7038 @section Thread-Local Storage
7039 @cindex Thread-Local Storage
7040 @cindex @acronym{TLS}
7043 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
7044 are allocated such that there is one instance of the variable per extant
7045 thread. The run-time model GCC uses to implement this originates
7046 in the IA-64 processor-specific ABI, but has since been migrated
7047 to other processors as well. It requires significant support from
7048 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
7049 system libraries (@file{libc.so} and @file{libpthread.so}), so it
7050 is not available everywhere.
7052 At the user level, the extension is visible with a new storage
7053 class keyword: @code{__thread}. For example:
7057 extern __thread struct state s;
7058 static __thread char *p;
7061 The @code{__thread} specifier may be used alone, with the @code{extern}
7062 or @code{static} specifiers, but with no other storage class specifier.
7063 When used with @code{extern} or @code{static}, @code{__thread} must appear
7064 immediately after the other storage class specifier.
7066 The @code{__thread} specifier may be applied to any global, file-scoped
7067 static, function-scoped static, or static data member of a class. It may
7068 not be applied to block-scoped automatic or non-static data member.
7070 When the address-of operator is applied to a thread-local variable, it is
7071 evaluated at run-time and returns the address of the current thread's
7072 instance of that variable. An address so obtained may be used by any
7073 thread. When a thread terminates, any pointers to thread-local variables
7074 in that thread become invalid.
7076 No static initialization may refer to the address of a thread-local variable.
7078 In C++, if an initializer is present for a thread-local variable, it must
7079 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
7082 See @uref{http://people.redhat.com/drepper/tls.pdf,
7083 ELF Handling For Thread-Local Storage} for a detailed explanation of
7084 the four thread-local storage addressing models, and how the run-time
7085 is expected to function.
7088 * C99 Thread-Local Edits::
7089 * C++98 Thread-Local Edits::
7092 @node C99 Thread-Local Edits
7093 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
7095 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
7096 that document the exact semantics of the language extension.
7100 @cite{5.1.2 Execution environments}
7102 Add new text after paragraph 1
7105 Within either execution environment, a @dfn{thread} is a flow of
7106 control within a program. It is implementation defined whether
7107 or not there may be more than one thread associated with a program.
7108 It is implementation defined how threads beyond the first are
7109 created, the name and type of the function called at thread
7110 startup, and how threads may be terminated. However, objects
7111 with thread storage duration shall be initialized before thread
7116 @cite{6.2.4 Storage durations of objects}
7118 Add new text before paragraph 3
7121 An object whose identifier is declared with the storage-class
7122 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
7123 Its lifetime is the entire execution of the thread, and its
7124 stored value is initialized only once, prior to thread startup.
7128 @cite{6.4.1 Keywords}
7130 Add @code{__thread}.
7133 @cite{6.7.1 Storage-class specifiers}
7135 Add @code{__thread} to the list of storage class specifiers in
7138 Change paragraph 2 to
7141 With the exception of @code{__thread}, at most one storage-class
7142 specifier may be given [@dots{}]. The @code{__thread} specifier may
7143 be used alone, or immediately following @code{extern} or
7147 Add new text after paragraph 6
7150 The declaration of an identifier for a variable that has
7151 block scope that specifies @code{__thread} shall also
7152 specify either @code{extern} or @code{static}.
7154 The @code{__thread} specifier shall be used only with
7159 @node C++98 Thread-Local Edits
7160 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
7162 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
7163 that document the exact semantics of the language extension.
7167 @b{[intro.execution]}
7169 New text after paragraph 4
7172 A @dfn{thread} is a flow of control within the abstract machine.
7173 It is implementation defined whether or not there may be more than
7177 New text after paragraph 7
7180 It is unspecified whether additional action must be taken to
7181 ensure when and whether side effects are visible to other threads.
7187 Add @code{__thread}.
7190 @b{[basic.start.main]}
7192 Add after paragraph 5
7195 The thread that begins execution at the @code{main} function is called
7196 the @dfn{main thread}. It is implementation defined how functions
7197 beginning threads other than the main thread are designated or typed.
7198 A function so designated, as well as the @code{main} function, is called
7199 a @dfn{thread startup function}. It is implementation defined what
7200 happens if a thread startup function returns. It is implementation
7201 defined what happens to other threads when any thread calls @code{exit}.
7205 @b{[basic.start.init]}
7207 Add after paragraph 4
7210 The storage for an object of thread storage duration shall be
7211 statically initialized before the first statement of the thread startup
7212 function. An object of thread storage duration shall not require
7213 dynamic initialization.
7217 @b{[basic.start.term]}
7219 Add after paragraph 3
7222 The type of an object with thread storage duration shall not have a
7223 non-trivial destructor, nor shall it be an array type whose elements
7224 (directly or indirectly) have non-trivial destructors.
7230 Add ``thread storage duration'' to the list in paragraph 1.
7235 Thread, static, and automatic storage durations are associated with
7236 objects introduced by declarations [@dots{}].
7239 Add @code{__thread} to the list of specifiers in paragraph 3.
7242 @b{[basic.stc.thread]}
7244 New section before @b{[basic.stc.static]}
7247 The keyword @code{__thread} applied to a non-local object gives the
7248 object thread storage duration.
7250 A local variable or class data member declared both @code{static}
7251 and @code{__thread} gives the variable or member thread storage
7256 @b{[basic.stc.static]}
7261 All objects which have neither thread storage duration, dynamic
7262 storage duration nor are local [@dots{}].
7268 Add @code{__thread} to the list in paragraph 1.
7273 With the exception of @code{__thread}, at most one
7274 @var{storage-class-specifier} shall appear in a given
7275 @var{decl-specifier-seq}. The @code{__thread} specifier may
7276 be used alone, or immediately following the @code{extern} or
7277 @code{static} specifiers. [@dots{}]
7280 Add after paragraph 5
7283 The @code{__thread} specifier can be applied only to the names of objects
7284 and to anonymous unions.
7290 Add after paragraph 6
7293 Non-@code{static} members shall not be @code{__thread}.
7297 @node C++ Extensions
7298 @chapter Extensions to the C++ Language
7299 @cindex extensions, C++ language
7300 @cindex C++ language extensions
7302 The GNU compiler provides these extensions to the C++ language (and you
7303 can also use most of the C language extensions in your C++ programs). If you
7304 want to write code that checks whether these features are available, you can
7305 test for the GNU compiler the same way as for C programs: check for a
7306 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
7307 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
7308 Predefined Macros,cpp,The GNU C Preprocessor}).
7311 * Min and Max:: C++ Minimum and maximum operators.
7312 * Volatiles:: What constitutes an access to a volatile object.
7313 * Restricted Pointers:: C99 restricted pointers and references.
7314 * Vague Linkage:: Where G++ puts inlines, vtables and such.
7315 * C++ Interface:: You can use a single C++ header file for both
7316 declarations and definitions.
7317 * Template Instantiation:: Methods for ensuring that exactly one copy of
7318 each needed template instantiation is emitted.
7319 * Bound member functions:: You can extract a function pointer to the
7320 method denoted by a @samp{->*} or @samp{.*} expression.
7321 * C++ Attributes:: Variable, function, and type attributes for C++ only.
7322 * Strong Using:: Strong using-directives for namespace composition.
7323 * Java Exceptions:: Tweaking exception handling to work with Java.
7324 * Deprecated Features:: Things will disappear from g++.
7325 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
7329 @section Minimum and Maximum Operators in C++
7331 It is very convenient to have operators which return the ``minimum'' or the
7332 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
7335 @item @var{a} <? @var{b}
7337 @cindex minimum operator
7338 is the @dfn{minimum}, returning the smaller of the numeric values
7339 @var{a} and @var{b};
7341 @item @var{a} >? @var{b}
7343 @cindex maximum operator
7344 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
7348 These operations are not primitive in ordinary C++, since you can
7349 use a macro to return the minimum of two things in C++, as in the
7353 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
7357 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
7358 the minimum value of variables @var{i} and @var{j}.
7360 However, side effects in @code{X} or @code{Y} may cause unintended
7361 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
7362 the smaller counter twice. The GNU C @code{typeof} extension allows you
7363 to write safe macros that avoid this kind of problem (@pxref{Typeof}).
7364 However, writing @code{MIN} and @code{MAX} as macros also forces you to
7365 use function-call notation for a fundamental arithmetic operation.
7366 Using GNU C++ extensions, you can write @w{@samp{int min = i <? j;}}
7369 Since @code{<?} and @code{>?} are built into the compiler, they properly
7370 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
7374 @section When is a Volatile Object Accessed?
7375 @cindex accessing volatiles
7376 @cindex volatile read
7377 @cindex volatile write
7378 @cindex volatile access
7380 Both the C and C++ standard have the concept of volatile objects. These
7381 are normally accessed by pointers and used for accessing hardware. The
7382 standards encourage compilers to refrain from optimizations
7383 concerning accesses to volatile objects that it might perform on
7384 non-volatile objects. The C standard leaves it implementation defined
7385 as to what constitutes a volatile access. The C++ standard omits to
7386 specify this, except to say that C++ should behave in a similar manner
7387 to C with respect to volatiles, where possible. The minimum either
7388 standard specifies is that at a sequence point all previous accesses to
7389 volatile objects have stabilized and no subsequent accesses have
7390 occurred. Thus an implementation is free to reorder and combine
7391 volatile accesses which occur between sequence points, but cannot do so
7392 for accesses across a sequence point. The use of volatiles does not
7393 allow you to violate the restriction on updating objects multiple times
7394 within a sequence point.
7396 In most expressions, it is intuitively obvious what is a read and what is
7397 a write. For instance
7400 volatile int *dst = @var{somevalue};
7401 volatile int *src = @var{someothervalue};
7406 will cause a read of the volatile object pointed to by @var{src} and stores the
7407 value into the volatile object pointed to by @var{dst}. There is no
7408 guarantee that these reads and writes are atomic, especially for objects
7409 larger than @code{int}.
7411 Less obvious expressions are where something which looks like an access
7412 is used in a void context. An example would be,
7415 volatile int *src = @var{somevalue};
7419 With C, such expressions are rvalues, and as rvalues cause a read of
7420 the object, GCC interprets this as a read of the volatile being pointed
7421 to. The C++ standard specifies that such expressions do not undergo
7422 lvalue to rvalue conversion, and that the type of the dereferenced
7423 object may be incomplete. The C++ standard does not specify explicitly
7424 that it is this lvalue to rvalue conversion which is responsible for
7425 causing an access. However, there is reason to believe that it is,
7426 because otherwise certain simple expressions become undefined. However,
7427 because it would surprise most programmers, G++ treats dereferencing a
7428 pointer to volatile object of complete type in a void context as a read
7429 of the object. When the object has incomplete type, G++ issues a
7434 struct T @{int m;@};
7435 volatile S *ptr1 = @var{somevalue};
7436 volatile T *ptr2 = @var{somevalue};
7441 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
7442 causes a read of the object pointed to. If you wish to force an error on
7443 the first case, you must force a conversion to rvalue with, for instance
7444 a static cast, @code{static_cast<S>(*ptr1)}.
7446 When using a reference to volatile, G++ does not treat equivalent
7447 expressions as accesses to volatiles, but instead issues a warning that
7448 no volatile is accessed. The rationale for this is that otherwise it
7449 becomes difficult to determine where volatile access occur, and not
7450 possible to ignore the return value from functions returning volatile
7451 references. Again, if you wish to force a read, cast the reference to
7454 @node Restricted Pointers
7455 @section Restricting Pointer Aliasing
7456 @cindex restricted pointers
7457 @cindex restricted references
7458 @cindex restricted this pointer
7460 As with the C front end, G++ understands the C99 feature of restricted pointers,
7461 specified with the @code{__restrict__}, or @code{__restrict} type
7462 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
7463 language flag, @code{restrict} is not a keyword in C++.
7465 In addition to allowing restricted pointers, you can specify restricted
7466 references, which indicate that the reference is not aliased in the local
7470 void fn (int *__restrict__ rptr, int &__restrict__ rref)
7477 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
7478 @var{rref} refers to a (different) unaliased integer.
7480 You may also specify whether a member function's @var{this} pointer is
7481 unaliased by using @code{__restrict__} as a member function qualifier.
7484 void T::fn () __restrict__
7491 Within the body of @code{T::fn}, @var{this} will have the effective
7492 definition @code{T *__restrict__ const this}. Notice that the
7493 interpretation of a @code{__restrict__} member function qualifier is
7494 different to that of @code{const} or @code{volatile} qualifier, in that it
7495 is applied to the pointer rather than the object. This is consistent with
7496 other compilers which implement restricted pointers.
7498 As with all outermost parameter qualifiers, @code{__restrict__} is
7499 ignored in function definition matching. This means you only need to
7500 specify @code{__restrict__} in a function definition, rather than
7501 in a function prototype as well.
7504 @section Vague Linkage
7505 @cindex vague linkage
7507 There are several constructs in C++ which require space in the object
7508 file but are not clearly tied to a single translation unit. We say that
7509 these constructs have ``vague linkage''. Typically such constructs are
7510 emitted wherever they are needed, though sometimes we can be more
7514 @item Inline Functions
7515 Inline functions are typically defined in a header file which can be
7516 included in many different compilations. Hopefully they can usually be
7517 inlined, but sometimes an out-of-line copy is necessary, if the address
7518 of the function is taken or if inlining fails. In general, we emit an
7519 out-of-line copy in all translation units where one is needed. As an
7520 exception, we only emit inline virtual functions with the vtable, since
7521 it will always require a copy.
7523 Local static variables and string constants used in an inline function
7524 are also considered to have vague linkage, since they must be shared
7525 between all inlined and out-of-line instances of the function.
7529 C++ virtual functions are implemented in most compilers using a lookup
7530 table, known as a vtable. The vtable contains pointers to the virtual
7531 functions provided by a class, and each object of the class contains a
7532 pointer to its vtable (or vtables, in some multiple-inheritance
7533 situations). If the class declares any non-inline, non-pure virtual
7534 functions, the first one is chosen as the ``key method'' for the class,
7535 and the vtable is only emitted in the translation unit where the key
7538 @emph{Note:} If the chosen key method is later defined as inline, the
7539 vtable will still be emitted in every translation unit which defines it.
7540 Make sure that any inline virtuals are declared inline in the class
7541 body, even if they are not defined there.
7543 @item type_info objects
7546 C++ requires information about types to be written out in order to
7547 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
7548 For polymorphic classes (classes with virtual functions), the type_info
7549 object is written out along with the vtable so that @samp{dynamic_cast}
7550 can determine the dynamic type of a class object at runtime. For all
7551 other types, we write out the type_info object when it is used: when
7552 applying @samp{typeid} to an expression, throwing an object, or
7553 referring to a type in a catch clause or exception specification.
7555 @item Template Instantiations
7556 Most everything in this section also applies to template instantiations,
7557 but there are other options as well.
7558 @xref{Template Instantiation,,Where's the Template?}.
7562 When used with GNU ld version 2.8 or later on an ELF system such as
7563 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
7564 these constructs will be discarded at link time. This is known as
7567 On targets that don't support COMDAT, but do support weak symbols, GCC
7568 will use them. This way one copy will override all the others, but
7569 the unused copies will still take up space in the executable.
7571 For targets which do not support either COMDAT or weak symbols,
7572 most entities with vague linkage will be emitted as local symbols to
7573 avoid duplicate definition errors from the linker. This will not happen
7574 for local statics in inlines, however, as having multiple copies will
7575 almost certainly break things.
7577 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
7578 another way to control placement of these constructs.
7581 @section #pragma interface and implementation
7583 @cindex interface and implementation headers, C++
7584 @cindex C++ interface and implementation headers
7585 @cindex pragmas, interface and implementation
7587 @code{#pragma interface} and @code{#pragma implementation} provide the
7588 user with a way of explicitly directing the compiler to emit entities
7589 with vague linkage (and debugging information) in a particular
7592 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
7593 most cases, because of COMDAT support and the ``key method'' heuristic
7594 mentioned in @ref{Vague Linkage}. Using them can actually cause your
7595 program to grow due to unnecesary out-of-line copies of inline
7596 functions. Currently (3.4) the only benefit of these
7597 @code{#pragma}s is reduced duplication of debugging information, and
7598 that should be addressed soon on DWARF 2 targets with the use of
7602 @item #pragma interface
7603 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
7604 @kindex #pragma interface
7605 Use this directive in @emph{header files} that define object classes, to save
7606 space in most of the object files that use those classes. Normally,
7607 local copies of certain information (backup copies of inline member
7608 functions, debugging information, and the internal tables that implement
7609 virtual functions) must be kept in each object file that includes class
7610 definitions. You can use this pragma to avoid such duplication. When a
7611 header file containing @samp{#pragma interface} is included in a
7612 compilation, this auxiliary information will not be generated (unless
7613 the main input source file itself uses @samp{#pragma implementation}).
7614 Instead, the object files will contain references to be resolved at link
7617 The second form of this directive is useful for the case where you have
7618 multiple headers with the same name in different directories. If you
7619 use this form, you must specify the same string to @samp{#pragma
7622 @item #pragma implementation
7623 @itemx #pragma implementation "@var{objects}.h"
7624 @kindex #pragma implementation
7625 Use this pragma in a @emph{main input file}, when you want full output from
7626 included header files to be generated (and made globally visible). The
7627 included header file, in turn, should use @samp{#pragma interface}.
7628 Backup copies of inline member functions, debugging information, and the
7629 internal tables used to implement virtual functions are all generated in
7630 implementation files.
7632 @cindex implied @code{#pragma implementation}
7633 @cindex @code{#pragma implementation}, implied
7634 @cindex naming convention, implementation headers
7635 If you use @samp{#pragma implementation} with no argument, it applies to
7636 an include file with the same basename@footnote{A file's @dfn{basename}
7637 was the name stripped of all leading path information and of trailing
7638 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
7639 file. For example, in @file{allclass.cc}, giving just
7640 @samp{#pragma implementation}
7641 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
7643 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
7644 an implementation file whenever you would include it from
7645 @file{allclass.cc} even if you never specified @samp{#pragma
7646 implementation}. This was deemed to be more trouble than it was worth,
7647 however, and disabled.
7649 Use the string argument if you want a single implementation file to
7650 include code from multiple header files. (You must also use
7651 @samp{#include} to include the header file; @samp{#pragma
7652 implementation} only specifies how to use the file---it doesn't actually
7655 There is no way to split up the contents of a single header file into
7656 multiple implementation files.
7659 @cindex inlining and C++ pragmas
7660 @cindex C++ pragmas, effect on inlining
7661 @cindex pragmas in C++, effect on inlining
7662 @samp{#pragma implementation} and @samp{#pragma interface} also have an
7663 effect on function inlining.
7665 If you define a class in a header file marked with @samp{#pragma
7666 interface}, the effect on an inline function defined in that class is
7667 similar to an explicit @code{extern} declaration---the compiler emits
7668 no code at all to define an independent version of the function. Its
7669 definition is used only for inlining with its callers.
7671 @opindex fno-implement-inlines
7672 Conversely, when you include the same header file in a main source file
7673 that declares it as @samp{#pragma implementation}, the compiler emits
7674 code for the function itself; this defines a version of the function
7675 that can be found via pointers (or by callers compiled without
7676 inlining). If all calls to the function can be inlined, you can avoid
7677 emitting the function by compiling with @option{-fno-implement-inlines}.
7678 If any calls were not inlined, you will get linker errors.
7680 @node Template Instantiation
7681 @section Where's the Template?
7682 @cindex template instantiation
7684 C++ templates are the first language feature to require more
7685 intelligence from the environment than one usually finds on a UNIX
7686 system. Somehow the compiler and linker have to make sure that each
7687 template instance occurs exactly once in the executable if it is needed,
7688 and not at all otherwise. There are two basic approaches to this
7689 problem, which are referred to as the Borland model and the Cfront model.
7693 Borland C++ solved the template instantiation problem by adding the code
7694 equivalent of common blocks to their linker; the compiler emits template
7695 instances in each translation unit that uses them, and the linker
7696 collapses them together. The advantage of this model is that the linker
7697 only has to consider the object files themselves; there is no external
7698 complexity to worry about. This disadvantage is that compilation time
7699 is increased because the template code is being compiled repeatedly.
7700 Code written for this model tends to include definitions of all
7701 templates in the header file, since they must be seen to be
7705 The AT&T C++ translator, Cfront, solved the template instantiation
7706 problem by creating the notion of a template repository, an
7707 automatically maintained place where template instances are stored. A
7708 more modern version of the repository works as follows: As individual
7709 object files are built, the compiler places any template definitions and
7710 instantiations encountered in the repository. At link time, the link
7711 wrapper adds in the objects in the repository and compiles any needed
7712 instances that were not previously emitted. The advantages of this
7713 model are more optimal compilation speed and the ability to use the
7714 system linker; to implement the Borland model a compiler vendor also
7715 needs to replace the linker. The disadvantages are vastly increased
7716 complexity, and thus potential for error; for some code this can be
7717 just as transparent, but in practice it can been very difficult to build
7718 multiple programs in one directory and one program in multiple
7719 directories. Code written for this model tends to separate definitions
7720 of non-inline member templates into a separate file, which should be
7721 compiled separately.
7724 When used with GNU ld version 2.8 or later on an ELF system such as
7725 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
7726 Borland model. On other systems, G++ implements neither automatic
7729 A future version of G++ will support a hybrid model whereby the compiler
7730 will emit any instantiations for which the template definition is
7731 included in the compile, and store template definitions and
7732 instantiation context information into the object file for the rest.
7733 The link wrapper will extract that information as necessary and invoke
7734 the compiler to produce the remaining instantiations. The linker will
7735 then combine duplicate instantiations.
7737 In the mean time, you have the following options for dealing with
7738 template instantiations:
7743 Compile your template-using code with @option{-frepo}. The compiler will
7744 generate files with the extension @samp{.rpo} listing all of the
7745 template instantiations used in the corresponding object files which
7746 could be instantiated there; the link wrapper, @samp{collect2}, will
7747 then update the @samp{.rpo} files to tell the compiler where to place
7748 those instantiations and rebuild any affected object files. The
7749 link-time overhead is negligible after the first pass, as the compiler
7750 will continue to place the instantiations in the same files.
7752 This is your best option for application code written for the Borland
7753 model, as it will just work. Code written for the Cfront model will
7754 need to be modified so that the template definitions are available at
7755 one or more points of instantiation; usually this is as simple as adding
7756 @code{#include <tmethods.cc>} to the end of each template header.
7758 For library code, if you want the library to provide all of the template
7759 instantiations it needs, just try to link all of its object files
7760 together; the link will fail, but cause the instantiations to be
7761 generated as a side effect. Be warned, however, that this may cause
7762 conflicts if multiple libraries try to provide the same instantiations.
7763 For greater control, use explicit instantiation as described in the next
7767 @opindex fno-implicit-templates
7768 Compile your code with @option{-fno-implicit-templates} to disable the
7769 implicit generation of template instances, and explicitly instantiate
7770 all the ones you use. This approach requires more knowledge of exactly
7771 which instances you need than do the others, but it's less
7772 mysterious and allows greater control. You can scatter the explicit
7773 instantiations throughout your program, perhaps putting them in the
7774 translation units where the instances are used or the translation units
7775 that define the templates themselves; you can put all of the explicit
7776 instantiations you need into one big file; or you can create small files
7783 template class Foo<int>;
7784 template ostream& operator <<
7785 (ostream&, const Foo<int>&);
7788 for each of the instances you need, and create a template instantiation
7791 If you are using Cfront-model code, you can probably get away with not
7792 using @option{-fno-implicit-templates} when compiling files that don't
7793 @samp{#include} the member template definitions.
7795 If you use one big file to do the instantiations, you may want to
7796 compile it without @option{-fno-implicit-templates} so you get all of the
7797 instances required by your explicit instantiations (but not by any
7798 other files) without having to specify them as well.
7800 G++ has extended the template instantiation syntax given in the ISO
7801 standard to allow forward declaration of explicit instantiations
7802 (with @code{extern}), instantiation of the compiler support data for a
7803 template class (i.e.@: the vtable) without instantiating any of its
7804 members (with @code{inline}), and instantiation of only the static data
7805 members of a template class, without the support data or member
7806 functions (with (@code{static}):
7809 extern template int max (int, int);
7810 inline template class Foo<int>;
7811 static template class Foo<int>;
7815 Do nothing. Pretend G++ does implement automatic instantiation
7816 management. Code written for the Borland model will work fine, but
7817 each translation unit will contain instances of each of the templates it
7818 uses. In a large program, this can lead to an unacceptable amount of code
7822 @node Bound member functions
7823 @section Extracting the function pointer from a bound pointer to member function
7825 @cindex pointer to member function
7826 @cindex bound pointer to member function
7828 In C++, pointer to member functions (PMFs) are implemented using a wide
7829 pointer of sorts to handle all the possible call mechanisms; the PMF
7830 needs to store information about how to adjust the @samp{this} pointer,
7831 and if the function pointed to is virtual, where to find the vtable, and
7832 where in the vtable to look for the member function. If you are using
7833 PMFs in an inner loop, you should really reconsider that decision. If
7834 that is not an option, you can extract the pointer to the function that
7835 would be called for a given object/PMF pair and call it directly inside
7836 the inner loop, to save a bit of time.
7838 Note that you will still be paying the penalty for the call through a
7839 function pointer; on most modern architectures, such a call defeats the
7840 branch prediction features of the CPU@. This is also true of normal
7841 virtual function calls.
7843 The syntax for this extension is
7847 extern int (A::*fp)();
7848 typedef int (*fptr)(A *);
7850 fptr p = (fptr)(a.*fp);
7853 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
7854 no object is needed to obtain the address of the function. They can be
7855 converted to function pointers directly:
7858 fptr p1 = (fptr)(&A::foo);
7861 @opindex Wno-pmf-conversions
7862 You must specify @option{-Wno-pmf-conversions} to use this extension.
7864 @node C++ Attributes
7865 @section C++-Specific Variable, Function, and Type Attributes
7867 Some attributes only make sense for C++ programs.
7870 @item init_priority (@var{priority})
7871 @cindex init_priority attribute
7874 In Standard C++, objects defined at namespace scope are guaranteed to be
7875 initialized in an order in strict accordance with that of their definitions
7876 @emph{in a given translation unit}. No guarantee is made for initializations
7877 across translation units. However, GNU C++ allows users to control the
7878 order of initialization of objects defined at namespace scope with the
7879 @code{init_priority} attribute by specifying a relative @var{priority},
7880 a constant integral expression currently bounded between 101 and 65535
7881 inclusive. Lower numbers indicate a higher priority.
7883 In the following example, @code{A} would normally be created before
7884 @code{B}, but the @code{init_priority} attribute has reversed that order:
7887 Some_Class A __attribute__ ((init_priority (2000)));
7888 Some_Class B __attribute__ ((init_priority (543)));
7892 Note that the particular values of @var{priority} do not matter; only their
7895 @item java_interface
7896 @cindex java_interface attribute
7898 This type attribute informs C++ that the class is a Java interface. It may
7899 only be applied to classes declared within an @code{extern "Java"} block.
7900 Calls to methods declared in this interface will be dispatched using GCJ's
7901 interface table mechanism, instead of regular virtual table dispatch.
7905 See also @xref{Strong Using}.
7908 @section Strong Using
7910 @strong{Caution:} The semantics of this extension are not fully
7911 defined. Users should refrain from using this extension as its
7912 semantics may change subtly over time. It is possible that this
7913 extension wil be removed in future versions of G++.
7915 A using-directive with @code{__attribute ((strong))} is stronger
7916 than a normal using-directive in two ways:
7920 Templates from the used namespace can be specialized as though they were members of the using namespace.
7923 The using namespace is considered an associated namespace of all
7924 templates in the used namespace for purposes of argument-dependent
7928 This is useful for composing a namespace transparently from
7929 implementation namespaces. For example:
7934 template <class T> struct A @{ @};
7936 using namespace debug __attribute ((__strong__));
7937 template <> struct A<int> @{ @}; // ok to specialize
7939 template <class T> void f (A<T>);
7944 f (std::A<float>()); // lookup finds std::f
7949 @node Java Exceptions
7950 @section Java Exceptions
7952 The Java language uses a slightly different exception handling model
7953 from C++. Normally, GNU C++ will automatically detect when you are
7954 writing C++ code that uses Java exceptions, and handle them
7955 appropriately. However, if C++ code only needs to execute destructors
7956 when Java exceptions are thrown through it, GCC will guess incorrectly.
7957 Sample problematic code is:
7960 struct S @{ ~S(); @};
7961 extern void bar(); // is written in Java, and may throw exceptions
7970 The usual effect of an incorrect guess is a link failure, complaining of
7971 a missing routine called @samp{__gxx_personality_v0}.
7973 You can inform the compiler that Java exceptions are to be used in a
7974 translation unit, irrespective of what it might think, by writing
7975 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
7976 @samp{#pragma} must appear before any functions that throw or catch
7977 exceptions, or run destructors when exceptions are thrown through them.
7979 You cannot mix Java and C++ exceptions in the same translation unit. It
7980 is believed to be safe to throw a C++ exception from one file through
7981 another file compiled for the Java exception model, or vice versa, but
7982 there may be bugs in this area.
7984 @node Deprecated Features
7985 @section Deprecated Features
7987 In the past, the GNU C++ compiler was extended to experiment with new
7988 features, at a time when the C++ language was still evolving. Now that
7989 the C++ standard is complete, some of those features are superseded by
7990 superior alternatives. Using the old features might cause a warning in
7991 some cases that the feature will be dropped in the future. In other
7992 cases, the feature might be gone already.
7994 While the list below is not exhaustive, it documents some of the options
7995 that are now deprecated:
7998 @item -fexternal-templates
7999 @itemx -falt-external-templates
8000 These are two of the many ways for G++ to implement template
8001 instantiation. @xref{Template Instantiation}. The C++ standard clearly
8002 defines how template definitions have to be organized across
8003 implementation units. G++ has an implicit instantiation mechanism that
8004 should work just fine for standard-conforming code.
8006 @item -fstrict-prototype
8007 @itemx -fno-strict-prototype
8008 Previously it was possible to use an empty prototype parameter list to
8009 indicate an unspecified number of parameters (like C), rather than no
8010 parameters, as C++ demands. This feature has been removed, except where
8011 it is required for backwards compatibility @xref{Backwards Compatibility}.
8014 The named return value extension has been deprecated, and is now
8017 The use of initializer lists with new expressions has been deprecated,
8018 and is now removed from G++.
8020 Floating and complex non-type template parameters have been deprecated,
8021 and are now removed from G++.
8023 The implicit typename extension has been deprecated and is now
8026 The use of default arguments in function pointers, function typedefs and
8027 and other places where they are not permitted by the standard is
8028 deprecated and will be removed from a future version of G++.
8030 @node Backwards Compatibility
8031 @section Backwards Compatibility
8032 @cindex Backwards Compatibility
8033 @cindex ARM [Annotated C++ Reference Manual]
8035 Now that there is a definitive ISO standard C++, G++ has a specification
8036 to adhere to. The C++ language evolved over time, and features that
8037 used to be acceptable in previous drafts of the standard, such as the ARM
8038 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
8039 compilation of C++ written to such drafts, G++ contains some backwards
8040 compatibilities. @emph{All such backwards compatibility features are
8041 liable to disappear in future versions of G++.} They should be considered
8042 deprecated @xref{Deprecated Features}.
8046 If a variable is declared at for scope, it used to remain in scope until
8047 the end of the scope which contained the for statement (rather than just
8048 within the for scope). G++ retains this, but issues a warning, if such a
8049 variable is accessed outside the for scope.
8051 @item Implicit C language
8052 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
8053 scope to set the language. On such systems, all header files are
8054 implicitly scoped inside a C language scope. Also, an empty prototype
8055 @code{()} will be treated as an unspecified number of arguments, rather
8056 than no arguments, as C++ demands.