1 @c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001,2002 Free Software Foundation, Inc.
2 @c This is part of the GCC manual.
3 @c For copying conditions, see the file gcc.texi.
6 @chapter C Implementation-defined behavior
7 @cindex implementation-defined behavior, C language
9 A conforming implementation of ISO C is required to document its
10 choice of behavior in each of the areas that are designated
11 ``implementation defined.'' The following lists all such areas,
12 along with the section number from the ISO/IEC 9899:1999 standard.
15 * Translation implementation::
16 * Environment implementation::
17 * Identifiers implementation::
18 * Characters implementation::
19 * Integers implementation::
20 * Floating point implementation::
21 * Arrays and pointers implementation::
22 * Hints implementation::
23 * Structures unions enumerations and bit-fields implementation::
24 * Qualifiers implementation::
25 * Preprocessing directives implementation::
26 * Library functions implementation::
27 * Architecture implementation::
28 * Locale-specific behavior implementation::
31 @node Translation implementation
36 @cite{How a diagnostic is identified (3.10, 5.1.1.3).}
39 @cite{Whether each nonempty sequence of white-space characters other than
40 new-line is retained or replaced by one space character in translation
44 @node Environment implementation
47 The behavior of these points are dependent on the implementation
48 of the C library, and are not defined by GCC itself.
50 @node Identifiers implementation
55 @cite{Which additional multibyte characters may appear in identifiers
56 and their correspondence to universal character names (6.4.2).}
59 @cite{The number of significant initial characters in an identifier
63 @node Characters implementation
68 @cite{The number of bits in a byte (3.6).}
71 @cite{The values of the members of the execution character set (5.2.1).}
74 @cite{The unique value of the member of the execution character set produced
75 for each of the standard alphabetic escape sequences (5.2.2).}
78 @cite{The value of a @code{char} object into which has been stored any
79 character other than a member of the basic execution character set (6.2.5).}
82 @cite{Which of @code{signed char} or @code{unsigned char} has the same range,
83 representation, and behavior as ``plain'' @code{char} (6.2.5, 6.3.1.1).}
86 @cite{The mapping of members of the source character set (in character
87 constants and string literals) to members of the execution character
88 set (6.4.4.4, 5.1.1.2).}
91 @cite{The value of an integer character constant containing more than one
92 character or containing a character or escape sequence that does not map
93 to a single-byte execution character (6.4.4.4).}
96 @cite{The value of a wide character constant containing more than one
97 multibyte character, or containing a multibyte character or escape
98 sequence not represented in the extended execution character set (6.4.4.4).}
101 @cite{The current locale used to convert a wide character constant consisting
102 of a single multibyte character that maps to a member of the extended
103 execution character set into a corresponding wide character code (6.4.4.4).}
106 @cite{The current locale used to convert a wide string literal into
107 corresponding wide character codes (6.4.5).}
110 @cite{The value of a string literal containing a multibyte character or escape
111 sequence not represented in the execution character set (6.4.5).}
114 @node Integers implementation
119 @cite{Any extended integer types that exist in the implementation (6.2.5).}
122 @cite{Whether signed integer types are represented using sign and magnitude,
123 two's complement, or one's complement, and whether the extraordinary value
124 is a trap representation or an ordinary value (6.2.6.2).}
127 @cite{The rank of any extended integer type relative to another extended
128 integer type with the same precision (6.3.1.1).}
131 @cite{The result of, or the signal raised by, converting an integer to a
132 signed integer type when the value cannot be represented in an object of
133 that type (6.3.1.3).}
136 @cite{The results of some bitwise operations on signed integers (6.5).}
139 @node Floating point implementation
140 @section Floating point
144 @cite{The accuracy of the floating-point operations and of the library
145 functions in @code{<math.h>} and @code{<complex.h>} that return floating-point
146 results (5.2.4.2.2).}
149 @cite{The rounding behaviors characterized by non-standard values
150 of @code{FLT_ROUNDS} @gol
154 @cite{The evaluation methods characterized by non-standard negative
155 values of @code{FLT_EVAL_METHOD} (5.2.4.2.2).}
158 @cite{The direction of rounding when an integer is converted to a
159 floating-point number that cannot exactly represent the original
163 @cite{The direction of rounding when a floating-point number is
164 converted to a narrower floating-point number (6.3.1.5).}
167 @cite{How the nearest representable value or the larger or smaller
168 representable value immediately adjacent to the nearest representable
169 value is chosen for certain floating constants (6.4.4.2).}
172 @cite{Whether and how floating expressions are contracted when not
173 disallowed by the @code{FP_CONTRACT} pragma (6.5).}
176 @cite{The default state for the @code{FENV_ACCESS} pragma (7.6.1).}
179 @cite{Additional floating-point exceptions, rounding modes, environments,
180 and classifications, and their macro names (7.6, 7.12).}
183 @cite{The default state for the @code{FP_CONTRACT} pragma (7.12.2).}
186 @cite{Whether the ``inexact'' floating-point exception can be raised
187 when the rounded result actually does equal the mathematical result
188 in an IEC 60559 conformant implementation (F.9).}
191 @cite{Whether the ``underflow'' (and ``inexact'') floating-point
192 exception can be raised when a result is tiny but not inexact in an
193 IEC 60559 conformant implementation (F.9).}
197 @node Arrays and pointers implementation
198 @section Arrays and pointers
202 @cite{The result of converting a pointer to an integer or
203 vice versa (6.3.2.3).}
205 A cast from pointer to integer discards most-significant bits if the
206 pointer representation is larger than the integer type,
207 sign-extends@footnote{Future versions of GCC may zero-extend, or use
208 a target-defined @code{ptr_extend} pattern. Do not rely on sign extension.}
209 if the pointer representation is smaller than the integer type, otherwise
210 the bits are unchanged.
211 @c ??? We've always claimed that pointers were unsigned entities.
212 @c Shouldn't we therefore be doing zero-extension? If so, the bug
213 @c is in convert_to_integer, where we call type_for_size and request
214 @c a signed integral type. On the other hand, it might be most useful
215 @c for the target if we extend according to POINTERS_EXTEND_UNSIGNED.
217 A cast from integer to pointer discards most-significant bits if the
218 pointer representation is smaller than the integer type, extends according
219 to the signedness of the integer type if the pointer representation
220 is larger than the integer type, otherwise the bits are unchanged.
222 When casting from pointer to integer and back again, the resulting
223 pointer must reference the same object as the original pointer, otherwise
224 the behavior is undefined. That is, one may not use integer arithmetic to
225 avoid the undefined behavior of pointer arithmetic as proscribed in 6.5.6/8.
228 @cite{The size of the result of subtracting two pointers to elements
229 of the same array (6.5.6).}
233 @node Hints implementation
238 @cite{The extent to which suggestions made by using the @code{register}
239 storage-class specifier are effective (6.7.1).}
242 @cite{The extent to which suggestions made by using the inline function
243 specifier are effective (6.7.4).}
247 @node Structures unions enumerations and bit-fields implementation
248 @section Structures, unions, enumerations, and bit-fields
252 @cite{Whether a ``plain'' int bit-field is treated as a @code{signed int}
253 bit-field or as an @code{unsigned int} bit-field (6.7.2, 6.7.2.1).}
256 @cite{Allowable bit-field types other than @code{_Bool}, @code{signed int},
257 and @code{unsigned int} (6.7.2.1).}
260 @cite{Whether a bit-field can straddle a storage-unit boundary (6.7.2.1).}
263 @cite{The order of allocation of bit-fields within a unit (6.7.2.1).}
266 @cite{The alignment of non-bit-field members of structures (6.7.2.1).}
269 @cite{The integer type compatible with each enumerated type (6.7.2.2).}
273 @node Qualifiers implementation
278 @cite{What constitutes an access to an object that has volatile-qualified
283 @node Preprocessing directives implementation
284 @section Preprocessing directives
288 @cite{How sequences in both forms of header names are mapped to headers
289 or external source file names (6.4.7).}
292 @cite{Whether the value of a character constant in a constant expression
293 that controls conditional inclusion matches the value of the same character
294 constant in the execution character set (6.10.1).}
297 @cite{Whether the value of a single-character character constant in a
298 constant expression that controls conditional inclusion may have a
299 negative value (6.10.1).}
302 @cite{The places that are searched for an included @samp{<>} delimited
303 header, and how the places are specified or the header is
304 identified (6.10.2).}
307 @cite{How the named source file is searched for in an included @samp{""}
308 delimited header (6.10.2).}
311 @cite{The method by which preprocessing tokens (possibly resulting from
312 macro expansion) in a @code{#include} directive are combined into a header
316 @cite{The nesting limit for @code{#include} processing (6.10.2).}
319 @cite{Whether the @samp{#} operator inserts a @samp{\} character before
320 the @samp{\} character that begins a universal character name in a
321 character constant or string literal (6.10.3.2).}
324 @cite{The behavior on each recognized non-@code{STDC #pragma}
328 @cite{The definitions for @code{__DATE__} and @code{__TIME__} when
329 respectively, the date and time of translation are not available (6.10.8).}
333 @node Library functions implementation
334 @section Library functions
336 The behavior of these points are dependent on the implementation
337 of the C library, and are not defined by GCC itself.
339 @node Architecture implementation
340 @section Architecture
344 @cite{The values or expressions assigned to the macros specified in the
345 headers @code{<float.h>}, @code{<limits.h>}, and @code{<stdint.h>}
346 (5.2.4.2, 7.18.2, 7.18.3).}
349 @cite{The number, order, and encoding of bytes in any object
350 (when not explicitly specified in this International Standard) (6.2.6.1).}
353 @cite{The value of the result of the sizeof operator (6.5.3.4).}
357 @node Locale-specific behavior implementation
358 @section Locale-specific behavior
360 The behavior of these points are dependent on the implementation
361 of the C library, and are not defined by GCC itself.
364 @chapter Extensions to the C Language Family
365 @cindex extensions, C language
366 @cindex C language extensions
369 GNU C provides several language features not found in ISO standard C@.
370 (The @option{-pedantic} option directs GCC to print a warning message if
371 any of these features is used.) To test for the availability of these
372 features in conditional compilation, check for a predefined macro
373 @code{__GNUC__}, which is always defined under GCC@.
375 These extensions are available in C and Objective-C@. Most of them are
376 also available in C++. @xref{C++ Extensions,,Extensions to the
377 C++ Language}, for extensions that apply @emph{only} to C++.
379 Some features that are in ISO C99 but not C89 or C++ are also, as
380 extensions, accepted by GCC in C89 mode and in C++.
383 * Statement Exprs:: Putting statements and declarations inside expressions.
384 * Local Labels:: Labels local to a statement-expression.
385 * Labels as Values:: Getting pointers to labels, and computed gotos.
386 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
387 * Constructing Calls:: Dispatching a call to another function.
388 * Naming Types:: Giving a name to the type of some expression.
389 * Typeof:: @code{typeof}: referring to the type of an expression.
390 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
391 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
392 * Long Long:: Double-word integers---@code{long long int}.
393 * Complex:: Data types for complex numbers.
394 * Hex Floats:: Hexadecimal floating-point constants.
395 * Zero Length:: Zero-length arrays.
396 * Variable Length:: Arrays whose length is computed at run time.
397 * Variadic Macros:: Macros with a variable number of arguments.
398 * Escaped Newlines:: Slightly looser rules for escaped newlines.
399 * Multi-line Strings:: String literals with embedded newlines.
400 * Subscripting:: Any array can be subscripted, even if not an lvalue.
401 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
402 * Initializers:: Non-constant initializers.
403 * Compound Literals:: Compound literals give structures, unions
405 * Designated Inits:: Labeling elements of initializers.
406 * Cast to Union:: Casting to union type from any member of the union.
407 * Case Ranges:: `case 1 ... 9' and such.
408 * Mixed Declarations:: Mixing declarations and code.
409 * Function Attributes:: Declaring that functions have no side effects,
410 or that they can never return.
411 * Attribute Syntax:: Formal syntax for attributes.
412 * Function Prototypes:: Prototype declarations and old-style definitions.
413 * C++ Comments:: C++ comments are recognized.
414 * Dollar Signs:: Dollar sign is allowed in identifiers.
415 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
416 * Variable Attributes:: Specifying attributes of variables.
417 * Type Attributes:: Specifying attributes of types.
418 * Alignment:: Inquiring about the alignment of a type or variable.
419 * Inline:: Defining inline functions (as fast as macros).
420 * Extended Asm:: Assembler instructions with C expressions as operands.
421 (With them you can define ``built-in'' functions.)
422 * Constraints:: Constraints for asm operands
423 * Asm Labels:: Specifying the assembler name to use for a C symbol.
424 * Explicit Reg Vars:: Defining variables residing in specified registers.
425 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
426 * Incomplete Enums:: @code{enum foo;}, with details to follow.
427 * Function Names:: Printable strings which are the name of the current
429 * Return Address:: Getting the return or frame address of a function.
430 * Vector Extensions:: Using vector instructions through built-in functions.
431 * Other Builtins:: Other built-in functions.
432 * Target Builtins:: Built-in functions specific to particular targets.
433 * Pragmas:: Pragmas accepted by GCC.
434 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
435 * Thread-Local:: Per-thread variables.
438 @node Statement Exprs
439 @section Statements and Declarations in Expressions
440 @cindex statements inside expressions
441 @cindex declarations inside expressions
442 @cindex expressions containing statements
443 @cindex macros, statements in expressions
445 @c the above section title wrapped and causes an underfull hbox.. i
446 @c changed it from "within" to "in". --mew 4feb93
448 A compound statement enclosed in parentheses may appear as an expression
449 in GNU C@. This allows you to use loops, switches, and local variables
450 within an expression.
452 Recall that a compound statement is a sequence of statements surrounded
453 by braces; in this construct, parentheses go around the braces. For
457 (@{ int y = foo (); int z;
464 is a valid (though slightly more complex than necessary) expression
465 for the absolute value of @code{foo ()}.
467 The last thing in the compound statement should be an expression
468 followed by a semicolon; the value of this subexpression serves as the
469 value of the entire construct. (If you use some other kind of statement
470 last within the braces, the construct has type @code{void}, and thus
471 effectively no value.)
473 This feature is especially useful in making macro definitions ``safe'' (so
474 that they evaluate each operand exactly once). For example, the
475 ``maximum'' function is commonly defined as a macro in standard C as
479 #define max(a,b) ((a) > (b) ? (a) : (b))
483 @cindex side effects, macro argument
484 But this definition computes either @var{a} or @var{b} twice, with bad
485 results if the operand has side effects. In GNU C, if you know the
486 type of the operands (here let's assume @code{int}), you can define
487 the macro safely as follows:
490 #define maxint(a,b) \
491 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
494 Embedded statements are not allowed in constant expressions, such as
495 the value of an enumeration constant, the width of a bit-field, or
496 the initial value of a static variable.
498 If you don't know the type of the operand, you can still do this, but you
499 must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
502 Statement expressions are not supported fully in G++, and their fate
503 there is unclear. (It is possible that they will become fully supported
504 at some point, or that they will be deprecated, or that the bugs that
505 are present will continue to exist indefinitely.) Presently, statement
506 expressions do not work well as default arguments.
508 In addition, there are semantic issues with statement-expressions in
509 C++. If you try to use statement-expressions instead of inline
510 functions in C++, you may be surprised at the way object destruction is
511 handled. For example:
514 #define foo(a) (@{int b = (a); b + 3; @})
518 does not work the same way as:
521 inline int foo(int a) @{ int b = a; return b + 3; @}
525 In particular, if the expression passed into @code{foo} involves the
526 creation of temporaries, the destructors for those temporaries will be
527 run earlier in the case of the macro than in the case of the function.
529 These considerations mean that it is probably a bad idea to use
530 statement-expressions of this form in header files that are designed to
531 work with C++. (Note that some versions of the GNU C Library contained
532 header files using statement-expression that lead to precisely this
536 @section Locally Declared Labels
538 @cindex macros, local labels
540 Each statement expression is a scope in which @dfn{local labels} can be
541 declared. A local label is simply an identifier; you can jump to it
542 with an ordinary @code{goto} statement, but only from within the
543 statement expression it belongs to.
545 A local label declaration looks like this:
548 __label__ @var{label};
555 __label__ @var{label1}, @var{label2}, @dots{};
558 Local label declarations must come at the beginning of the statement
559 expression, right after the @samp{(@{}, before any ordinary
562 The label declaration defines the label @emph{name}, but does not define
563 the label itself. You must do this in the usual way, with
564 @code{@var{label}:}, within the statements of the statement expression.
566 The local label feature is useful because statement expressions are
567 often used in macros. If the macro contains nested loops, a @code{goto}
568 can be useful for breaking out of them. However, an ordinary label
569 whose scope is the whole function cannot be used: if the macro can be
570 expanded several times in one function, the label will be multiply
571 defined in that function. A local label avoids this problem. For
575 #define SEARCH(array, target) \
578 typeof (target) _SEARCH_target = (target); \
579 typeof (*(array)) *_SEARCH_array = (array); \
582 for (i = 0; i < max; i++) \
583 for (j = 0; j < max; j++) \
584 if (_SEARCH_array[i][j] == _SEARCH_target) \
585 @{ value = i; goto found; @} \
592 @node Labels as Values
593 @section Labels as Values
594 @cindex labels as values
595 @cindex computed gotos
596 @cindex goto with computed label
597 @cindex address of a label
599 You can get the address of a label defined in the current function
600 (or a containing function) with the unary operator @samp{&&}. The
601 value has type @code{void *}. This value is a constant and can be used
602 wherever a constant of that type is valid. For example:
610 To use these values, you need to be able to jump to one. This is done
611 with the computed goto statement@footnote{The analogous feature in
612 Fortran is called an assigned goto, but that name seems inappropriate in
613 C, where one can do more than simply store label addresses in label
614 variables.}, @code{goto *@var{exp};}. For example,
621 Any expression of type @code{void *} is allowed.
623 One way of using these constants is in initializing a static array that
624 will serve as a jump table:
627 static void *array[] = @{ &&foo, &&bar, &&hack @};
630 Then you can select a label with indexing, like this:
637 Note that this does not check whether the subscript is in bounds---array
638 indexing in C never does that.
640 Such an array of label values serves a purpose much like that of the
641 @code{switch} statement. The @code{switch} statement is cleaner, so
642 use that rather than an array unless the problem does not fit a
643 @code{switch} statement very well.
645 Another use of label values is in an interpreter for threaded code.
646 The labels within the interpreter function can be stored in the
647 threaded code for super-fast dispatching.
649 You may not use this mechanism to jump to code in a different function.
650 If you do that, totally unpredictable things will happen. The best way to
651 avoid this is to store the label address only in automatic variables and
652 never pass it as an argument.
654 An alternate way to write the above example is
657 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
659 goto *(&&foo + array[i]);
663 This is more friendly to code living in shared libraries, as it reduces
664 the number of dynamic relocations that are needed, and by consequence,
665 allows the data to be read-only.
667 @node Nested Functions
668 @section Nested Functions
669 @cindex nested functions
670 @cindex downward funargs
673 A @dfn{nested function} is a function defined inside another function.
674 (Nested functions are not supported for GNU C++.) The nested function's
675 name is local to the block where it is defined. For example, here we
676 define a nested function named @code{square}, and call it twice:
680 foo (double a, double b)
682 double square (double z) @{ return z * z; @}
684 return square (a) + square (b);
689 The nested function can access all the variables of the containing
690 function that are visible at the point of its definition. This is
691 called @dfn{lexical scoping}. For example, here we show a nested
692 function which uses an inherited variable named @code{offset}:
696 bar (int *array, int offset, int size)
698 int access (int *array, int index)
699 @{ return array[index + offset]; @}
702 for (i = 0; i < size; i++)
703 @dots{} access (array, i) @dots{}
708 Nested function definitions are permitted within functions in the places
709 where variable definitions are allowed; that is, in any block, before
710 the first statement in the block.
712 It is possible to call the nested function from outside the scope of its
713 name by storing its address or passing the address to another function:
716 hack (int *array, int size)
718 void store (int index, int value)
719 @{ array[index] = value; @}
721 intermediate (store, size);
725 Here, the function @code{intermediate} receives the address of
726 @code{store} as an argument. If @code{intermediate} calls @code{store},
727 the arguments given to @code{store} are used to store into @code{array}.
728 But this technique works only so long as the containing function
729 (@code{hack}, in this example) does not exit.
731 If you try to call the nested function through its address after the
732 containing function has exited, all hell will break loose. If you try
733 to call it after a containing scope level has exited, and if it refers
734 to some of the variables that are no longer in scope, you may be lucky,
735 but it's not wise to take the risk. If, however, the nested function
736 does not refer to anything that has gone out of scope, you should be
739 GCC implements taking the address of a nested function using a technique
740 called @dfn{trampolines}. A paper describing them is available as
743 @uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}.
745 A nested function can jump to a label inherited from a containing
746 function, provided the label was explicitly declared in the containing
747 function (@pxref{Local Labels}). Such a jump returns instantly to the
748 containing function, exiting the nested function which did the
749 @code{goto} and any intermediate functions as well. Here is an example:
753 bar (int *array, int offset, int size)
756 int access (int *array, int index)
760 return array[index + offset];
764 for (i = 0; i < size; i++)
765 @dots{} access (array, i) @dots{}
769 /* @r{Control comes here from @code{access}
770 if it detects an error.} */
777 A nested function always has internal linkage. Declaring one with
778 @code{extern} is erroneous. If you need to declare the nested function
779 before its definition, use @code{auto} (which is otherwise meaningless
780 for function declarations).
783 bar (int *array, int offset, int size)
786 auto int access (int *, int);
788 int access (int *array, int index)
792 return array[index + offset];
798 @node Constructing Calls
799 @section Constructing Function Calls
800 @cindex constructing calls
801 @cindex forwarding calls
803 Using the built-in functions described below, you can record
804 the arguments a function received, and call another function
805 with the same arguments, without knowing the number or types
808 You can also record the return value of that function call,
809 and later return that value, without knowing what data type
810 the function tried to return (as long as your caller expects
813 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
814 This built-in function returns a pointer to data
815 describing how to perform a call with the same arguments as were passed
816 to the current function.
818 The function saves the arg pointer register, structure value address,
819 and all registers that might be used to pass arguments to a function
820 into a block of memory allocated on the stack. Then it returns the
821 address of that block.
824 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
825 This built-in function invokes @var{function}
826 with a copy of the parameters described by @var{arguments}
829 The value of @var{arguments} should be the value returned by
830 @code{__builtin_apply_args}. The argument @var{size} specifies the size
831 of the stack argument data, in bytes.
833 This function returns a pointer to data describing
834 how to return whatever value was returned by @var{function}. The data
835 is saved in a block of memory allocated on the stack.
837 It is not always simple to compute the proper value for @var{size}. The
838 value is used by @code{__builtin_apply} to compute the amount of data
839 that should be pushed on the stack and copied from the incoming argument
843 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
844 This built-in function returns the value described by @var{result} from
845 the containing function. You should specify, for @var{result}, a value
846 returned by @code{__builtin_apply}.
850 @section Naming an Expression's Type
853 You can give a name to the type of an expression using a @code{typedef}
854 declaration with an initializer. Here is how to define @var{name} as a
855 type name for the type of @var{exp}:
858 typedef @var{name} = @var{exp};
861 This is useful in conjunction with the statements-within-expressions
862 feature. Here is how the two together can be used to define a safe
863 ``maximum'' macro that operates on any arithmetic type:
867 (@{typedef _ta = (a), _tb = (b); \
868 _ta _a = (a); _tb _b = (b); \
869 _a > _b ? _a : _b; @})
872 @cindex underscores in variables in macros
873 @cindex @samp{_} in variables in macros
874 @cindex local variables in macros
875 @cindex variables, local, in macros
876 @cindex macros, local variables in
878 The reason for using names that start with underscores for the local
879 variables is to avoid conflicts with variable names that occur within the
880 expressions that are substituted for @code{a} and @code{b}. Eventually we
881 hope to design a new form of declaration syntax that allows you to declare
882 variables whose scopes start only after their initializers; this will be a
883 more reliable way to prevent such conflicts.
886 @section Referring to a Type with @code{typeof}
889 @cindex macros, types of arguments
891 Another way to refer to the type of an expression is with @code{typeof}.
892 The syntax of using of this keyword looks like @code{sizeof}, but the
893 construct acts semantically like a type name defined with @code{typedef}.
895 There are two ways of writing the argument to @code{typeof}: with an
896 expression or with a type. Here is an example with an expression:
903 This assumes that @code{x} is an array of pointers to functions;
904 the type described is that of the values of the functions.
906 Here is an example with a typename as the argument:
913 Here the type described is that of pointers to @code{int}.
915 If you are writing a header file that must work when included in ISO C
916 programs, write @code{__typeof__} instead of @code{typeof}.
917 @xref{Alternate Keywords}.
919 A @code{typeof}-construct can be used anywhere a typedef name could be
920 used. For example, you can use it in a declaration, in a cast, or inside
921 of @code{sizeof} or @code{typeof}.
925 This declares @code{y} with the type of what @code{x} points to.
932 This declares @code{y} as an array of such values.
939 This declares @code{y} as an array of pointers to characters:
942 typeof (typeof (char *)[4]) y;
946 It is equivalent to the following traditional C declaration:
952 To see the meaning of the declaration using @code{typeof}, and why it
953 might be a useful way to write, let's rewrite it with these macros:
956 #define pointer(T) typeof(T *)
957 #define array(T, N) typeof(T [N])
961 Now the declaration can be rewritten this way:
964 array (pointer (char), 4) y;
968 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
969 pointers to @code{char}.
973 @section Generalized Lvalues
974 @cindex compound expressions as lvalues
975 @cindex expressions, compound, as lvalues
976 @cindex conditional expressions as lvalues
977 @cindex expressions, conditional, as lvalues
978 @cindex casts as lvalues
979 @cindex generalized lvalues
980 @cindex lvalues, generalized
981 @cindex extensions, @code{?:}
982 @cindex @code{?:} extensions
983 Compound expressions, conditional expressions and casts are allowed as
984 lvalues provided their operands are lvalues. This means that you can take
985 their addresses or store values into them.
987 Standard C++ allows compound expressions and conditional expressions as
988 lvalues, and permits casts to reference type, so use of this extension
989 is deprecated for C++ code.
991 For example, a compound expression can be assigned, provided the last
992 expression in the sequence is an lvalue. These two expressions are
1000 Similarly, the address of the compound expression can be taken. These two
1001 expressions are equivalent:
1008 A conditional expression is a valid lvalue if its type is not void and the
1009 true and false branches are both valid lvalues. For example, these two
1010 expressions are equivalent:
1014 (a ? b = 5 : (c = 5))
1017 A cast is a valid lvalue if its operand is an lvalue. A simple
1018 assignment whose left-hand side is a cast works by converting the
1019 right-hand side first to the specified type, then to the type of the
1020 inner left-hand side expression. After this is stored, the value is
1021 converted back to the specified type to become the value of the
1022 assignment. Thus, if @code{a} has type @code{char *}, the following two
1023 expressions are equivalent:
1027 (int)(a = (char *)(int)5)
1030 An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1031 performs the arithmetic using the type resulting from the cast, and then
1032 continues as in the previous case. Therefore, these two expressions are
1037 (int)(a = (char *)(int) ((int)a + 5))
1040 You cannot take the address of an lvalue cast, because the use of its
1041 address would not work out coherently. Suppose that @code{&(int)f} were
1042 permitted, where @code{f} has type @code{float}. Then the following
1043 statement would try to store an integer bit-pattern where a floating
1044 point number belongs:
1050 This is quite different from what @code{(int)f = 1} would do---that
1051 would convert 1 to floating point and store it. Rather than cause this
1052 inconsistency, we think it is better to prohibit use of @samp{&} on a cast.
1054 If you really do want an @code{int *} pointer with the address of
1055 @code{f}, you can simply write @code{(int *)&f}.
1058 @section Conditionals with Omitted Operands
1059 @cindex conditional expressions, extensions
1060 @cindex omitted middle-operands
1061 @cindex middle-operands, omitted
1062 @cindex extensions, @code{?:}
1063 @cindex @code{?:} extensions
1065 The middle operand in a conditional expression may be omitted. Then
1066 if the first operand is nonzero, its value is the value of the conditional
1069 Therefore, the expression
1076 has the value of @code{x} if that is nonzero; otherwise, the value of
1079 This example is perfectly equivalent to
1085 @cindex side effect in ?:
1086 @cindex ?: side effect
1088 In this simple case, the ability to omit the middle operand is not
1089 especially useful. When it becomes useful is when the first operand does,
1090 or may (if it is a macro argument), contain a side effect. Then repeating
1091 the operand in the middle would perform the side effect twice. Omitting
1092 the middle operand uses the value already computed without the undesirable
1093 effects of recomputing it.
1096 @section Double-Word Integers
1097 @cindex @code{long long} data types
1098 @cindex double-word arithmetic
1099 @cindex multiprecision arithmetic
1100 @cindex @code{LL} integer suffix
1101 @cindex @code{ULL} integer suffix
1103 ISO C99 supports data types for integers that are at least 64 bits wide,
1104 and as an extension GCC supports them in C89 mode and in C++.
1105 Simply write @code{long long int} for a signed integer, or
1106 @code{unsigned long long int} for an unsigned integer. To make an
1107 integer constant of type @code{long long int}, add the suffix @samp{LL}
1108 to the integer. To make an integer constant of type @code{unsigned long
1109 long int}, add the suffix @samp{ULL} to the integer.
1111 You can use these types in arithmetic like any other integer types.
1112 Addition, subtraction, and bitwise boolean operations on these types
1113 are open-coded on all types of machines. Multiplication is open-coded
1114 if the machine supports fullword-to-doubleword a widening multiply
1115 instruction. Division and shifts are open-coded only on machines that
1116 provide special support. The operations that are not open-coded use
1117 special library routines that come with GCC@.
1119 There may be pitfalls when you use @code{long long} types for function
1120 arguments, unless you declare function prototypes. If a function
1121 expects type @code{int} for its argument, and you pass a value of type
1122 @code{long long int}, confusion will result because the caller and the
1123 subroutine will disagree about the number of bytes for the argument.
1124 Likewise, if the function expects @code{long long int} and you pass
1125 @code{int}. The best way to avoid such problems is to use prototypes.
1128 @section Complex Numbers
1129 @cindex complex numbers
1130 @cindex @code{_Complex} keyword
1131 @cindex @code{__complex__} keyword
1133 ISO C99 supports complex floating data types, and as an extension GCC
1134 supports them in C89 mode and in C++, and supports complex integer data
1135 types which are not part of ISO C99. You can declare complex types
1136 using the keyword @code{_Complex}. As an extension, the older GNU
1137 keyword @code{__complex__} is also supported.
1139 For example, @samp{_Complex double x;} declares @code{x} as a
1140 variable whose real part and imaginary part are both of type
1141 @code{double}. @samp{_Complex short int y;} declares @code{y} to
1142 have real and imaginary parts of type @code{short int}; this is not
1143 likely to be useful, but it shows that the set of complex types is
1146 To write a constant with a complex data type, use the suffix @samp{i} or
1147 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
1148 has type @code{_Complex float} and @code{3i} has type
1149 @code{_Complex int}. Such a constant always has a pure imaginary
1150 value, but you can form any complex value you like by adding one to a
1151 real constant. This is a GNU extension; if you have an ISO C99
1152 conforming C library (such as GNU libc), and want to construct complex
1153 constants of floating type, you should include @code{<complex.h>} and
1154 use the macros @code{I} or @code{_Complex_I} instead.
1156 @cindex @code{__real__} keyword
1157 @cindex @code{__imag__} keyword
1158 To extract the real part of a complex-valued expression @var{exp}, write
1159 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
1160 extract the imaginary part. This is a GNU extension; for values of
1161 floating type, you should use the ISO C99 functions @code{crealf},
1162 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
1163 @code{cimagl}, declared in @code{<complex.h>} and also provided as
1164 built-in functions by GCC@.
1166 @cindex complex conjugation
1167 The operator @samp{~} performs complex conjugation when used on a value
1168 with a complex type. This is a GNU extension; for values of
1169 floating type, you should use the ISO C99 functions @code{conjf},
1170 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
1171 provided as built-in functions by GCC@.
1173 GCC can allocate complex automatic variables in a noncontiguous
1174 fashion; it's even possible for the real part to be in a register while
1175 the imaginary part is on the stack (or vice-versa). None of the
1176 supported debugging info formats has a way to represent noncontiguous
1177 allocation like this, so GCC describes a noncontiguous complex
1178 variable as if it were two separate variables of noncomplex type.
1179 If the variable's actual name is @code{foo}, the two fictitious
1180 variables are named @code{foo$real} and @code{foo$imag}. You can
1181 examine and set these two fictitious variables with your debugger.
1183 A future version of GDB will know how to recognize such pairs and treat
1184 them as a single variable with a complex type.
1190 ISO C99 supports floating-point numbers written not only in the usual
1191 decimal notation, such as @code{1.55e1}, but also numbers such as
1192 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1193 supports this in C89 mode (except in some cases when strictly
1194 conforming) and in C++. In that format the
1195 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1196 mandatory. The exponent is a decimal number that indicates the power of
1197 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1204 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1205 is the same as @code{1.55e1}.
1207 Unlike for floating-point numbers in the decimal notation the exponent
1208 is always required in the hexadecimal notation. Otherwise the compiler
1209 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1210 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1211 extension for floating-point constants of type @code{float}.
1214 @section Arrays of Length Zero
1215 @cindex arrays of length zero
1216 @cindex zero-length arrays
1217 @cindex length-zero arrays
1218 @cindex flexible array members
1220 Zero-length arrays are allowed in GNU C@. They are very useful as the
1221 last element of a structure which is really a header for a variable-length
1230 struct line *thisline = (struct line *)
1231 malloc (sizeof (struct line) + this_length);
1232 thisline->length = this_length;
1235 In ISO C89, you would have to give @code{contents} a length of 1, which
1236 means either you waste space or complicate the argument to @code{malloc}.
1238 In ISO C99, you would use a @dfn{flexible array member}, which is
1239 slightly different in syntax and semantics:
1243 Flexible array members are written as @code{contents[]} without
1247 Flexible array members have incomplete type, and so the @code{sizeof}
1248 operator may not be applied. As a quirk of the original implementation
1249 of zero-length arrays, @code{sizeof} evaluates to zero.
1252 Flexible array members may only appear as the last member of a
1253 @code{struct} that is otherwise non-empty.
1256 GCC versions before 3.0 allowed zero-length arrays to be statically
1257 initialized, as if they were flexible arrays. In addition to those
1258 cases that were useful, it also allowed initializations in situations
1259 that would corrupt later data. Non-empty initialization of zero-length
1260 arrays is now treated like any case where there are more initializer
1261 elements than the array holds, in that a suitable warning about "excess
1262 elements in array" is given, and the excess elements (all of them, in
1263 this case) are ignored.
1265 Instead GCC allows static initialization of flexible array members.
1266 This is equivalent to defining a new structure containing the original
1267 structure followed by an array of sufficient size to contain the data.
1268 I.e.@: in the following, @code{f1} is constructed as if it were declared
1274 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1277 struct f1 f1; int data[3];
1278 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1282 The convenience of this extension is that @code{f1} has the desired
1283 type, eliminating the need to consistently refer to @code{f2.f1}.
1285 This has symmetry with normal static arrays, in that an array of
1286 unknown size is also written with @code{[]}.
1288 Of course, this extension only makes sense if the extra data comes at
1289 the end of a top-level object, as otherwise we would be overwriting
1290 data at subsequent offsets. To avoid undue complication and confusion
1291 with initialization of deeply nested arrays, we simply disallow any
1292 non-empty initialization except when the structure is the top-level
1293 object. For example:
1296 struct foo @{ int x; int y[]; @};
1297 struct bar @{ struct foo z; @};
1299 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1300 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1301 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1302 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1305 @node Variable Length
1306 @section Arrays of Variable Length
1307 @cindex variable-length arrays
1308 @cindex arrays of variable length
1311 Variable-length automatic arrays are allowed in ISO C99, and as an
1312 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1313 implementation of variable-length arrays does not yet conform in detail
1314 to the ISO C99 standard.) These arrays are
1315 declared like any other automatic arrays, but with a length that is not
1316 a constant expression. The storage is allocated at the point of
1317 declaration and deallocated when the brace-level is exited. For
1322 concat_fopen (char *s1, char *s2, char *mode)
1324 char str[strlen (s1) + strlen (s2) + 1];
1327 return fopen (str, mode);
1331 @cindex scope of a variable length array
1332 @cindex variable-length array scope
1333 @cindex deallocating variable length arrays
1334 Jumping or breaking out of the scope of the array name deallocates the
1335 storage. Jumping into the scope is not allowed; you get an error
1338 @cindex @code{alloca} vs variable-length arrays
1339 You can use the function @code{alloca} to get an effect much like
1340 variable-length arrays. The function @code{alloca} is available in
1341 many other C implementations (but not in all). On the other hand,
1342 variable-length arrays are more elegant.
1344 There are other differences between these two methods. Space allocated
1345 with @code{alloca} exists until the containing @emph{function} returns.
1346 The space for a variable-length array is deallocated as soon as the array
1347 name's scope ends. (If you use both variable-length arrays and
1348 @code{alloca} in the same function, deallocation of a variable-length array
1349 will also deallocate anything more recently allocated with @code{alloca}.)
1351 You can also use variable-length arrays as arguments to functions:
1355 tester (int len, char data[len][len])
1361 The length of an array is computed once when the storage is allocated
1362 and is remembered for the scope of the array in case you access it with
1365 If you want to pass the array first and the length afterward, you can
1366 use a forward declaration in the parameter list---another GNU extension.
1370 tester (int len; char data[len][len], int len)
1376 @cindex parameter forward declaration
1377 The @samp{int len} before the semicolon is a @dfn{parameter forward
1378 declaration}, and it serves the purpose of making the name @code{len}
1379 known when the declaration of @code{data} is parsed.
1381 You can write any number of such parameter forward declarations in the
1382 parameter list. They can be separated by commas or semicolons, but the
1383 last one must end with a semicolon, which is followed by the ``real''
1384 parameter declarations. Each forward declaration must match a ``real''
1385 declaration in parameter name and data type. ISO C99 does not support
1386 parameter forward declarations.
1388 @node Variadic Macros
1389 @section Macros with a Variable Number of Arguments.
1390 @cindex variable number of arguments
1391 @cindex macro with variable arguments
1392 @cindex rest argument (in macro)
1393 @cindex variadic macros
1395 In the ISO C standard of 1999, a macro can be declared to accept a
1396 variable number of arguments much as a function can. The syntax for
1397 defining the macro is similar to that of a function. Here is an
1401 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1404 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1405 such a macro, it represents the zero or more tokens until the closing
1406 parenthesis that ends the invocation, including any commas. This set of
1407 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1408 wherever it appears. See the CPP manual for more information.
1410 GCC has long supported variadic macros, and used a different syntax that
1411 allowed you to give a name to the variable arguments just like any other
1412 argument. Here is an example:
1415 #define debug(format, args...) fprintf (stderr, format, args)
1418 This is in all ways equivalent to the ISO C example above, but arguably
1419 more readable and descriptive.
1421 GNU CPP has two further variadic macro extensions, and permits them to
1422 be used with either of the above forms of macro definition.
1424 In standard C, you are not allowed to leave the variable argument out
1425 entirely; but you are allowed to pass an empty argument. For example,
1426 this invocation is invalid in ISO C, because there is no comma after
1433 GNU CPP permits you to completely omit the variable arguments in this
1434 way. In the above examples, the compiler would complain, though since
1435 the expansion of the macro still has the extra comma after the format
1438 To help solve this problem, CPP behaves specially for variable arguments
1439 used with the token paste operator, @samp{##}. If instead you write
1442 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1445 and if the variable arguments are omitted or empty, the @samp{##}
1446 operator causes the preprocessor to remove the comma before it. If you
1447 do provide some variable arguments in your macro invocation, GNU CPP
1448 does not complain about the paste operation and instead places the
1449 variable arguments after the comma. Just like any other pasted macro
1450 argument, these arguments are not macro expanded.
1452 @node Escaped Newlines
1453 @section Slightly Looser Rules for Escaped Newlines
1454 @cindex escaped newlines
1455 @cindex newlines (escaped)
1457 Recently, the preprocessor has relaxed its treatment of escaped
1458 newlines. Previously, the newline had to immediately follow a
1459 backslash. The current implementation allows whitespace in the form of
1460 spaces, horizontal and vertical tabs, and form feeds between the
1461 backslash and the subsequent newline. The preprocessor issues a
1462 warning, but treats it as a valid escaped newline and combines the two
1463 lines to form a single logical line. This works within comments and
1464 tokens, including multi-line strings, as well as between tokens.
1465 Comments are @emph{not} treated as whitespace for the purposes of this
1466 relaxation, since they have not yet been replaced with spaces.
1468 @node Multi-line Strings
1469 @section String Literals with Embedded Newlines
1470 @cindex multi-line string literals
1472 As an extension, GNU CPP permits string literals to cross multiple lines
1473 without escaping the embedded newlines. Each embedded newline is
1474 replaced with a single @samp{\n} character in the resulting string
1475 literal, regardless of what form the newline took originally.
1477 CPP currently allows such strings in directives as well (other than the
1478 @samp{#include} family). This is deprecated and will eventually be
1482 @section Non-Lvalue Arrays May Have Subscripts
1483 @cindex subscripting
1484 @cindex arrays, non-lvalue
1486 @cindex subscripting and function values
1487 In ISO C99, arrays that are not lvalues still decay to pointers, and
1488 may be subscripted, although they may not be modified or used after
1489 the next sequence point and the unary @samp{&} operator may not be
1490 applied to them. As an extension, GCC allows such arrays to be
1491 subscripted in C89 mode, though otherwise they do not decay to
1492 pointers outside C99 mode. For example,
1493 this is valid in GNU C though not valid in C89:
1497 struct foo @{int a[4];@};
1503 return f().a[index];
1509 @section Arithmetic on @code{void}- and Function-Pointers
1510 @cindex void pointers, arithmetic
1511 @cindex void, size of pointer to
1512 @cindex function pointers, arithmetic
1513 @cindex function, size of pointer to
1515 In GNU C, addition and subtraction operations are supported on pointers to
1516 @code{void} and on pointers to functions. This is done by treating the
1517 size of a @code{void} or of a function as 1.
1519 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1520 and on function types, and returns 1.
1522 @opindex Wpointer-arith
1523 The option @option{-Wpointer-arith} requests a warning if these extensions
1527 @section Non-Constant Initializers
1528 @cindex initializers, non-constant
1529 @cindex non-constant initializers
1531 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1532 automatic variable are not required to be constant expressions in GNU C@.
1533 Here is an example of an initializer with run-time varying elements:
1536 foo (float f, float g)
1538 float beat_freqs[2] = @{ f-g, f+g @};
1543 @node Compound Literals
1544 @section Compound Literals
1545 @cindex constructor expressions
1546 @cindex initializations in expressions
1547 @cindex structures, constructor expression
1548 @cindex expressions, constructor
1549 @cindex compound literals
1550 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1552 ISO C99 supports compound literals. A compound literal looks like
1553 a cast containing an initializer. Its value is an object of the
1554 type specified in the cast, containing the elements specified in
1555 the initializer; it is an lvalue. As an extension, GCC supports
1556 compound literals in C89 mode and in C++.
1558 Usually, the specified type is a structure. Assume that
1559 @code{struct foo} and @code{structure} are declared as shown:
1562 struct foo @{int a; char b[2];@} structure;
1566 Here is an example of constructing a @code{struct foo} with a compound literal:
1569 structure = ((struct foo) @{x + y, 'a', 0@});
1573 This is equivalent to writing the following:
1577 struct foo temp = @{x + y, 'a', 0@};
1582 You can also construct an array. If all the elements of the compound literal
1583 are (made up of) simple constant expressions, suitable for use in
1584 initializers of objects of static storage duration, then the compound
1585 literal can be coerced to a pointer to its first element and used in
1586 such an initializer, as shown here:
1589 char **foo = (char *[]) @{ "x", "y", "z" @};
1592 Compound literals for scalar types and union types are is
1593 also allowed, but then the compound literal is equivalent
1596 As a GNU extension, GCC allows initialization of objects with static storage
1597 duration by compound literals (which is not possible in ISO C99, because
1598 the initializer is not a constant).
1599 It is handled as if the object was initialized only with the bracket
1600 enclosed list if compound literal's and object types match.
1601 The initializer list of the compound literal must be constant.
1602 If the object being initialized has array type of unknown size, the size is
1603 determined by compound literal size.
1606 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1607 static int y[] = (int []) @{1, 2, 3@};
1608 static int z[] = (int [3]) @{1@};
1612 The above lines are equivalent to the following:
1614 static struct foo x = @{1, 'a', 'b'@};
1615 static int y[] = @{1, 2, 3@};
1616 static int z[] = @{1, 0, 0@};
1619 @node Designated Inits
1620 @section Designated Initializers
1621 @cindex initializers with labeled elements
1622 @cindex labeled elements in initializers
1623 @cindex case labels in initializers
1624 @cindex designated initializers
1626 Standard C89 requires the elements of an initializer to appear in a fixed
1627 order, the same as the order of the elements in the array or structure
1630 In ISO C99 you can give the elements in any order, specifying the array
1631 indices or structure field names they apply to, and GNU C allows this as
1632 an extension in C89 mode as well. This extension is not
1633 implemented in GNU C++.
1635 To specify an array index, write
1636 @samp{[@var{index}] =} before the element value. For example,
1639 int a[6] = @{ [4] = 29, [2] = 15 @};
1646 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1650 The index values must be constant expressions, even if the array being
1651 initialized is automatic.
1653 An alternative syntax for this which has been obsolete since GCC 2.5 but
1654 GCC still accepts is to write @samp{[@var{index}]} before the element
1655 value, with no @samp{=}.
1657 To initialize a range of elements to the same value, write
1658 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1659 extension. For example,
1662 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1666 If the value in it has side-effects, the side-effects will happen only once,
1667 not for each initialized field by the range initializer.
1670 Note that the length of the array is the highest value specified
1673 In a structure initializer, specify the name of a field to initialize
1674 with @samp{.@var{fieldname} =} before the element value. For example,
1675 given the following structure,
1678 struct point @{ int x, y; @};
1682 the following initialization
1685 struct point p = @{ .y = yvalue, .x = xvalue @};
1692 struct point p = @{ xvalue, yvalue @};
1695 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1696 @samp{@var{fieldname}:}, as shown here:
1699 struct point p = @{ y: yvalue, x: xvalue @};
1703 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1704 @dfn{designator}. You can also use a designator (or the obsolete colon
1705 syntax) when initializing a union, to specify which element of the union
1706 should be used. For example,
1709 union foo @{ int i; double d; @};
1711 union foo f = @{ .d = 4 @};
1715 will convert 4 to a @code{double} to store it in the union using
1716 the second element. By contrast, casting 4 to type @code{union foo}
1717 would store it into the union as the integer @code{i}, since it is
1718 an integer. (@xref{Cast to Union}.)
1720 You can combine this technique of naming elements with ordinary C
1721 initialization of successive elements. Each initializer element that
1722 does not have a designator applies to the next consecutive element of the
1723 array or structure. For example,
1726 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1733 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1736 Labeling the elements of an array initializer is especially useful
1737 when the indices are characters or belong to an @code{enum} type.
1742 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1743 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1746 @cindex designator lists
1747 You can also write a series of @samp{.@var{fieldname}} and
1748 @samp{[@var{index}]} designators before an @samp{=} to specify a
1749 nested subobject to initialize; the list is taken relative to the
1750 subobject corresponding to the closest surrounding brace pair. For
1751 example, with the @samp{struct point} declaration above:
1754 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1758 If the same field is initialized multiple times, it will have value from
1759 the last initialization. If any such overridden initialization has
1760 side-effect, it is unspecified whether the side-effect happens or not.
1761 Currently, gcc will discard them and issue a warning.
1764 @section Case Ranges
1766 @cindex ranges in case statements
1768 You can specify a range of consecutive values in a single @code{case} label,
1772 case @var{low} ... @var{high}:
1776 This has the same effect as the proper number of individual @code{case}
1777 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1779 This feature is especially useful for ranges of ASCII character codes:
1785 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1786 it may be parsed wrong when you use it with integer values. For example,
1801 @section Cast to a Union Type
1802 @cindex cast to a union
1803 @cindex union, casting to a
1805 A cast to union type is similar to other casts, except that the type
1806 specified is a union type. You can specify the type either with
1807 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1808 a constructor though, not a cast, and hence does not yield an lvalue like
1809 normal casts. (@xref{Compound Literals}.)
1811 The types that may be cast to the union type are those of the members
1812 of the union. Thus, given the following union and variables:
1815 union foo @{ int i; double d; @};
1821 both @code{x} and @code{y} can be cast to type @code{union foo}.
1823 Using the cast as the right-hand side of an assignment to a variable of
1824 union type is equivalent to storing in a member of the union:
1829 u = (union foo) x @equiv{} u.i = x
1830 u = (union foo) y @equiv{} u.d = y
1833 You can also use the union cast as a function argument:
1836 void hack (union foo);
1838 hack ((union foo) x);
1841 @node Mixed Declarations
1842 @section Mixed Declarations and Code
1843 @cindex mixed declarations and code
1844 @cindex declarations, mixed with code
1845 @cindex code, mixed with declarations
1847 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1848 within compound statements. As an extension, GCC also allows this in
1849 C89 mode. For example, you could do:
1858 Each identifier is visible from where it is declared until the end of
1859 the enclosing block.
1861 @node Function Attributes
1862 @section Declaring Attributes of Functions
1863 @cindex function attributes
1864 @cindex declaring attributes of functions
1865 @cindex functions that never return
1866 @cindex functions that have no side effects
1867 @cindex functions in arbitrary sections
1868 @cindex functions that behave like malloc
1869 @cindex @code{volatile} applied to function
1870 @cindex @code{const} applied to function
1871 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1872 @cindex functions with non-null pointer arguments
1873 @cindex functions that are passed arguments in registers on the 386
1874 @cindex functions that pop the argument stack on the 386
1875 @cindex functions that do not pop the argument stack on the 386
1877 In GNU C, you declare certain things about functions called in your program
1878 which help the compiler optimize function calls and check your code more
1881 The keyword @code{__attribute__} allows you to specify special
1882 attributes when making a declaration. This keyword is followed by an
1883 attribute specification inside double parentheses. The following
1884 attributes are currently defined for functions on all targets:
1885 @code{noreturn}, @code{noinline}, @code{always_inline},
1886 @code{pure}, @code{const}, @code{nothrow},
1887 @code{format}, @code{format_arg}, @code{no_instrument_function},
1888 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1889 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1890 @code{alias}, and @code{nonnull}. Several other attributes are defined
1891 for functions on particular target systems. Other attributes, including
1892 @code{section} are supported for variables declarations
1893 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1895 You may also specify attributes with @samp{__} preceding and following
1896 each keyword. This allows you to use them in header files without
1897 being concerned about a possible macro of the same name. For example,
1898 you may use @code{__noreturn__} instead of @code{noreturn}.
1900 @xref{Attribute Syntax}, for details of the exact syntax for using
1904 @cindex @code{noreturn} function attribute
1906 A few standard library functions, such as @code{abort} and @code{exit},
1907 cannot return. GCC knows this automatically. Some programs define
1908 their own functions that never return. You can declare them
1909 @code{noreturn} to tell the compiler this fact. For example,
1913 void fatal () __attribute__ ((noreturn));
1918 @dots{} /* @r{Print error message.} */ @dots{}
1924 The @code{noreturn} keyword tells the compiler to assume that
1925 @code{fatal} cannot return. It can then optimize without regard to what
1926 would happen if @code{fatal} ever did return. This makes slightly
1927 better code. More importantly, it helps avoid spurious warnings of
1928 uninitialized variables.
1930 Do not assume that registers saved by the calling function are
1931 restored before calling the @code{noreturn} function.
1933 It does not make sense for a @code{noreturn} function to have a return
1934 type other than @code{void}.
1936 The attribute @code{noreturn} is not implemented in GCC versions
1937 earlier than 2.5. An alternative way to declare that a function does
1938 not return, which works in the current version and in some older
1939 versions, is as follows:
1942 typedef void voidfn ();
1944 volatile voidfn fatal;
1947 @cindex @code{noinline} function attribute
1949 This function attribute prevents a function from being considered for
1952 @cindex @code{always_inline} function attribute
1954 Generally, functions are not inlined unless optimization is specified.
1955 For functions declared inline, this attribute inlines the function even
1956 if no optimization level was specified.
1958 @cindex @code{pure} function attribute
1960 Many functions have no effects except the return value and their
1961 return value depends only on the parameters and/or global variables.
1962 Such a function can be subject
1963 to common subexpression elimination and loop optimization just as an
1964 arithmetic operator would be. These functions should be declared
1965 with the attribute @code{pure}. For example,
1968 int square (int) __attribute__ ((pure));
1972 says that the hypothetical function @code{square} is safe to call
1973 fewer times than the program says.
1975 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
1976 Interesting non-pure functions are functions with infinite loops or those
1977 depending on volatile memory or other system resource, that may change between
1978 two consecutive calls (such as @code{feof} in a multithreading environment).
1980 The attribute @code{pure} is not implemented in GCC versions earlier
1982 @cindex @code{const} function attribute
1984 Many functions do not examine any values except their arguments, and
1985 have no effects except the return value. Basically this is just slightly
1986 more strict class than the @code{pure} attribute above, since function is not
1987 allowed to read global memory.
1989 @cindex pointer arguments
1990 Note that a function that has pointer arguments and examines the data
1991 pointed to must @emph{not} be declared @code{const}. Likewise, a
1992 function that calls a non-@code{const} function usually must not be
1993 @code{const}. It does not make sense for a @code{const} function to
1996 The attribute @code{const} is not implemented in GCC versions earlier
1997 than 2.5. An alternative way to declare that a function has no side
1998 effects, which works in the current version and in some older versions,
2002 typedef int intfn ();
2004 extern const intfn square;
2007 This approach does not work in GNU C++ from 2.6.0 on, since the language
2008 specifies that the @samp{const} must be attached to the return value.
2010 @cindex @code{nothrow} function attribute
2012 The @code{nothrow} attribute is used to inform the compiler that a
2013 function cannot throw an exception. For example, most functions in
2014 the standard C library can be guaranteed not to throw an exception
2015 with the notable exceptions of @code{qsort} and @code{bsearch} that
2016 take function pointer arguments. The @code{nothrow} attribute is not
2017 implemented in GCC versions earlier than 3.2.
2019 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2020 @cindex @code{format} function attribute
2022 The @code{format} attribute specifies that a function takes @code{printf},
2023 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2024 should be type-checked against a format string. For example, the
2029 my_printf (void *my_object, const char *my_format, ...)
2030 __attribute__ ((format (printf, 2, 3)));
2034 causes the compiler to check the arguments in calls to @code{my_printf}
2035 for consistency with the @code{printf} style format string argument
2038 The parameter @var{archetype} determines how the format string is
2039 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
2040 or @code{strfmon}. (You can also use @code{__printf__},
2041 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
2042 parameter @var{string-index} specifies which argument is the format
2043 string argument (starting from 1), while @var{first-to-check} is the
2044 number of the first argument to check against the format string. For
2045 functions where the arguments are not available to be checked (such as
2046 @code{vprintf}), specify the third parameter as zero. In this case the
2047 compiler only checks the format string for consistency. For
2048 @code{strftime} formats, the third parameter is required to be zero.
2050 In the example above, the format string (@code{my_format}) is the second
2051 argument of the function @code{my_print}, and the arguments to check
2052 start with the third argument, so the correct parameters for the format
2053 attribute are 2 and 3.
2055 @opindex ffreestanding
2056 The @code{format} attribute allows you to identify your own functions
2057 which take format strings as arguments, so that GCC can check the
2058 calls to these functions for errors. The compiler always (unless
2059 @option{-ffreestanding} is used) checks formats
2060 for the standard library functions @code{printf}, @code{fprintf},
2061 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2062 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2063 warnings are requested (using @option{-Wformat}), so there is no need to
2064 modify the header file @file{stdio.h}. In C99 mode, the functions
2065 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2066 @code{vsscanf} are also checked. Except in strictly conforming C
2067 standard modes, the X/Open function @code{strfmon} is also checked as
2068 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2069 @xref{C Dialect Options,,Options Controlling C Dialect}.
2071 @item format_arg (@var{string-index})
2072 @cindex @code{format_arg} function attribute
2073 @opindex Wformat-nonliteral
2074 The @code{format_arg} attribute specifies that a function takes a format
2075 string for a @code{printf}, @code{scanf}, @code{strftime} or
2076 @code{strfmon} style function and modifies it (for example, to translate
2077 it into another language), so the result can be passed to a
2078 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2079 function (with the remaining arguments to the format function the same
2080 as they would have been for the unmodified string). For example, the
2085 my_dgettext (char *my_domain, const char *my_format)
2086 __attribute__ ((format_arg (2)));
2090 causes the compiler to check the arguments in calls to a @code{printf},
2091 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2092 format string argument is a call to the @code{my_dgettext} function, for
2093 consistency with the format string argument @code{my_format}. If the
2094 @code{format_arg} attribute had not been specified, all the compiler
2095 could tell in such calls to format functions would be that the format
2096 string argument is not constant; this would generate a warning when
2097 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2098 without the attribute.
2100 The parameter @var{string-index} specifies which argument is the format
2101 string argument (starting from 1).
2103 The @code{format-arg} attribute allows you to identify your own
2104 functions which modify format strings, so that GCC can check the
2105 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2106 type function whose operands are a call to one of your own function.
2107 The compiler always treats @code{gettext}, @code{dgettext}, and
2108 @code{dcgettext} in this manner except when strict ISO C support is
2109 requested by @option{-ansi} or an appropriate @option{-std} option, or
2110 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
2111 Controlling C Dialect}.
2113 @item nonnull (@var{arg-index,...})
2114 @cindex @code{nonnull} function attribute
2115 The @code{nonnull} attribute specifies that some function parameters should
2116 be non-null pointers. For instance, the declaration:
2120 my_memcpy (void *dest, const void *src, size_t len)
2121 __attribute__((nonnull (1, 2)));
2125 causes the compiler to check that, in calls to @code{my_memcpy},
2126 arguments @var{dest} and @var{src} are non-null. If the compiler
2127 determines that a null pointer is passed in an argument slot marked
2128 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2129 is issued. The compiler may also choose to make optimizations based
2130 on the knowledge that certain function arguments will not be null.
2132 If no argument index list is given to the @code{nonnull} attribute,
2133 all pointer arguments are marked as non-null. To illustrate, the
2134 following declaration is equivalent to the previous example:
2138 my_memcpy (void *dest, const void *src, size_t len)
2139 __attribute__((nonnull));
2142 @item no_instrument_function
2143 @cindex @code{no_instrument_function} function attribute
2144 @opindex finstrument-functions
2145 If @option{-finstrument-functions} is given, profiling function calls will
2146 be generated at entry and exit of most user-compiled functions.
2147 Functions with this attribute will not be so instrumented.
2149 @item section ("@var{section-name}")
2150 @cindex @code{section} function attribute
2151 Normally, the compiler places the code it generates in the @code{text} section.
2152 Sometimes, however, you need additional sections, or you need certain
2153 particular functions to appear in special sections. The @code{section}
2154 attribute specifies that a function lives in a particular section.
2155 For example, the declaration:
2158 extern void foobar (void) __attribute__ ((section ("bar")));
2162 puts the function @code{foobar} in the @code{bar} section.
2164 Some file formats do not support arbitrary sections so the @code{section}
2165 attribute is not available on all platforms.
2166 If you need to map the entire contents of a module to a particular
2167 section, consider using the facilities of the linker instead.
2171 @cindex @code{constructor} function attribute
2172 @cindex @code{destructor} function attribute
2173 The @code{constructor} attribute causes the function to be called
2174 automatically before execution enters @code{main ()}. Similarly, the
2175 @code{destructor} attribute causes the function to be called
2176 automatically after @code{main ()} has completed or @code{exit ()} has
2177 been called. Functions with these attributes are useful for
2178 initializing data that will be used implicitly during the execution of
2181 These attributes are not currently implemented for Objective-C@.
2183 @cindex @code{unused} attribute.
2185 This attribute, attached to a function, means that the function is meant
2186 to be possibly unused. GCC will not produce a warning for this
2187 function. GNU C++ does not currently support this attribute as
2188 definitions without parameters are valid in C++.
2190 @cindex @code{used} attribute.
2192 This attribute, attached to a function, means that code must be emitted
2193 for the function even if it appears that the function is not referenced.
2194 This is useful, for example, when the function is referenced only in
2197 @cindex @code{deprecated} attribute.
2199 The @code{deprecated} attribute results in a warning if the function
2200 is used anywhere in the source file. This is useful when identifying
2201 functions that are expected to be removed in a future version of a
2202 program. The warning also includes the location of the declaration
2203 of the deprecated function, to enable users to easily find further
2204 information about why the function is deprecated, or what they should
2205 do instead. Note that the warnings only occurs for uses:
2208 int old_fn () __attribute__ ((deprecated));
2210 int (*fn_ptr)() = old_fn;
2213 results in a warning on line 3 but not line 2.
2215 The @code{deprecated} attribute can also be used for variables and
2216 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2219 @cindex @code{weak} attribute
2220 The @code{weak} attribute causes the declaration to be emitted as a weak
2221 symbol rather than a global. This is primarily useful in defining
2222 library functions which can be overridden in user code, though it can
2223 also be used with non-function declarations. Weak symbols are supported
2224 for ELF targets, and also for a.out targets when using the GNU assembler
2228 @cindex @code{malloc} attribute
2229 The @code{malloc} attribute is used to tell the compiler that a function
2230 may be treated as if it were the malloc function. The compiler assumes
2231 that calls to malloc result in a pointers that cannot alias anything.
2232 This will often improve optimization.
2234 @item alias ("@var{target}")
2235 @cindex @code{alias} attribute
2236 The @code{alias} attribute causes the declaration to be emitted as an
2237 alias for another symbol, which must be specified. For instance,
2240 void __f () @{ /* @r{Do something.} */; @}
2241 void f () __attribute__ ((weak, alias ("__f")));
2244 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
2245 mangled name for the target must be used.
2247 Not all target machines support this attribute.
2249 @item visibility ("@var{visibility_type}")
2250 @cindex @code{visibility} attribute
2251 The @code{visibility} attribute on ELF targets causes the declaration
2252 to be emitted with hidden, protected or internal visibility.
2255 void __attribute__ ((visibility ("protected")))
2256 f () @{ /* @r{Do something.} */; @}
2257 int i __attribute__ ((visibility ("hidden")));
2260 See the ELF gABI for complete details, but the short story is
2264 Hidden visibility indicates that the symbol will not be placed into
2265 the dynamic symbol table, so no other @dfn{module} (executable or
2266 shared library) can reference it directly.
2269 Protected visibility indicates that the symbol will be placed in the
2270 dynamic symbol table, but that references within the defining module
2271 will bind to the local symbol. That is, the symbol cannot be overridden
2275 Internal visibility is like hidden visibility, but with additional
2276 processor specific semantics. Unless otherwise specified by the psABI,
2277 gcc defines internal visibility to mean that the function is @emph{never}
2278 called from another module. Note that hidden symbols, while then cannot
2279 be referenced directly by other modules, can be referenced indirectly via
2280 function pointers. By indicating that a symbol cannot be called from
2281 outside the module, gcc may for instance omit the load of a PIC register
2282 since it is known that the calling function loaded the correct value.
2285 Not all ELF targets support this attribute.
2287 @item regparm (@var{number})
2288 @cindex functions that are passed arguments in registers on the 386
2289 On the Intel 386, the @code{regparm} attribute causes the compiler to
2290 pass up to @var{number} integer arguments in registers EAX,
2291 EDX, and ECX instead of on the stack. Functions that take a
2292 variable number of arguments will continue to be passed all of their
2293 arguments on the stack.
2296 @cindex functions that pop the argument stack on the 386
2297 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2298 assume that the called function will pop off the stack space used to
2299 pass arguments, unless it takes a variable number of arguments.
2301 The PowerPC compiler for Windows NT currently ignores the @code{stdcall}
2305 @cindex functions that do pop the argument stack on the 386
2307 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2308 assume that the calling function will pop off the stack space used to
2309 pass arguments. This is
2310 useful to override the effects of the @option{-mrtd} switch.
2312 The PowerPC compiler for Windows NT currently ignores the @code{cdecl}
2315 @item longcall/shortcall
2316 @cindex functions called via pointer on the RS/6000 and PowerPC
2317 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
2318 compiler to always call this function via a pointer, just as it would if
2319 the @option{-mlongcall} option had been specified. The @code{shortcall}
2320 attribute causes the compiler not to do this. These attributes override
2321 both the @option{-mlongcall} switch and the @code{#pragma longcall}
2324 @xref{RS/6000 and PowerPC Options}, for more information on when long
2325 calls are and are not necessary.
2327 @item long_call/short_call
2328 @cindex indirect calls on ARM
2329 This attribute allows to specify how to call a particular function on
2330 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2331 command line switch and @code{#pragma long_calls} settings. The
2332 @code{long_call} attribute causes the compiler to always call the
2333 function by first loading its address into a register and then using the
2334 contents of that register. The @code{short_call} attribute always places
2335 the offset to the function from the call site into the @samp{BL}
2336 instruction directly.
2339 @cindex functions which are imported from a dll on PowerPC Windows NT
2340 On the PowerPC running Windows NT, the @code{dllimport} attribute causes
2341 the compiler to call the function via a global pointer to the function
2342 pointer that is set up by the Windows NT dll library. The pointer name
2343 is formed by combining @code{__imp_} and the function name.
2346 @cindex functions which are exported from a dll on PowerPC Windows NT
2347 On the PowerPC running Windows NT, the @code{dllexport} attribute causes
2348 the compiler to provide a global pointer to the function pointer, so
2349 that it can be called with the @code{dllimport} attribute. The pointer
2350 name is formed by combining @code{__imp_} and the function name.
2352 @item exception (@var{except-func} [, @var{except-arg}])
2353 @cindex functions which specify exception handling on PowerPC Windows NT
2354 On the PowerPC running Windows NT, the @code{exception} attribute causes
2355 the compiler to modify the structured exception table entry it emits for
2356 the declared function. The string or identifier @var{except-func} is
2357 placed in the third entry of the structured exception table. It
2358 represents a function, which is called by the exception handling
2359 mechanism if an exception occurs. If it was specified, the string or
2360 identifier @var{except-arg} is placed in the fourth entry of the
2361 structured exception table.
2363 @item function_vector
2364 @cindex calling functions through the function vector on the H8/300 processors
2365 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2366 function should be called through the function vector. Calling a
2367 function through the function vector will reduce code size, however;
2368 the function vector has a limited size (maximum 128 entries on the H8/300
2369 and 64 entries on the H8/300H) and shares space with the interrupt vector.
2371 You must use GAS and GLD from GNU binutils version 2.7 or later for
2372 this attribute to work correctly.
2375 @cindex interrupt handler functions
2376 Use this attribute on the ARM, AVR, M32R/D and Xstormy16 ports to indicate
2377 that the specified function is an interrupt handler. The compiler will
2378 generate function entry and exit sequences suitable for use in an
2379 interrupt handler when this attribute is present.
2381 Note, interrupt handlers for the H8/300, H8/300H and SH processors can
2382 be specified via the @code{interrupt_handler} attribute.
2384 Note, on the AVR interrupts will be enabled inside the function.
2386 Note, for the ARM you can specify the kind of interrupt to be handled by
2387 adding an optional parameter to the interrupt attribute like this:
2390 void f () __attribute__ ((interrupt ("IRQ")));
2393 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2395 @item interrupt_handler
2396 @cindex interrupt handler functions on the H8/300 and SH processors
2397 Use this attribute on the H8/300, H8/300H and SH to indicate that the
2398 specified function is an interrupt handler. The compiler will generate
2399 function entry and exit sequences suitable for use in an interrupt
2400 handler when this attribute is present.
2403 Use this attribute on the SH to indicate an @code{interrupt_handler}
2404 function should switch to an alternate stack. It expects a string
2405 argument that names a global variable holding the address of the
2410 void f () __attribute__ ((interrupt_handler,
2411 sp_switch ("alt_stack")));
2415 Use this attribute on the SH for an @code{interrupt_handle} to return using
2416 @code{trapa} instead of @code{rte}. This attribute expects an integer
2417 argument specifying the trap number to be used.
2420 @cindex eight bit data on the H8/300 and H8/300H
2421 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2422 variable should be placed into the eight bit data section.
2423 The compiler will generate more efficient code for certain operations
2424 on data in the eight bit data area. Note the eight bit data area is limited to
2427 You must use GAS and GLD from GNU binutils version 2.7 or later for
2428 this attribute to work correctly.
2431 @cindex tiny data section on the H8/300H
2432 Use this attribute on the H8/300H to indicate that the specified
2433 variable should be placed into the tiny data section.
2434 The compiler will generate more efficient code for loads and stores
2435 on data in the tiny data section. Note the tiny data area is limited to
2436 slightly under 32kbytes of data.
2439 @cindex signal handler functions on the AVR processors
2440 Use this attribute on the AVR to indicate that the specified
2441 function is an signal handler. The compiler will generate function
2442 entry and exit sequences suitable for use in an signal handler when this
2443 attribute is present. Interrupts will be disabled inside function.
2446 @cindex function without a prologue/epilogue code
2447 Use this attribute on the ARM or AVR ports to indicate that the specified
2448 function do not need prologue/epilogue sequences generated by the
2449 compiler. It is up to the programmer to provide these sequences.
2451 @item model (@var{model-name})
2452 @cindex function addressability on the M32R/D
2453 Use this attribute on the M32R/D to set the addressability of an object,
2454 and the code generated for a function.
2455 The identifier @var{model-name} is one of @code{small}, @code{medium},
2456 or @code{large}, representing each of the code models.
2458 Small model objects live in the lower 16MB of memory (so that their
2459 addresses can be loaded with the @code{ld24} instruction), and are
2460 callable with the @code{bl} instruction.
2462 Medium model objects may live anywhere in the 32-bit address space (the
2463 compiler will generate @code{seth/add3} instructions to load their addresses),
2464 and are callable with the @code{bl} instruction.
2466 Large model objects may live anywhere in the 32-bit address space (the
2467 compiler will generate @code{seth/add3} instructions to load their addresses),
2468 and may not be reachable with the @code{bl} instruction (the compiler will
2469 generate the much slower @code{seth/add3/jl} instruction sequence).
2473 You can specify multiple attributes in a declaration by separating them
2474 by commas within the double parentheses or by immediately following an
2475 attribute declaration with another attribute declaration.
2477 @cindex @code{#pragma}, reason for not using
2478 @cindex pragma, reason for not using
2479 Some people object to the @code{__attribute__} feature, suggesting that
2480 ISO C's @code{#pragma} should be used instead. At the time
2481 @code{__attribute__} was designed, there were two reasons for not doing
2486 It is impossible to generate @code{#pragma} commands from a macro.
2489 There is no telling what the same @code{#pragma} might mean in another
2493 These two reasons applied to almost any application that might have been
2494 proposed for @code{#pragma}. It was basically a mistake to use
2495 @code{#pragma} for @emph{anything}.
2497 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2498 to be generated from macros. In addition, a @code{#pragma GCC}
2499 namespace is now in use for GCC-specific pragmas. However, it has been
2500 found convenient to use @code{__attribute__} to achieve a natural
2501 attachment of attributes to their corresponding declarations, whereas
2502 @code{#pragma GCC} is of use for constructs that do not naturally form
2503 part of the grammar. @xref{Other Directives,,Miscellaneous
2504 Preprocessing Directives, cpp, The C Preprocessor}.
2506 @node Attribute Syntax
2507 @section Attribute Syntax
2508 @cindex attribute syntax
2510 This section describes the syntax with which @code{__attribute__} may be
2511 used, and the constructs to which attribute specifiers bind, for the C
2512 language. Some details may vary for C++ and Objective-C@. Because of
2513 infelicities in the grammar for attributes, some forms described here
2514 may not be successfully parsed in all cases.
2516 There are some problems with the semantics of attributes in C++. For
2517 example, there are no manglings for attributes, although they may affect
2518 code generation, so problems may arise when attributed types are used in
2519 conjunction with templates or overloading. Similarly, @code{typeid}
2520 does not distinguish between types with different attributes. Support
2521 for attributes in C++ may be restricted in future to attributes on
2522 declarations only, but not on nested declarators.
2524 @xref{Function Attributes}, for details of the semantics of attributes
2525 applying to functions. @xref{Variable Attributes}, for details of the
2526 semantics of attributes applying to variables. @xref{Type Attributes},
2527 for details of the semantics of attributes applying to structure, union
2528 and enumerated types.
2530 An @dfn{attribute specifier} is of the form
2531 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2532 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2533 each attribute is one of the following:
2537 Empty. Empty attributes are ignored.
2540 A word (which may be an identifier such as @code{unused}, or a reserved
2541 word such as @code{const}).
2544 A word, followed by, in parentheses, parameters for the attribute.
2545 These parameters take one of the following forms:
2549 An identifier. For example, @code{mode} attributes use this form.
2552 An identifier followed by a comma and a non-empty comma-separated list
2553 of expressions. For example, @code{format} attributes use this form.
2556 A possibly empty comma-separated list of expressions. For example,
2557 @code{format_arg} attributes use this form with the list being a single
2558 integer constant expression, and @code{alias} attributes use this form
2559 with the list being a single string constant.
2563 An @dfn{attribute specifier list} is a sequence of one or more attribute
2564 specifiers, not separated by any other tokens.
2566 An attribute specifier list may appear after the colon following a
2567 label, other than a @code{case} or @code{default} label. The only
2568 attribute it makes sense to use after a label is @code{unused}. This
2569 feature is intended for code generated by programs which contains labels
2570 that may be unused but which is compiled with @option{-Wall}. It would
2571 not normally be appropriate to use in it human-written code, though it
2572 could be useful in cases where the code that jumps to the label is
2573 contained within an @code{#ifdef} conditional.
2575 An attribute specifier list may appear as part of a @code{struct},
2576 @code{union} or @code{enum} specifier. It may go either immediately
2577 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2578 the closing brace. It is ignored if the content of the structure, union
2579 or enumerated type is not defined in the specifier in which the
2580 attribute specifier list is used---that is, in usages such as
2581 @code{struct __attribute__((foo)) bar} with no following opening brace.
2582 Where attribute specifiers follow the closing brace, they are considered
2583 to relate to the structure, union or enumerated type defined, not to any
2584 enclosing declaration the type specifier appears in, and the type
2585 defined is not complete until after the attribute specifiers.
2586 @c Otherwise, there would be the following problems: a shift/reduce
2587 @c conflict between attributes binding the struct/union/enum and
2588 @c binding to the list of specifiers/qualifiers; and "aligned"
2589 @c attributes could use sizeof for the structure, but the size could be
2590 @c changed later by "packed" attributes.
2592 Otherwise, an attribute specifier appears as part of a declaration,
2593 counting declarations of unnamed parameters and type names, and relates
2594 to that declaration (which may be nested in another declaration, for
2595 example in the case of a parameter declaration), or to a particular declarator
2596 within a declaration. Where an
2597 attribute specifier is applied to a parameter declared as a function or
2598 an array, it should apply to the function or array rather than the
2599 pointer to which the parameter is implicitly converted, but this is not
2600 yet correctly implemented.
2602 Any list of specifiers and qualifiers at the start of a declaration may
2603 contain attribute specifiers, whether or not such a list may in that
2604 context contain storage class specifiers. (Some attributes, however,
2605 are essentially in the nature of storage class specifiers, and only make
2606 sense where storage class specifiers may be used; for example,
2607 @code{section}.) There is one necessary limitation to this syntax: the
2608 first old-style parameter declaration in a function definition cannot
2609 begin with an attribute specifier, because such an attribute applies to
2610 the function instead by syntax described below (which, however, is not
2611 yet implemented in this case). In some other cases, attribute
2612 specifiers are permitted by this grammar but not yet supported by the
2613 compiler. All attribute specifiers in this place relate to the
2614 declaration as a whole. In the obsolescent usage where a type of
2615 @code{int} is implied by the absence of type specifiers, such a list of
2616 specifiers and qualifiers may be an attribute specifier list with no
2617 other specifiers or qualifiers.
2619 An attribute specifier list may appear immediately before a declarator
2620 (other than the first) in a comma-separated list of declarators in a
2621 declaration of more than one identifier using a single list of
2622 specifiers and qualifiers. Such attribute specifiers apply
2623 only to the identifier before whose declarator they appear. For
2627 __attribute__((noreturn)) void d0 (void),
2628 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2633 the @code{noreturn} attribute applies to all the functions
2634 declared; the @code{format} attribute only applies to @code{d1}.
2636 An attribute specifier list may appear immediately before the comma,
2637 @code{=} or semicolon terminating the declaration of an identifier other
2638 than a function definition. At present, such attribute specifiers apply
2639 to the declared object or function, but in future they may attach to the
2640 outermost adjacent declarator. In simple cases there is no difference,
2641 but, for example, in
2644 void (****f)(void) __attribute__((noreturn));
2648 at present the @code{noreturn} attribute applies to @code{f}, which
2649 causes a warning since @code{f} is not a function, but in future it may
2650 apply to the function @code{****f}. The precise semantics of what
2651 attributes in such cases will apply to are not yet specified. Where an
2652 assembler name for an object or function is specified (@pxref{Asm
2653 Labels}), at present the attribute must follow the @code{asm}
2654 specification; in future, attributes before the @code{asm} specification
2655 may apply to the adjacent declarator, and those after it to the declared
2658 An attribute specifier list may, in future, be permitted to appear after
2659 the declarator in a function definition (before any old-style parameter
2660 declarations or the function body).
2662 Attribute specifiers may be mixed with type qualifiers appearing inside
2663 the @code{[]} of a parameter array declarator, in the C99 construct by
2664 which such qualifiers are applied to the pointer to which the array is
2665 implicitly converted. Such attribute specifiers apply to the pointer,
2666 not to the array, but at present this is not implemented and they are
2669 An attribute specifier list may appear at the start of a nested
2670 declarator. At present, there are some limitations in this usage: the
2671 attributes correctly apply to the declarator, but for most individual
2672 attributes the semantics this implies are not implemented.
2673 When attribute specifiers follow the @code{*} of a pointer
2674 declarator, they may be mixed with any type qualifiers present.
2675 The following describes the formal semantics of this syntax. It will make the
2676 most sense if you are familiar with the formal specification of
2677 declarators in the ISO C standard.
2679 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2680 D1}, where @code{T} contains declaration specifiers that specify a type
2681 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2682 contains an identifier @var{ident}. The type specified for @var{ident}
2683 for derived declarators whose type does not include an attribute
2684 specifier is as in the ISO C standard.
2686 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2687 and the declaration @code{T D} specifies the type
2688 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2689 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2690 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2692 If @code{D1} has the form @code{*
2693 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2694 declaration @code{T D} specifies the type
2695 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2696 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2697 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2703 void (__attribute__((noreturn)) ****f) (void);
2707 specifies the type ``pointer to pointer to pointer to pointer to
2708 non-returning function returning @code{void}''. As another example,
2711 char *__attribute__((aligned(8))) *f;
2715 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2716 Note again that this does not work with most attributes; for example,
2717 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2718 is not yet supported.
2720 For compatibility with existing code written for compiler versions that
2721 did not implement attributes on nested declarators, some laxity is
2722 allowed in the placing of attributes. If an attribute that only applies
2723 to types is applied to a declaration, it will be treated as applying to
2724 the type of that declaration. If an attribute that only applies to
2725 declarations is applied to the type of a declaration, it will be treated
2726 as applying to that declaration; and, for compatibility with code
2727 placing the attributes immediately before the identifier declared, such
2728 an attribute applied to a function return type will be treated as
2729 applying to the function type, and such an attribute applied to an array
2730 element type will be treated as applying to the array type. If an
2731 attribute that only applies to function types is applied to a
2732 pointer-to-function type, it will be treated as applying to the pointer
2733 target type; if such an attribute is applied to a function return type
2734 that is not a pointer-to-function type, it will be treated as applying
2735 to the function type.
2737 @node Function Prototypes
2738 @section Prototypes and Old-Style Function Definitions
2739 @cindex function prototype declarations
2740 @cindex old-style function definitions
2741 @cindex promotion of formal parameters
2743 GNU C extends ISO C to allow a function prototype to override a later
2744 old-style non-prototype definition. Consider the following example:
2747 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2754 /* @r{Prototype function declaration.} */
2755 int isroot P((uid_t));
2757 /* @r{Old-style function definition.} */
2759 isroot (x) /* ??? lossage here ??? */
2766 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2767 not allow this example, because subword arguments in old-style
2768 non-prototype definitions are promoted. Therefore in this example the
2769 function definition's argument is really an @code{int}, which does not
2770 match the prototype argument type of @code{short}.
2772 This restriction of ISO C makes it hard to write code that is portable
2773 to traditional C compilers, because the programmer does not know
2774 whether the @code{uid_t} type is @code{short}, @code{int}, or
2775 @code{long}. Therefore, in cases like these GNU C allows a prototype
2776 to override a later old-style definition. More precisely, in GNU C, a
2777 function prototype argument type overrides the argument type specified
2778 by a later old-style definition if the former type is the same as the
2779 latter type before promotion. Thus in GNU C the above example is
2780 equivalent to the following:
2793 GNU C++ does not support old-style function definitions, so this
2794 extension is irrelevant.
2797 @section C++ Style Comments
2799 @cindex C++ comments
2800 @cindex comments, C++ style
2802 In GNU C, you may use C++ style comments, which start with @samp{//} and
2803 continue until the end of the line. Many other C implementations allow
2804 such comments, and they are included in the 1999 C standard. However,
2805 C++ style comments are not recognized if you specify an @option{-std}
2806 option specifying a version of ISO C before C99, or @option{-ansi}
2807 (equivalent to @option{-std=c89}).
2810 @section Dollar Signs in Identifier Names
2812 @cindex dollar signs in identifier names
2813 @cindex identifier names, dollar signs in
2815 In GNU C, you may normally use dollar signs in identifier names.
2816 This is because many traditional C implementations allow such identifiers.
2817 However, dollar signs in identifiers are not supported on a few target
2818 machines, typically because the target assembler does not allow them.
2820 @node Character Escapes
2821 @section The Character @key{ESC} in Constants
2823 You can use the sequence @samp{\e} in a string or character constant to
2824 stand for the ASCII character @key{ESC}.
2827 @section Inquiring on Alignment of Types or Variables
2829 @cindex type alignment
2830 @cindex variable alignment
2832 The keyword @code{__alignof__} allows you to inquire about how an object
2833 is aligned, or the minimum alignment usually required by a type. Its
2834 syntax is just like @code{sizeof}.
2836 For example, if the target machine requires a @code{double} value to be
2837 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2838 This is true on many RISC machines. On more traditional machine
2839 designs, @code{__alignof__ (double)} is 4 or even 2.
2841 Some machines never actually require alignment; they allow reference to any
2842 data type even at an odd addresses. For these machines, @code{__alignof__}
2843 reports the @emph{recommended} alignment of a type.
2845 If the operand of @code{__alignof__} is an lvalue rather than a type,
2846 its value is the required alignment for its type, taking into account
2847 any minimum alignment specified with GCC's @code{__attribute__}
2848 extension (@pxref{Variable Attributes}). For example, after this
2852 struct foo @{ int x; char y; @} foo1;
2856 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
2857 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
2859 It is an error to ask for the alignment of an incomplete type.
2861 @node Variable Attributes
2862 @section Specifying Attributes of Variables
2863 @cindex attribute of variables
2864 @cindex variable attributes
2866 The keyword @code{__attribute__} allows you to specify special
2867 attributes of variables or structure fields. This keyword is followed
2868 by an attribute specification inside double parentheses. Ten
2869 attributes are currently defined for variables: @code{aligned},
2870 @code{mode}, @code{nocommon}, @code{packed}, @code{section},
2871 @code{transparent_union}, @code{unused}, @code{deprecated},
2872 @code{vector_size}, and @code{weak}. Some other attributes are defined
2873 for variables on particular target systems. Other attributes are
2874 available for functions (@pxref{Function Attributes}) and for types
2875 (@pxref{Type Attributes}). Other front ends might define more
2876 attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2878 You may also specify attributes with @samp{__} preceding and following
2879 each keyword. This allows you to use them in header files without
2880 being concerned about a possible macro of the same name. For example,
2881 you may use @code{__aligned__} instead of @code{aligned}.
2883 @xref{Attribute Syntax}, for details of the exact syntax for using
2887 @cindex @code{aligned} attribute
2888 @item aligned (@var{alignment})
2889 This attribute specifies a minimum alignment for the variable or
2890 structure field, measured in bytes. For example, the declaration:
2893 int x __attribute__ ((aligned (16))) = 0;
2897 causes the compiler to allocate the global variable @code{x} on a
2898 16-byte boundary. On a 68040, this could be used in conjunction with
2899 an @code{asm} expression to access the @code{move16} instruction which
2900 requires 16-byte aligned operands.
2902 You can also specify the alignment of structure fields. For example, to
2903 create a double-word aligned @code{int} pair, you could write:
2906 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2910 This is an alternative to creating a union with a @code{double} member
2911 that forces the union to be double-word aligned.
2913 As in the preceding examples, you can explicitly specify the alignment
2914 (in bytes) that you wish the compiler to use for a given variable or
2915 structure field. Alternatively, you can leave out the alignment factor
2916 and just ask the compiler to align a variable or field to the maximum
2917 useful alignment for the target machine you are compiling for. For
2918 example, you could write:
2921 short array[3] __attribute__ ((aligned));
2924 Whenever you leave out the alignment factor in an @code{aligned} attribute
2925 specification, the compiler automatically sets the alignment for the declared
2926 variable or field to the largest alignment which is ever used for any data
2927 type on the target machine you are compiling for. Doing this can often make
2928 copy operations more efficient, because the compiler can use whatever
2929 instructions copy the biggest chunks of memory when performing copies to
2930 or from the variables or fields that you have aligned this way.
2932 The @code{aligned} attribute can only increase the alignment; but you
2933 can decrease it by specifying @code{packed} as well. See below.
2935 Note that the effectiveness of @code{aligned} attributes may be limited
2936 by inherent limitations in your linker. On many systems, the linker is
2937 only able to arrange for variables to be aligned up to a certain maximum
2938 alignment. (For some linkers, the maximum supported alignment may
2939 be very very small.) If your linker is only able to align variables
2940 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2941 in an @code{__attribute__} will still only provide you with 8 byte
2942 alignment. See your linker documentation for further information.
2944 @item mode (@var{mode})
2945 @cindex @code{mode} attribute
2946 This attribute specifies the data type for the declaration---whichever
2947 type corresponds to the mode @var{mode}. This in effect lets you
2948 request an integer or floating point type according to its width.
2950 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2951 indicate the mode corresponding to a one-byte integer, @samp{word} or
2952 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2953 or @samp{__pointer__} for the mode used to represent pointers.
2956 @cindex @code{nocommon} attribute
2958 This attribute specifies requests GCC not to place a variable
2959 ``common'' but instead to allocate space for it directly. If you
2960 specify the @option{-fno-common} flag, GCC will do this for all
2963 Specifying the @code{nocommon} attribute for a variable provides an
2964 initialization of zeros. A variable may only be initialized in one
2968 @cindex @code{packed} attribute
2969 The @code{packed} attribute specifies that a variable or structure field
2970 should have the smallest possible alignment---one byte for a variable,
2971 and one bit for a field, unless you specify a larger value with the
2972 @code{aligned} attribute.
2974 Here is a structure in which the field @code{x} is packed, so that it
2975 immediately follows @code{a}:
2981 int x[2] __attribute__ ((packed));
2985 @item section ("@var{section-name}")
2986 @cindex @code{section} variable attribute
2987 Normally, the compiler places the objects it generates in sections like
2988 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
2989 or you need certain particular variables to appear in special sections,
2990 for example to map to special hardware. The @code{section}
2991 attribute specifies that a variable (or function) lives in a particular
2992 section. For example, this small program uses several specific section names:
2995 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
2996 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
2997 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
2998 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3002 /* Initialize stack pointer */
3003 init_sp (stack + sizeof (stack));
3005 /* Initialize initialized data */
3006 memcpy (&init_data, &data, &edata - &data);
3008 /* Turn on the serial ports */
3015 Use the @code{section} attribute with an @emph{initialized} definition
3016 of a @emph{global} variable, as shown in the example. GCC issues
3017 a warning and otherwise ignores the @code{section} attribute in
3018 uninitialized variable declarations.
3020 You may only use the @code{section} attribute with a fully initialized
3021 global definition because of the way linkers work. The linker requires
3022 each object be defined once, with the exception that uninitialized
3023 variables tentatively go in the @code{common} (or @code{bss}) section
3024 and can be multiply ``defined''. You can force a variable to be
3025 initialized with the @option{-fno-common} flag or the @code{nocommon}
3028 Some file formats do not support arbitrary sections so the @code{section}
3029 attribute is not available on all platforms.
3030 If you need to map the entire contents of a module to a particular
3031 section, consider using the facilities of the linker instead.
3034 @cindex @code{shared} variable attribute
3035 On Windows NT, in addition to putting variable definitions in a named
3036 section, the section can also be shared among all running copies of an
3037 executable or DLL@. For example, this small program defines shared data
3038 by putting it in a named section @code{shared} and marking the section
3042 int foo __attribute__((section ("shared"), shared)) = 0;
3047 /* Read and write foo. All running
3048 copies see the same value. */
3054 You may only use the @code{shared} attribute along with @code{section}
3055 attribute with a fully initialized global definition because of the way
3056 linkers work. See @code{section} attribute for more information.
3058 The @code{shared} attribute is only available on Windows NT@.
3060 @item transparent_union
3061 This attribute, attached to a function parameter which is a union, means
3062 that the corresponding argument may have the type of any union member,
3063 but the argument is passed as if its type were that of the first union
3064 member. For more details see @xref{Type Attributes}. You can also use
3065 this attribute on a @code{typedef} for a union data type; then it
3066 applies to all function parameters with that type.
3069 This attribute, attached to a variable, means that the variable is meant
3070 to be possibly unused. GCC will not produce a warning for this
3074 The @code{deprecated} attribute results in a warning if the variable
3075 is used anywhere in the source file. This is useful when identifying
3076 variables that are expected to be removed in a future version of a
3077 program. The warning also includes the location of the declaration
3078 of the deprecated variable, to enable users to easily find further
3079 information about why the variable is deprecated, or what they should
3080 do instead. Note that the warnings only occurs for uses:
3083 extern int old_var __attribute__ ((deprecated));
3085 int new_fn () @{ return old_var; @}
3088 results in a warning on line 3 but not line 2.
3090 The @code{deprecated} attribute can also be used for functions and
3091 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
3093 @item vector_size (@var{bytes})
3094 This attribute specifies the vector size for the variable, measured in
3095 bytes. For example, the declaration:
3098 int foo __attribute__ ((vector_size (16)));
3102 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3103 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3104 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3106 This attribute is only applicable to integral and float scalars,
3107 although arrays, pointers, and function return values are allowed in
3108 conjunction with this construct.
3110 Aggregates with this attribute are invalid, even if they are of the same
3111 size as a corresponding scalar. For example, the declaration:
3114 struct S @{ int a; @};
3115 struct S __attribute__ ((vector_size (16))) foo;
3119 is invalid even if the size of the structure is the same as the size of
3123 The @code{weak} attribute is described in @xref{Function Attributes}.
3125 @item model (@var{model-name})
3126 @cindex variable addressability on the M32R/D
3127 Use this attribute on the M32R/D to set the addressability of an object.
3128 The identifier @var{model-name} is one of @code{small}, @code{medium},
3129 or @code{large}, representing each of the code models.
3131 Small model objects live in the lower 16MB of memory (so that their
3132 addresses can be loaded with the @code{ld24} instruction).
3134 Medium and large model objects may live anywhere in the 32-bit address space
3135 (the compiler will generate @code{seth/add3} instructions to load their
3140 To specify multiple attributes, separate them by commas within the
3141 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3144 @node Type Attributes
3145 @section Specifying Attributes of Types
3146 @cindex attribute of types
3147 @cindex type attributes
3149 The keyword @code{__attribute__} allows you to specify special
3150 attributes of @code{struct} and @code{union} types when you define such
3151 types. This keyword is followed by an attribute specification inside
3152 double parentheses. Six attributes are currently defined for types:
3153 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
3154 @code{deprecated} and @code{may_alias}. Other attributes are defined for
3155 functions (@pxref{Function Attributes}) and for variables
3156 (@pxref{Variable Attributes}).
3158 You may also specify any one of these attributes with @samp{__}
3159 preceding and following its keyword. This allows you to use these
3160 attributes in header files without being concerned about a possible
3161 macro of the same name. For example, you may use @code{__aligned__}
3162 instead of @code{aligned}.
3164 You may specify the @code{aligned} and @code{transparent_union}
3165 attributes either in a @code{typedef} declaration or just past the
3166 closing curly brace of a complete enum, struct or union type
3167 @emph{definition} and the @code{packed} attribute only past the closing
3168 brace of a definition.
3170 You may also specify attributes between the enum, struct or union
3171 tag and the name of the type rather than after the closing brace.
3173 @xref{Attribute Syntax}, for details of the exact syntax for using
3177 @cindex @code{aligned} attribute
3178 @item aligned (@var{alignment})
3179 This attribute specifies a minimum alignment (in bytes) for variables
3180 of the specified type. For example, the declarations:
3183 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3184 typedef int more_aligned_int __attribute__ ((aligned (8)));
3188 force the compiler to insure (as far as it can) that each variable whose
3189 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3190 aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all
3191 variables of type @code{struct S} aligned to 8-byte boundaries allows
3192 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3193 store) instructions when copying one variable of type @code{struct S} to
3194 another, thus improving run-time efficiency.
3196 Note that the alignment of any given @code{struct} or @code{union} type
3197 is required by the ISO C standard to be at least a perfect multiple of
3198 the lowest common multiple of the alignments of all of the members of
3199 the @code{struct} or @code{union} in question. This means that you @emph{can}
3200 effectively adjust the alignment of a @code{struct} or @code{union}
3201 type by attaching an @code{aligned} attribute to any one of the members
3202 of such a type, but the notation illustrated in the example above is a
3203 more obvious, intuitive, and readable way to request the compiler to
3204 adjust the alignment of an entire @code{struct} or @code{union} type.
3206 As in the preceding example, you can explicitly specify the alignment
3207 (in bytes) that you wish the compiler to use for a given @code{struct}
3208 or @code{union} type. Alternatively, you can leave out the alignment factor
3209 and just ask the compiler to align a type to the maximum
3210 useful alignment for the target machine you are compiling for. For
3211 example, you could write:
3214 struct S @{ short f[3]; @} __attribute__ ((aligned));
3217 Whenever you leave out the alignment factor in an @code{aligned}
3218 attribute specification, the compiler automatically sets the alignment
3219 for the type to the largest alignment which is ever used for any data
3220 type on the target machine you are compiling for. Doing this can often
3221 make copy operations more efficient, because the compiler can use
3222 whatever instructions copy the biggest chunks of memory when performing
3223 copies to or from the variables which have types that you have aligned
3226 In the example above, if the size of each @code{short} is 2 bytes, then
3227 the size of the entire @code{struct S} type is 6 bytes. The smallest
3228 power of two which is greater than or equal to that is 8, so the
3229 compiler sets the alignment for the entire @code{struct S} type to 8
3232 Note that although you can ask the compiler to select a time-efficient
3233 alignment for a given type and then declare only individual stand-alone
3234 objects of that type, the compiler's ability to select a time-efficient
3235 alignment is primarily useful only when you plan to create arrays of
3236 variables having the relevant (efficiently aligned) type. If you
3237 declare or use arrays of variables of an efficiently-aligned type, then
3238 it is likely that your program will also be doing pointer arithmetic (or
3239 subscripting, which amounts to the same thing) on pointers to the
3240 relevant type, and the code that the compiler generates for these
3241 pointer arithmetic operations will often be more efficient for
3242 efficiently-aligned types than for other types.
3244 The @code{aligned} attribute can only increase the alignment; but you
3245 can decrease it by specifying @code{packed} as well. See below.
3247 Note that the effectiveness of @code{aligned} attributes may be limited
3248 by inherent limitations in your linker. On many systems, the linker is
3249 only able to arrange for variables to be aligned up to a certain maximum
3250 alignment. (For some linkers, the maximum supported alignment may
3251 be very very small.) If your linker is only able to align variables
3252 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3253 in an @code{__attribute__} will still only provide you with 8 byte
3254 alignment. See your linker documentation for further information.
3257 This attribute, attached to an @code{enum}, @code{struct}, or
3258 @code{union} type definition, specified that the minimum required memory
3259 be used to represent the type.
3261 @opindex fshort-enums
3262 Specifying this attribute for @code{struct} and @code{union} types is
3263 equivalent to specifying the @code{packed} attribute on each of the
3264 structure or union members. Specifying the @option{-fshort-enums}
3265 flag on the line is equivalent to specifying the @code{packed}
3266 attribute on all @code{enum} definitions.
3268 You may only specify this attribute after a closing curly brace on an
3269 @code{enum} definition, not in a @code{typedef} declaration, unless that
3270 declaration also contains the definition of the @code{enum}.
3272 @item transparent_union
3273 This attribute, attached to a @code{union} type definition, indicates
3274 that any function parameter having that union type causes calls to that
3275 function to be treated in a special way.
3277 First, the argument corresponding to a transparent union type can be of
3278 any type in the union; no cast is required. Also, if the union contains
3279 a pointer type, the corresponding argument can be a null pointer
3280 constant or a void pointer expression; and if the union contains a void
3281 pointer type, the corresponding argument can be any pointer expression.
3282 If the union member type is a pointer, qualifiers like @code{const} on
3283 the referenced type must be respected, just as with normal pointer
3286 Second, the argument is passed to the function using the calling
3287 conventions of first member of the transparent union, not the calling
3288 conventions of the union itself. All members of the union must have the
3289 same machine representation; this is necessary for this argument passing
3292 Transparent unions are designed for library functions that have multiple
3293 interfaces for compatibility reasons. For example, suppose the
3294 @code{wait} function must accept either a value of type @code{int *} to
3295 comply with Posix, or a value of type @code{union wait *} to comply with
3296 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3297 @code{wait} would accept both kinds of arguments, but it would also
3298 accept any other pointer type and this would make argument type checking
3299 less useful. Instead, @code{<sys/wait.h>} might define the interface
3307 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3309 pid_t wait (wait_status_ptr_t);
3312 This interface allows either @code{int *} or @code{union wait *}
3313 arguments to be passed, using the @code{int *} calling convention.
3314 The program can call @code{wait} with arguments of either type:
3317 int w1 () @{ int w; return wait (&w); @}
3318 int w2 () @{ union wait w; return wait (&w); @}
3321 With this interface, @code{wait}'s implementation might look like this:
3324 pid_t wait (wait_status_ptr_t p)
3326 return waitpid (-1, p.__ip, 0);
3331 When attached to a type (including a @code{union} or a @code{struct}),
3332 this attribute means that variables of that type are meant to appear
3333 possibly unused. GCC will not produce a warning for any variables of
3334 that type, even if the variable appears to do nothing. This is often
3335 the case with lock or thread classes, which are usually defined and then
3336 not referenced, but contain constructors and destructors that have
3337 nontrivial bookkeeping functions.
3340 The @code{deprecated} attribute results in a warning if the type
3341 is used anywhere in the source file. This is useful when identifying
3342 types that are expected to be removed in a future version of a program.
3343 If possible, the warning also includes the location of the declaration
3344 of the deprecated type, to enable users to easily find further
3345 information about why the type is deprecated, or what they should do
3346 instead. Note that the warnings only occur for uses and then only
3347 if the type is being applied to an identifier that itself is not being
3348 declared as deprecated.
3351 typedef int T1 __attribute__ ((deprecated));
3355 typedef T1 T3 __attribute__ ((deprecated));
3356 T3 z __attribute__ ((deprecated));
3359 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3360 warning is issued for line 4 because T2 is not explicitly
3361 deprecated. Line 5 has no warning because T3 is explicitly
3362 deprecated. Similarly for line 6.
3364 The @code{deprecated} attribute can also be used for functions and
3365 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3368 Accesses to objects with types with this attribute are not subjected to
3369 type-based alias analysis, but are instead assumed to be able to alias
3370 any other type of objects, just like the @code{char} type. See
3371 @option{-fstrict-aliasing} for more information on aliasing issues.
3376 typedef short __attribute__((__may_alias__)) short_a;
3382 short_a *b = (short_a *) &a;
3386 if (a == 0x12345678)
3393 If you replaced @code{short_a} with @code{short} in the variable
3394 declaration, the above program would abort when compiled with
3395 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3396 above in recent GCC versions.
3399 To specify multiple attributes, separate them by commas within the
3400 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3404 @section An Inline Function is As Fast As a Macro
3405 @cindex inline functions
3406 @cindex integrating function code
3408 @cindex macros, inline alternative
3410 By declaring a function @code{inline}, you can direct GCC to
3411 integrate that function's code into the code for its callers. This
3412 makes execution faster by eliminating the function-call overhead; in
3413 addition, if any of the actual argument values are constant, their known
3414 values may permit simplifications at compile time so that not all of the
3415 inline function's code needs to be included. The effect on code size is
3416 less predictable; object code may be larger or smaller with function
3417 inlining, depending on the particular case. Inlining of functions is an
3418 optimization and it really ``works'' only in optimizing compilation. If
3419 you don't use @option{-O}, no function is really inline.
3421 Inline functions are included in the ISO C99 standard, but there are
3422 currently substantial differences between what GCC implements and what
3423 the ISO C99 standard requires.
3425 To declare a function inline, use the @code{inline} keyword in its
3426 declaration, like this:
3436 (If you are writing a header file to be included in ISO C programs, write
3437 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3438 You can also make all ``simple enough'' functions inline with the option
3439 @option{-finline-functions}.
3442 Note that certain usages in a function definition can make it unsuitable
3443 for inline substitution. Among these usages are: use of varargs, use of
3444 alloca, use of variable sized data types (@pxref{Variable Length}),
3445 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3446 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3447 will warn when a function marked @code{inline} could not be substituted,
3448 and will give the reason for the failure.
3450 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3451 does not affect the linkage of the function.
3453 @cindex automatic @code{inline} for C++ member fns
3454 @cindex @code{inline} automatic for C++ member fns
3455 @cindex member fns, automatically @code{inline}
3456 @cindex C++ member fns, automatically @code{inline}
3457 @opindex fno-default-inline
3458 GCC automatically inlines member functions defined within the class
3459 body of C++ programs even if they are not explicitly declared
3460 @code{inline}. (You can override this with @option{-fno-default-inline};
3461 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3463 @cindex inline functions, omission of
3464 @opindex fkeep-inline-functions
3465 When a function is both inline and @code{static}, if all calls to the
3466 function are integrated into the caller, and the function's address is
3467 never used, then the function's own assembler code is never referenced.
3468 In this case, GCC does not actually output assembler code for the
3469 function, unless you specify the option @option{-fkeep-inline-functions}.
3470 Some calls cannot be integrated for various reasons (in particular,
3471 calls that precede the function's definition cannot be integrated, and
3472 neither can recursive calls within the definition). If there is a
3473 nonintegrated call, then the function is compiled to assembler code as
3474 usual. The function must also be compiled as usual if the program
3475 refers to its address, because that can't be inlined.
3477 @cindex non-static inline function
3478 When an inline function is not @code{static}, then the compiler must assume
3479 that there may be calls from other source files; since a global symbol can
3480 be defined only once in any program, the function must not be defined in
3481 the other source files, so the calls therein cannot be integrated.
3482 Therefore, a non-@code{static} inline function is always compiled on its
3483 own in the usual fashion.
3485 If you specify both @code{inline} and @code{extern} in the function
3486 definition, then the definition is used only for inlining. In no case
3487 is the function compiled on its own, not even if you refer to its
3488 address explicitly. Such an address becomes an external reference, as
3489 if you had only declared the function, and had not defined it.
3491 This combination of @code{inline} and @code{extern} has almost the
3492 effect of a macro. The way to use it is to put a function definition in
3493 a header file with these keywords, and put another copy of the
3494 definition (lacking @code{inline} and @code{extern}) in a library file.
3495 The definition in the header file will cause most calls to the function
3496 to be inlined. If any uses of the function remain, they will refer to
3497 the single copy in the library.
3499 For future compatibility with when GCC implements ISO C99 semantics for
3500 inline functions, it is best to use @code{static inline} only. (The
3501 existing semantics will remain available when @option{-std=gnu89} is
3502 specified, but eventually the default will be @option{-std=gnu99} and
3503 that will implement the C99 semantics, though it does not do so yet.)
3505 GCC does not inline any functions when not optimizing unless you specify
3506 the @samp{always_inline} attribute for the function, like this:
3510 inline void foo (const char) __attribute__((always_inline));
3514 @section Assembler Instructions with C Expression Operands
3515 @cindex extended @code{asm}
3516 @cindex @code{asm} expressions
3517 @cindex assembler instructions
3520 In an assembler instruction using @code{asm}, you can specify the
3521 operands of the instruction using C expressions. This means you need not
3522 guess which registers or memory locations will contain the data you want
3525 You must specify an assembler instruction template much like what
3526 appears in a machine description, plus an operand constraint string for
3529 For example, here is how to use the 68881's @code{fsinx} instruction:
3532 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3536 Here @code{angle} is the C expression for the input operand while
3537 @code{result} is that of the output operand. Each has @samp{"f"} as its
3538 operand constraint, saying that a floating point register is required.
3539 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3540 output operands' constraints must use @samp{=}. The constraints use the
3541 same language used in the machine description (@pxref{Constraints}).
3543 Each operand is described by an operand-constraint string followed by
3544 the C expression in parentheses. A colon separates the assembler
3545 template from the first output operand and another separates the last
3546 output operand from the first input, if any. Commas separate the
3547 operands within each group. The total number of operands is currently
3548 limited to 30; this limitation may be lifted in some future version of
3551 If there are no output operands but there are input operands, you must
3552 place two consecutive colons surrounding the place where the output
3555 As of GCC version 3.1, it is also possible to specify input and output
3556 operands using symbolic names which can be referenced within the
3557 assembler code. These names are specified inside square brackets
3558 preceding the constraint string, and can be referenced inside the
3559 assembler code using @code{%[@var{name}]} instead of a percentage sign
3560 followed by the operand number. Using named operands the above example
3564 asm ("fsinx %[angle],%[output]"
3565 : [output] "=f" (result)
3566 : [angle] "f" (angle));
3570 Note that the symbolic operand names have no relation whatsoever to
3571 other C identifiers. You may use any name you like, even those of
3572 existing C symbols, but must ensure that no two operands within the same
3573 assembler construct use the same symbolic name.
3575 Output operand expressions must be lvalues; the compiler can check this.
3576 The input operands need not be lvalues. The compiler cannot check
3577 whether the operands have data types that are reasonable for the
3578 instruction being executed. It does not parse the assembler instruction
3579 template and does not know what it means or even whether it is valid
3580 assembler input. The extended @code{asm} feature is most often used for
3581 machine instructions the compiler itself does not know exist. If
3582 the output expression cannot be directly addressed (for example, it is a
3583 bit-field), your constraint must allow a register. In that case, GCC
3584 will use the register as the output of the @code{asm}, and then store
3585 that register into the output.
3587 The ordinary output operands must be write-only; GCC will assume that
3588 the values in these operands before the instruction are dead and need
3589 not be generated. Extended asm supports input-output or read-write
3590 operands. Use the constraint character @samp{+} to indicate such an
3591 operand and list it with the output operands.
3593 When the constraints for the read-write operand (or the operand in which
3594 only some of the bits are to be changed) allows a register, you may, as
3595 an alternative, logically split its function into two separate operands,
3596 one input operand and one write-only output operand. The connection
3597 between them is expressed by constraints which say they need to be in
3598 the same location when the instruction executes. You can use the same C
3599 expression for both operands, or different expressions. For example,
3600 here we write the (fictitious) @samp{combine} instruction with
3601 @code{bar} as its read-only source operand and @code{foo} as its
3602 read-write destination:
3605 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3609 The constraint @samp{"0"} for operand 1 says that it must occupy the
3610 same location as operand 0. A number in constraint is allowed only in
3611 an input operand and it must refer to an output operand.
3613 Only a number in the constraint can guarantee that one operand will be in
3614 the same place as another. The mere fact that @code{foo} is the value
3615 of both operands is not enough to guarantee that they will be in the
3616 same place in the generated assembler code. The following would not
3620 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3623 Various optimizations or reloading could cause operands 0 and 1 to be in
3624 different registers; GCC knows no reason not to do so. For example, the
3625 compiler might find a copy of the value of @code{foo} in one register and
3626 use it for operand 1, but generate the output operand 0 in a different
3627 register (copying it afterward to @code{foo}'s own address). Of course,
3628 since the register for operand 1 is not even mentioned in the assembler
3629 code, the result will not work, but GCC can't tell that.
3631 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3632 the operand number for a matching constraint. For example:
3635 asm ("cmoveq %1,%2,%[result]"
3636 : [result] "=r"(result)
3637 : "r" (test), "r"(new), "[result]"(old));
3640 Some instructions clobber specific hard registers. To describe this,
3641 write a third colon after the input operands, followed by the names of
3642 the clobbered hard registers (given as strings). Here is a realistic
3643 example for the VAX:
3646 asm volatile ("movc3 %0,%1,%2"
3648 : "g" (from), "g" (to), "g" (count)
3649 : "r0", "r1", "r2", "r3", "r4", "r5");
3652 You may not write a clobber description in a way that overlaps with an
3653 input or output operand. For example, you may not have an operand
3654 describing a register class with one member if you mention that register
3655 in the clobber list. There is no way for you to specify that an input
3656 operand is modified without also specifying it as an output
3657 operand. Note that if all the output operands you specify are for this
3658 purpose (and hence unused), you will then also need to specify
3659 @code{volatile} for the @code{asm} construct, as described below, to
3660 prevent GCC from deleting the @code{asm} statement as unused.
3662 If you refer to a particular hardware register from the assembler code,
3663 you will probably have to list the register after the third colon to
3664 tell the compiler the register's value is modified. In some assemblers,
3665 the register names begin with @samp{%}; to produce one @samp{%} in the
3666 assembler code, you must write @samp{%%} in the input.
3668 If your assembler instruction can alter the condition code register, add
3669 @samp{cc} to the list of clobbered registers. GCC on some machines
3670 represents the condition codes as a specific hardware register;
3671 @samp{cc} serves to name this register. On other machines, the
3672 condition code is handled differently, and specifying @samp{cc} has no
3673 effect. But it is valid no matter what the machine.
3675 If your assembler instruction modifies memory in an unpredictable
3676 fashion, add @samp{memory} to the list of clobbered registers. This
3677 will cause GCC to not keep memory values cached in registers across
3678 the assembler instruction. You will also want to add the
3679 @code{volatile} keyword if the memory affected is not listed in the
3680 inputs or outputs of the @code{asm}, as the @samp{memory} clobber does
3681 not count as a side-effect of the @code{asm}.
3683 You can put multiple assembler instructions together in a single
3684 @code{asm} template, separated by the characters normally used in assembly
3685 code for the system. A combination that works in most places is a newline
3686 to break the line, plus a tab character to move to the instruction field
3687 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3688 assembler allows semicolons as a line-breaking character. Note that some
3689 assembler dialects use semicolons to start a comment.
3690 The input operands are guaranteed not to use any of the clobbered
3691 registers, and neither will the output operands' addresses, so you can
3692 read and write the clobbered registers as many times as you like. Here
3693 is an example of multiple instructions in a template; it assumes the
3694 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3697 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3699 : "g" (from), "g" (to)
3703 Unless an output operand has the @samp{&} constraint modifier, GCC
3704 may allocate it in the same register as an unrelated input operand, on
3705 the assumption the inputs are consumed before the outputs are produced.
3706 This assumption may be false if the assembler code actually consists of
3707 more than one instruction. In such a case, use @samp{&} for each output
3708 operand that may not overlap an input. @xref{Modifiers}.
3710 If you want to test the condition code produced by an assembler
3711 instruction, you must include a branch and a label in the @code{asm}
3712 construct, as follows:
3715 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3721 This assumes your assembler supports local labels, as the GNU assembler
3722 and most Unix assemblers do.
3724 Speaking of labels, jumps from one @code{asm} to another are not
3725 supported. The compiler's optimizers do not know about these jumps, and
3726 therefore they cannot take account of them when deciding how to
3729 @cindex macros containing @code{asm}
3730 Usually the most convenient way to use these @code{asm} instructions is to
3731 encapsulate them in macros that look like functions. For example,
3735 (@{ double __value, __arg = (x); \
3736 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3741 Here the variable @code{__arg} is used to make sure that the instruction
3742 operates on a proper @code{double} value, and to accept only those
3743 arguments @code{x} which can convert automatically to a @code{double}.
3745 Another way to make sure the instruction operates on the correct data
3746 type is to use a cast in the @code{asm}. This is different from using a
3747 variable @code{__arg} in that it converts more different types. For
3748 example, if the desired type were @code{int}, casting the argument to
3749 @code{int} would accept a pointer with no complaint, while assigning the
3750 argument to an @code{int} variable named @code{__arg} would warn about
3751 using a pointer unless the caller explicitly casts it.
3753 If an @code{asm} has output operands, GCC assumes for optimization
3754 purposes the instruction has no side effects except to change the output
3755 operands. This does not mean instructions with a side effect cannot be
3756 used, but you must be careful, because the compiler may eliminate them
3757 if the output operands aren't used, or move them out of loops, or
3758 replace two with one if they constitute a common subexpression. Also,
3759 if your instruction does have a side effect on a variable that otherwise
3760 appears not to change, the old value of the variable may be reused later
3761 if it happens to be found in a register.
3763 You can prevent an @code{asm} instruction from being deleted, moved
3764 significantly, or combined, by writing the keyword @code{volatile} after
3765 the @code{asm}. For example:
3768 #define get_and_set_priority(new) \
3770 asm volatile ("get_and_set_priority %0, %1" \
3771 : "=g" (__old) : "g" (new)); \
3776 If you write an @code{asm} instruction with no outputs, GCC will know
3777 the instruction has side-effects and will not delete the instruction or
3778 move it outside of loops.
3780 The @code{volatile} keyword indicates that the instruction has
3781 important side-effects. GCC will not delete a volatile @code{asm} if
3782 it is reachable. (The instruction can still be deleted if GCC can
3783 prove that control-flow will never reach the location of the
3784 instruction.) In addition, GCC will not reschedule instructions
3785 across a volatile @code{asm} instruction. For example:
3788 *(volatile int *)addr = foo;
3789 asm volatile ("eieio" : : );
3793 Assume @code{addr} contains the address of a memory mapped device
3794 register. The PowerPC @code{eieio} instruction (Enforce In-order
3795 Execution of I/O) tells the CPU to make sure that the store to that
3796 device register happens before it issues any other I/O@.
3798 Note that even a volatile @code{asm} instruction can be moved in ways
3799 that appear insignificant to the compiler, such as across jump
3800 instructions. You can't expect a sequence of volatile @code{asm}
3801 instructions to remain perfectly consecutive. If you want consecutive
3802 output, use a single @code{asm}. Also, GCC will perform some
3803 optimizations across a volatile @code{asm} instruction; GCC does not
3804 ``forget everything'' when it encounters a volatile @code{asm}
3805 instruction the way some other compilers do.
3807 An @code{asm} instruction without any operands or clobbers (an ``old
3808 style'' @code{asm}) will be treated identically to a volatile
3809 @code{asm} instruction.
3811 It is a natural idea to look for a way to give access to the condition
3812 code left by the assembler instruction. However, when we attempted to
3813 implement this, we found no way to make it work reliably. The problem
3814 is that output operands might need reloading, which would result in
3815 additional following ``store'' instructions. On most machines, these
3816 instructions would alter the condition code before there was time to
3817 test it. This problem doesn't arise for ordinary ``test'' and
3818 ``compare'' instructions because they don't have any output operands.
3820 For reasons similar to those described above, it is not possible to give
3821 an assembler instruction access to the condition code left by previous
3824 If you are writing a header file that should be includable in ISO C
3825 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3828 @subsection i386 floating point asm operands
3830 There are several rules on the usage of stack-like regs in
3831 asm_operands insns. These rules apply only to the operands that are
3836 Given a set of input regs that die in an asm_operands, it is
3837 necessary to know which are implicitly popped by the asm, and
3838 which must be explicitly popped by gcc.
3840 An input reg that is implicitly popped by the asm must be
3841 explicitly clobbered, unless it is constrained to match an
3845 For any input reg that is implicitly popped by an asm, it is
3846 necessary to know how to adjust the stack to compensate for the pop.
3847 If any non-popped input is closer to the top of the reg-stack than
3848 the implicitly popped reg, it would not be possible to know what the
3849 stack looked like---it's not clear how the rest of the stack ``slides
3852 All implicitly popped input regs must be closer to the top of
3853 the reg-stack than any input that is not implicitly popped.
3855 It is possible that if an input dies in an insn, reload might
3856 use the input reg for an output reload. Consider this example:
3859 asm ("foo" : "=t" (a) : "f" (b));
3862 This asm says that input B is not popped by the asm, and that
3863 the asm pushes a result onto the reg-stack, i.e., the stack is one
3864 deeper after the asm than it was before. But, it is possible that
3865 reload will think that it can use the same reg for both the input and
3866 the output, if input B dies in this insn.
3868 If any input operand uses the @code{f} constraint, all output reg
3869 constraints must use the @code{&} earlyclobber.
3871 The asm above would be written as
3874 asm ("foo" : "=&t" (a) : "f" (b));
3878 Some operands need to be in particular places on the stack. All
3879 output operands fall in this category---there is no other way to
3880 know which regs the outputs appear in unless the user indicates
3881 this in the constraints.
3883 Output operands must specifically indicate which reg an output
3884 appears in after an asm. @code{=f} is not allowed: the operand
3885 constraints must select a class with a single reg.
3888 Output operands may not be ``inserted'' between existing stack regs.
3889 Since no 387 opcode uses a read/write operand, all output operands
3890 are dead before the asm_operands, and are pushed by the asm_operands.
3891 It makes no sense to push anywhere but the top of the reg-stack.
3893 Output operands must start at the top of the reg-stack: output
3894 operands may not ``skip'' a reg.
3897 Some asm statements may need extra stack space for internal
3898 calculations. This can be guaranteed by clobbering stack registers
3899 unrelated to the inputs and outputs.
3903 Here are a couple of reasonable asms to want to write. This asm
3904 takes one input, which is internally popped, and produces two outputs.
3907 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
3910 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
3911 and replaces them with one output. The user must code the @code{st(1)}
3912 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
3915 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
3921 @section Controlling Names Used in Assembler Code
3922 @cindex assembler names for identifiers
3923 @cindex names used in assembler code
3924 @cindex identifiers, names in assembler code
3926 You can specify the name to be used in the assembler code for a C
3927 function or variable by writing the @code{asm} (or @code{__asm__})
3928 keyword after the declarator as follows:
3931 int foo asm ("myfoo") = 2;
3935 This specifies that the name to be used for the variable @code{foo} in
3936 the assembler code should be @samp{myfoo} rather than the usual
3939 On systems where an underscore is normally prepended to the name of a C
3940 function or variable, this feature allows you to define names for the
3941 linker that do not start with an underscore.
3943 It does not make sense to use this feature with a non-static local
3944 variable since such variables do not have assembler names. If you are
3945 trying to put the variable in a particular register, see @ref{Explicit
3946 Reg Vars}. GCC presently accepts such code with a warning, but will
3947 probably be changed to issue an error, rather than a warning, in the
3950 You cannot use @code{asm} in this way in a function @emph{definition}; but
3951 you can get the same effect by writing a declaration for the function
3952 before its definition and putting @code{asm} there, like this:
3955 extern func () asm ("FUNC");
3962 It is up to you to make sure that the assembler names you choose do not
3963 conflict with any other assembler symbols. Also, you must not use a
3964 register name; that would produce completely invalid assembler code. GCC
3965 does not as yet have the ability to store static variables in registers.
3966 Perhaps that will be added.
3968 @node Explicit Reg Vars
3969 @section Variables in Specified Registers
3970 @cindex explicit register variables
3971 @cindex variables in specified registers
3972 @cindex specified registers
3973 @cindex registers, global allocation
3975 GNU C allows you to put a few global variables into specified hardware
3976 registers. You can also specify the register in which an ordinary
3977 register variable should be allocated.
3981 Global register variables reserve registers throughout the program.
3982 This may be useful in programs such as programming language
3983 interpreters which have a couple of global variables that are accessed
3987 Local register variables in specific registers do not reserve the
3988 registers. The compiler's data flow analysis is capable of determining
3989 where the specified registers contain live values, and where they are
3990 available for other uses. Stores into local register variables may be deleted
3991 when they appear to be dead according to dataflow analysis. References
3992 to local register variables may be deleted or moved or simplified.
3994 These local variables are sometimes convenient for use with the extended
3995 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
3996 output of the assembler instruction directly into a particular register.
3997 (This will work provided the register you specify fits the constraints
3998 specified for that operand in the @code{asm}.)
4006 @node Global Reg Vars
4007 @subsection Defining Global Register Variables
4008 @cindex global register variables
4009 @cindex registers, global variables in
4011 You can define a global register variable in GNU C like this:
4014 register int *foo asm ("a5");
4018 Here @code{a5} is the name of the register which should be used. Choose a
4019 register which is normally saved and restored by function calls on your
4020 machine, so that library routines will not clobber it.
4022 Naturally the register name is cpu-dependent, so you would need to
4023 conditionalize your program according to cpu type. The register
4024 @code{a5} would be a good choice on a 68000 for a variable of pointer
4025 type. On machines with register windows, be sure to choose a ``global''
4026 register that is not affected magically by the function call mechanism.
4028 In addition, operating systems on one type of cpu may differ in how they
4029 name the registers; then you would need additional conditionals. For
4030 example, some 68000 operating systems call this register @code{%a5}.
4032 Eventually there may be a way of asking the compiler to choose a register
4033 automatically, but first we need to figure out how it should choose and
4034 how to enable you to guide the choice. No solution is evident.
4036 Defining a global register variable in a certain register reserves that
4037 register entirely for this use, at least within the current compilation.
4038 The register will not be allocated for any other purpose in the functions
4039 in the current compilation. The register will not be saved and restored by
4040 these functions. Stores into this register are never deleted even if they
4041 would appear to be dead, but references may be deleted or moved or
4044 It is not safe to access the global register variables from signal
4045 handlers, or from more than one thread of control, because the system
4046 library routines may temporarily use the register for other things (unless
4047 you recompile them specially for the task at hand).
4049 @cindex @code{qsort}, and global register variables
4050 It is not safe for one function that uses a global register variable to
4051 call another such function @code{foo} by way of a third function
4052 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4053 different source file in which the variable wasn't declared). This is
4054 because @code{lose} might save the register and put some other value there.
4055 For example, you can't expect a global register variable to be available in
4056 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4057 might have put something else in that register. (If you are prepared to
4058 recompile @code{qsort} with the same global register variable, you can
4059 solve this problem.)
4061 If you want to recompile @code{qsort} or other source files which do not
4062 actually use your global register variable, so that they will not use that
4063 register for any other purpose, then it suffices to specify the compiler
4064 option @option{-ffixed-@var{reg}}. You need not actually add a global
4065 register declaration to their source code.
4067 A function which can alter the value of a global register variable cannot
4068 safely be called from a function compiled without this variable, because it
4069 could clobber the value the caller expects to find there on return.
4070 Therefore, the function which is the entry point into the part of the
4071 program that uses the global register variable must explicitly save and
4072 restore the value which belongs to its caller.
4074 @cindex register variable after @code{longjmp}
4075 @cindex global register after @code{longjmp}
4076 @cindex value after @code{longjmp}
4079 On most machines, @code{longjmp} will restore to each global register
4080 variable the value it had at the time of the @code{setjmp}. On some
4081 machines, however, @code{longjmp} will not change the value of global
4082 register variables. To be portable, the function that called @code{setjmp}
4083 should make other arrangements to save the values of the global register
4084 variables, and to restore them in a @code{longjmp}. This way, the same
4085 thing will happen regardless of what @code{longjmp} does.
4087 All global register variable declarations must precede all function
4088 definitions. If such a declaration could appear after function
4089 definitions, the declaration would be too late to prevent the register from
4090 being used for other purposes in the preceding functions.
4092 Global register variables may not have initial values, because an
4093 executable file has no means to supply initial contents for a register.
4095 On the Sparc, there are reports that g3 @dots{} g7 are suitable
4096 registers, but certain library functions, such as @code{getwd}, as well
4097 as the subroutines for division and remainder, modify g3 and g4. g1 and
4098 g2 are local temporaries.
4100 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4101 Of course, it will not do to use more than a few of those.
4103 @node Local Reg Vars
4104 @subsection Specifying Registers for Local Variables
4105 @cindex local variables, specifying registers
4106 @cindex specifying registers for local variables
4107 @cindex registers for local variables
4109 You can define a local register variable with a specified register
4113 register int *foo asm ("a5");
4117 Here @code{a5} is the name of the register which should be used. Note
4118 that this is the same syntax used for defining global register
4119 variables, but for a local variable it would appear within a function.
4121 Naturally the register name is cpu-dependent, but this is not a
4122 problem, since specific registers are most often useful with explicit
4123 assembler instructions (@pxref{Extended Asm}). Both of these things
4124 generally require that you conditionalize your program according to
4127 In addition, operating systems on one type of cpu may differ in how they
4128 name the registers; then you would need additional conditionals. For
4129 example, some 68000 operating systems call this register @code{%a5}.
4131 Defining such a register variable does not reserve the register; it
4132 remains available for other uses in places where flow control determines
4133 the variable's value is not live. However, these registers are made
4134 unavailable for use in the reload pass; excessive use of this feature
4135 leaves the compiler too few available registers to compile certain
4138 This option does not guarantee that GCC will generate code that has
4139 this variable in the register you specify at all times. You may not
4140 code an explicit reference to this register in an @code{asm} statement
4141 and assume it will always refer to this variable.
4143 Stores into local register variables may be deleted when they appear to be dead
4144 according to dataflow analysis. References to local register variables may
4145 be deleted or moved or simplified.
4147 @node Alternate Keywords
4148 @section Alternate Keywords
4149 @cindex alternate keywords
4150 @cindex keywords, alternate
4152 @option{-ansi} and the various @option{-std} options disable certain
4153 keywords. This causes trouble when you want to use GNU C extensions, or
4154 a general-purpose header file that should be usable by all programs,
4155 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4156 @code{inline} are not available in programs compiled with
4157 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4158 program compiled with @option{-std=c99}). The ISO C99 keyword
4159 @code{restrict} is only available when @option{-std=gnu99} (which will
4160 eventually be the default) or @option{-std=c99} (or the equivalent
4161 @option{-std=iso9899:1999}) is used.
4163 The way to solve these problems is to put @samp{__} at the beginning and
4164 end of each problematical keyword. For example, use @code{__asm__}
4165 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4167 Other C compilers won't accept these alternative keywords; if you want to
4168 compile with another compiler, you can define the alternate keywords as
4169 macros to replace them with the customary keywords. It looks like this:
4177 @findex __extension__
4179 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4181 prevent such warnings within one expression by writing
4182 @code{__extension__} before the expression. @code{__extension__} has no
4183 effect aside from this.
4185 @node Incomplete Enums
4186 @section Incomplete @code{enum} Types
4188 You can define an @code{enum} tag without specifying its possible values.
4189 This results in an incomplete type, much like what you get if you write
4190 @code{struct foo} without describing the elements. A later declaration
4191 which does specify the possible values completes the type.
4193 You can't allocate variables or storage using the type while it is
4194 incomplete. However, you can work with pointers to that type.
4196 This extension may not be very useful, but it makes the handling of
4197 @code{enum} more consistent with the way @code{struct} and @code{union}
4200 This extension is not supported by GNU C++.
4202 @node Function Names
4203 @section Function Names as Strings
4204 @cindex @code{__FUNCTION__} identifier
4205 @cindex @code{__PRETTY_FUNCTION__} identifier
4206 @cindex @code{__func__} identifier
4208 GCC predefines two magic identifiers to hold the name of the current
4209 function. The identifier @code{__FUNCTION__} holds the name of the function
4210 as it appears in the source. The identifier @code{__PRETTY_FUNCTION__}
4211 holds the name of the function pretty printed in a language specific
4214 These names are always the same in a C function, but in a C++ function
4215 they may be different. For example, this program:
4219 extern int printf (char *, ...);
4226 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4227 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4245 __PRETTY_FUNCTION__ = int a::sub (int)
4248 The compiler automagically replaces the identifiers with a string
4249 literal containing the appropriate name. Thus, they are neither
4250 preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor
4251 variables. This means that they catenate with other string literals, and
4252 that they can be used to initialize char arrays. For example
4255 char here[] = "Function " __FUNCTION__ " in " __FILE__;
4258 On the other hand, @samp{#ifdef __FUNCTION__} does not have any special
4259 meaning inside a function, since the preprocessor does not do anything
4260 special with the identifier @code{__FUNCTION__}.
4262 Note that these semantics are deprecated, and that GCC 3.2 will handle
4263 @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} the same way as
4264 @code{__func__}. @code{__func__} is defined by the ISO standard C99:
4267 The identifier @code{__func__} is implicitly declared by the translator
4268 as if, immediately following the opening brace of each function
4269 definition, the declaration
4272 static const char __func__[] = "function-name";
4275 appeared, where function-name is the name of the lexically-enclosing
4276 function. This name is the unadorned name of the function.
4279 By this definition, @code{__func__} is a variable, not a string literal.
4280 In particular, @code{__func__} does not catenate with other string
4283 In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are
4284 variables, declared in the same way as @code{__func__}.
4286 @node Return Address
4287 @section Getting the Return or Frame Address of a Function
4289 These functions may be used to get information about the callers of a
4292 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4293 This function returns the return address of the current function, or of
4294 one of its callers. The @var{level} argument is number of frames to
4295 scan up the call stack. A value of @code{0} yields the return address
4296 of the current function, a value of @code{1} yields the return address
4297 of the caller of the current function, and so forth.
4299 The @var{level} argument must be a constant integer.
4301 On some machines it may be impossible to determine the return address of
4302 any function other than the current one; in such cases, or when the top
4303 of the stack has been reached, this function will return @code{0} or a
4304 random value. In addition, @code{__builtin_frame_address} may be used
4305 to determine if the top of the stack has been reached.
4307 This function should only be used with a nonzero argument for debugging
4311 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4312 This function is similar to @code{__builtin_return_address}, but it
4313 returns the address of the function frame rather than the return address
4314 of the function. Calling @code{__builtin_frame_address} with a value of
4315 @code{0} yields the frame address of the current function, a value of
4316 @code{1} yields the frame address of the caller of the current function,
4319 The frame is the area on the stack which holds local variables and saved
4320 registers. The frame address is normally the address of the first word
4321 pushed on to the stack by the function. However, the exact definition
4322 depends upon the processor and the calling convention. If the processor
4323 has a dedicated frame pointer register, and the function has a frame,
4324 then @code{__builtin_frame_address} will return the value of the frame
4327 On some machines it may be impossible to determine the frame address of
4328 any function other than the current one; in such cases, or when the top
4329 of the stack has been reached, this function will return @code{0} if
4330 the first frame pointer is properly initialized by the startup code.
4332 This function should only be used with a nonzero argument for debugging
4336 @node Vector Extensions
4337 @section Using vector instructions through built-in functions
4339 On some targets, the instruction set contains SIMD vector instructions that
4340 operate on multiple values contained in one large register at the same time.
4341 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4344 The first step in using these extensions is to provide the necessary data
4345 types. This should be done using an appropriate @code{typedef}:
4348 typedef int v4si __attribute__ ((mode(V4SI)));
4351 The base type @code{int} is effectively ignored by the compiler, the
4352 actual properties of the new type @code{v4si} are defined by the
4353 @code{__attribute__}. It defines the machine mode to be used; for vector
4354 types these have the form @code{V@var{n}@var{B}}; @var{n} should be the
4355 number of elements in the vector, and @var{B} should be the base mode of the
4356 individual elements. The following can be used as base modes:
4360 An integer that is as wide as the smallest addressable unit, usually 8 bits.
4362 An integer, twice as wide as a QI mode integer, usually 16 bits.
4364 An integer, four times as wide as a QI mode integer, usually 32 bits.
4366 An integer, eight times as wide as a QI mode integer, usually 64 bits.
4368 A floating point value, as wide as a SI mode integer, usually 32 bits.
4370 A floating point value, as wide as a DI mode integer, usually 64 bits.
4373 Not all base types or combinations are always valid; which modes can be used
4374 is determined by the target machine. For example, if targetting the i386 MMX
4375 extensions, only @code{V8QI}, @code{V4HI} and @code{V2SI} are allowed modes.
4377 There are no @code{V1xx} vector modes - they would be identical to the
4378 corresponding base mode.
4380 There is no distinction between signed and unsigned vector modes. This
4381 distinction is made by the operations that perform on the vectors, not
4384 The types defined in this manner are somewhat special, they cannot be
4385 used with most normal C operations (i.e., a vector addition can @emph{not}
4386 be represented by a normal addition of two vector type variables). You
4387 can declare only variables and use them in function calls and returns, as
4388 well as in assignments and some casts. It is possible to cast from one
4389 vector type to another, provided they are of the same size (in fact, you
4390 can also cast vectors to and from other datatypes of the same size).
4392 A port that supports vector operations provides a set of built-in functions
4393 that can be used to operate on vectors. For example, a function to add two
4394 vectors and multiply the result by a third could look like this:
4397 v4si f (v4si a, v4si b, v4si c)
4399 v4si tmp = __builtin_addv4si (a, b);
4400 return __builtin_mulv4si (tmp, c);
4405 @node Other Builtins
4406 @section Other built-in functions provided by GCC
4407 @cindex built-in functions
4408 @findex __builtin_isgreater
4409 @findex __builtin_isgreaterequal
4410 @findex __builtin_isless
4411 @findex __builtin_islessequal
4412 @findex __builtin_islessgreater
4413 @findex __builtin_isunordered
4439 @findex fprintf_unlocked
4441 @findex fputs_unlocked
4450 @findex printf_unlocked
4472 GCC provides a large number of built-in functions other than the ones
4473 mentioned above. Some of these are for internal use in the processing
4474 of exceptions or variable-length argument lists and will not be
4475 documented here because they may change from time to time; we do not
4476 recommend general use of these functions.
4478 The remaining functions are provided for optimization purposes.
4480 @opindex fno-builtin
4481 GCC includes built-in versions of many of the functions in the standard
4482 C library. The versions prefixed with @code{__builtin_} will always be
4483 treated as having the same meaning as the C library function even if you
4484 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
4485 Many of these functions are only optimized in certain cases; if they are
4486 not optimized in a particular case, a call to the library function will
4491 The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit}
4492 are recognized and presumed not to return, but otherwise are not built
4493 in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi},
4494 @option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in
4495 strict C89 mode (@option{-ansi} or @option{-std=c89}).
4497 Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp},
4498 @code{bzero}, @code{index}, @code{rindex}, @code{ffs}, @code{fputs_unlocked},
4499 @code{printf_unlocked} and @code{fprintf_unlocked} may be handled as
4500 built-in functions. All these functions have corresponding versions
4501 prefixed with @code{__builtin_}, which may be used even in strict C89
4504 The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl},
4505 @code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf},
4506 @code{cimagl}, @code{llabs} and @code{imaxabs} are handled as built-in
4507 functions except in strict ISO C89 mode. There are also built-in
4508 versions of the ISO C99 functions @code{cosf}, @code{cosl},
4509 @code{fabsf}, @code{fabsl}, @code{sinf}, @code{sinl}, @code{sqrtf}, and
4510 @code{sqrtl}, that are recognized in any mode since ISO C89 reserves
4511 these names for the purpose to which ISO C99 puts them. All these
4512 functions have corresponding versions prefixed with @code{__builtin_}.
4514 The ISO C89 functions @code{abs}, @code{cos}, @code{fabs},
4515 @code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy},
4516 @code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat},
4517 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
4518 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
4519 @code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr} are all
4520 recognized as built-in functions unless @option{-fno-builtin} is
4521 specified (or @option{-fno-builtin-@var{function}} is specified for an
4522 individual function). All of these functions have corresponding
4523 versions prefixed with @code{__builtin_}.
4525 GCC provides built-in versions of the ISO C99 floating point comparison
4526 macros that avoid raising exceptions for unordered operands. They have
4527 the same names as the standard macros ( @code{isgreater},
4528 @code{isgreaterequal}, @code{isless}, @code{islessequal},
4529 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
4530 prefixed. We intend for a library implementor to be able to simply
4531 @code{#define} each standard macro to its built-in equivalent.
4533 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
4535 You can use the built-in function @code{__builtin_types_compatible_p} to
4536 determine whether two types are the same.
4538 This built-in function returns 1 if the unqualified versions of the
4539 types @var{type1} and @var{type2} (which are types, not expressions) are
4540 compatible, 0 otherwise. The result of this built-in function can be
4541 used in integer constant expressions.
4543 This built-in function ignores top level qualifiers (e.g., @code{const},
4544 @code{volatile}). For example, @code{int} is equivalent to @code{const
4547 The type @code{int[]} and @code{int[5]} are compatible. On the other
4548 hand, @code{int} and @code{char *} are not compatible, even if the size
4549 of their types, on the particular architecture are the same. Also, the
4550 amount of pointer indirection is taken into account when determining
4551 similarity. Consequently, @code{short *} is not similar to
4552 @code{short **}. Furthermore, two types that are typedefed are
4553 considered compatible if their underlying types are compatible.
4555 An @code{enum} type is considered to be compatible with another
4556 @code{enum} type. For example, @code{enum @{foo, bar@}} is similar to
4557 @code{enum @{hot, dog@}}.
4559 You would typically use this function in code whose execution varies
4560 depending on the arguments' types. For example:
4566 if (__builtin_types_compatible_p (typeof (x), long double)) \
4567 tmp = foo_long_double (tmp); \
4568 else if (__builtin_types_compatible_p (typeof (x), double)) \
4569 tmp = foo_double (tmp); \
4570 else if (__builtin_types_compatible_p (typeof (x), float)) \
4571 tmp = foo_float (tmp); \
4578 @emph{Note:} This construct is only available for C.
4582 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
4584 You can use the built-in function @code{__builtin_choose_expr} to
4585 evaluate code depending on the value of a constant expression. This
4586 built-in function returns @var{exp1} if @var{const_exp}, which is a
4587 constant expression that must be able to be determined at compile time,
4588 is nonzero. Otherwise it returns 0.
4590 This built-in function is analogous to the @samp{? :} operator in C,
4591 except that the expression returned has its type unaltered by promotion
4592 rules. Also, the built-in function does not evaluate the expression
4593 that was not chosen. For example, if @var{const_exp} evaluates to true,
4594 @var{exp2} is not evaluated even if it has side-effects.
4596 This built-in function can return an lvalue if the chosen argument is an
4599 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
4600 type. Similarly, if @var{exp2} is returned, its return type is the same
4607 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), double), \
4609 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), float), \
4611 /* @r{The void expression results in a compile-time error} \
4612 @r{when assigning the result to something.} */ \
4616 @emph{Note:} This construct is only available for C. Furthermore, the
4617 unused expression (@var{exp1} or @var{exp2} depending on the value of
4618 @var{const_exp}) may still generate syntax errors. This may change in
4623 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
4624 You can use the built-in function @code{__builtin_constant_p} to
4625 determine if a value is known to be constant at compile-time and hence
4626 that GCC can perform constant-folding on expressions involving that
4627 value. The argument of the function is the value to test. The function
4628 returns the integer 1 if the argument is known to be a compile-time
4629 constant and 0 if it is not known to be a compile-time constant. A
4630 return of 0 does not indicate that the value is @emph{not} a constant,
4631 but merely that GCC cannot prove it is a constant with the specified
4632 value of the @option{-O} option.
4634 You would typically use this function in an embedded application where
4635 memory was a critical resource. If you have some complex calculation,
4636 you may want it to be folded if it involves constants, but need to call
4637 a function if it does not. For example:
4640 #define Scale_Value(X) \
4641 (__builtin_constant_p (X) \
4642 ? ((X) * SCALE + OFFSET) : Scale (X))
4645 You may use this built-in function in either a macro or an inline
4646 function. However, if you use it in an inlined function and pass an
4647 argument of the function as the argument to the built-in, GCC will
4648 never return 1 when you call the inline function with a string constant
4649 or compound literal (@pxref{Compound Literals}) and will not return 1
4650 when you pass a constant numeric value to the inline function unless you
4651 specify the @option{-O} option.
4653 You may also use @code{__builtin_constant_p} in initializers for static
4654 data. For instance, you can write
4657 static const int table[] = @{
4658 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
4664 This is an acceptable initializer even if @var{EXPRESSION} is not a
4665 constant expression. GCC must be more conservative about evaluating the
4666 built-in in this case, because it has no opportunity to perform
4669 Previous versions of GCC did not accept this built-in in data
4670 initializers. The earliest version where it is completely safe is
4674 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
4675 @opindex fprofile-arcs
4676 You may use @code{__builtin_expect} to provide the compiler with
4677 branch prediction information. In general, you should prefer to
4678 use actual profile feedback for this (@option{-fprofile-arcs}), as
4679 programmers are notoriously bad at predicting how their programs
4680 actually perform. However, there are applications in which this
4681 data is hard to collect.
4683 The return value is the value of @var{exp}, which should be an
4684 integral expression. The value of @var{c} must be a compile-time
4685 constant. The semantics of the built-in are that it is expected
4686 that @var{exp} == @var{c}. For example:
4689 if (__builtin_expect (x, 0))
4694 would indicate that we do not expect to call @code{foo}, since
4695 we expect @code{x} to be zero. Since you are limited to integral
4696 expressions for @var{exp}, you should use constructions such as
4699 if (__builtin_expect (ptr != NULL, 1))
4704 when testing pointer or floating-point values.
4707 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
4708 This function is used to minimize cache-miss latency by moving data into
4709 a cache before it is accessed.
4710 You can insert calls to @code{__builtin_prefetch} into code for which
4711 you know addresses of data in memory that is likely to be accessed soon.
4712 If the target supports them, data prefetch instructions will be generated.
4713 If the prefetch is done early enough before the access then the data will
4714 be in the cache by the time it is accessed.
4716 The value of @var{addr} is the address of the memory to prefetch.
4717 There are two optional arguments, @var{rw} and @var{locality}.
4718 The value of @var{rw} is a compile-time constant one or zero; one
4719 means that the prefetch is preparing for a write to the memory address
4720 and zero, the default, means that the prefetch is preparing for a read.
4721 The value @var{locality} must be a compile-time constant integer between
4722 zero and three. A value of zero means that the data has no temporal
4723 locality, so it need not be left in the cache after the access. A value
4724 of three means that the data has a high degree of temporal locality and
4725 should be left in all levels of cache possible. Values of one and two
4726 mean, respectively, a low or moderate degree of temporal locality. The
4730 for (i = 0; i < n; i++)
4733 __builtin_prefetch (&a[i+j], 1, 1);
4734 __builtin_prefetch (&b[i+j], 0, 1);
4739 Data prefetch does not generate faults if @var{addr} is invalid, but
4740 the address expression itself must be valid. For example, a prefetch
4741 of @code{p->next} will not fault if @code{p->next} is not a valid
4742 address, but evaluation will fault if @code{p} is not a valid address.
4744 If the target does not support data prefetch, the address expression
4745 is evaluated if it includes side effects but no other code is generated
4746 and GCC does not issue a warning.
4749 @node Target Builtins
4750 @section Built-in Functions Specific to Particular Target Machines
4752 On some target machines, GCC supports many built-in functions specific
4753 to those machines. Generally these generate calls to specific machine
4754 instructions, but allow the compiler to schedule those calls.
4757 * Alpha Built-in Functions::
4758 * X86 Built-in Functions::
4759 * PowerPC AltiVec Built-in Functions::
4762 @node Alpha Built-in Functions
4763 @subsection Alpha Built-in Functions
4765 These built-in functions are available for the Alpha family of
4766 processors, depending on the command-line switches used.
4768 The following built-in functions are always available. They
4769 all generate the machine instruction that is part of the name.
4772 long __builtin_alpha_implver (void)
4773 long __builtin_alpha_rpcc (void)
4774 long __builtin_alpha_amask (long)
4775 long __builtin_alpha_cmpbge (long, long)
4776 long __builtin_alpha_extql (long, long)
4777 long __builtin_alpha_extqh (long, long)
4778 long __builtin_alpha_zap (long, long)
4779 long __builtin_alpha_zapnot (long, long)
4782 The following built-in functions are always with @option{-mmax}
4783 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
4784 later. They all generate the machine instruction that is part
4788 long __builtin_alpha_pklb (long)
4789 long __builtin_alpha_pkwb (long)
4790 long __builtin_alpha_unpkbl (long)
4791 long __builtin_alpha_unpkbw (long)
4792 long __builtin_alpha_minub8 (long, long)
4793 long __builtin_alpha_minsb8 (long, long)
4794 long __builtin_alpha_minuw4 (long, long)
4795 long __builtin_alpha_minsw4 (long, long)
4796 long __builtin_alpha_maxub8 (long, long)
4797 long __builtin_alpha_maxsb8 (long, long)
4798 long __builtin_alpha_maxuw4 (long, long)
4799 long __builtin_alpha_maxsw4 (long, long)
4800 long __builtin_alpha_perr (long, long)
4803 @node X86 Built-in Functions
4804 @subsection X86 Built-in Functions
4806 These built-in functions are available for the i386 and x86-64 family
4807 of computers, depending on the command-line switches used.
4809 The following machine modes are available for use with MMX built-in functions
4810 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
4811 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
4812 vector of eight 8-bit integers. Some of the built-in functions operate on
4813 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
4815 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
4816 of two 32-bit floating point values.
4818 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
4819 floating point values. Some instructions use a vector of four 32-bit
4820 integers, these use @code{V4SI}. Finally, some instructions operate on an
4821 entire vector register, interpreting it as a 128-bit integer, these use mode
4824 The following built-in functions are made available by @option{-mmmx}.
4825 All of them generate the machine instruction that is part of the name.
4828 v8qi __builtin_ia32_paddb (v8qi, v8qi)
4829 v4hi __builtin_ia32_paddw (v4hi, v4hi)
4830 v2si __builtin_ia32_paddd (v2si, v2si)
4831 v8qi __builtin_ia32_psubb (v8qi, v8qi)
4832 v4hi __builtin_ia32_psubw (v4hi, v4hi)
4833 v2si __builtin_ia32_psubd (v2si, v2si)
4834 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
4835 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
4836 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
4837 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
4838 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
4839 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
4840 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
4841 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
4842 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
4843 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
4844 di __builtin_ia32_pand (di, di)
4845 di __builtin_ia32_pandn (di,di)
4846 di __builtin_ia32_por (di, di)
4847 di __builtin_ia32_pxor (di, di)
4848 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
4849 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
4850 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
4851 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
4852 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
4853 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
4854 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
4855 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
4856 v2si __builtin_ia32_punpckhdq (v2si, v2si)
4857 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
4858 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
4859 v2si __builtin_ia32_punpckldq (v2si, v2si)
4860 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
4861 v4hi __builtin_ia32_packssdw (v2si, v2si)
4862 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
4865 The following built-in functions are made available either with
4866 @option{-msse}, or with a combination of @option{-m3dnow} and
4867 @option{-march=athlon}. All of them generate the machine
4868 instruction that is part of the name.
4871 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
4872 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
4873 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
4874 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
4875 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
4876 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
4877 v8qi __builtin_ia32_pminub (v8qi, v8qi)
4878 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
4879 int __builtin_ia32_pextrw (v4hi, int)
4880 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
4881 int __builtin_ia32_pmovmskb (v8qi)
4882 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
4883 void __builtin_ia32_movntq (di *, di)
4884 void __builtin_ia32_sfence (void)
4887 The following built-in functions are available when @option{-msse} is used.
4888 All of them generate the machine instruction that is part of the name.
4891 int __builtin_ia32_comieq (v4sf, v4sf)
4892 int __builtin_ia32_comineq (v4sf, v4sf)
4893 int __builtin_ia32_comilt (v4sf, v4sf)
4894 int __builtin_ia32_comile (v4sf, v4sf)
4895 int __builtin_ia32_comigt (v4sf, v4sf)
4896 int __builtin_ia32_comige (v4sf, v4sf)
4897 int __builtin_ia32_ucomieq (v4sf, v4sf)
4898 int __builtin_ia32_ucomineq (v4sf, v4sf)
4899 int __builtin_ia32_ucomilt (v4sf, v4sf)
4900 int __builtin_ia32_ucomile (v4sf, v4sf)
4901 int __builtin_ia32_ucomigt (v4sf, v4sf)
4902 int __builtin_ia32_ucomige (v4sf, v4sf)
4903 v4sf __builtin_ia32_addps (v4sf, v4sf)
4904 v4sf __builtin_ia32_subps (v4sf, v4sf)
4905 v4sf __builtin_ia32_mulps (v4sf, v4sf)
4906 v4sf __builtin_ia32_divps (v4sf, v4sf)
4907 v4sf __builtin_ia32_addss (v4sf, v4sf)
4908 v4sf __builtin_ia32_subss (v4sf, v4sf)
4909 v4sf __builtin_ia32_mulss (v4sf, v4sf)
4910 v4sf __builtin_ia32_divss (v4sf, v4sf)
4911 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
4912 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
4913 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
4914 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
4915 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
4916 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
4917 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
4918 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
4919 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
4920 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
4921 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
4922 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
4923 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
4924 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
4925 v4si __builtin_ia32_cmpless (v4sf, v4sf)
4926 v4si __builtin_ia32_cmpgtss (v4sf, v4sf)
4927 v4si __builtin_ia32_cmpgess (v4sf, v4sf)
4928 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
4929 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
4930 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
4931 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
4932 v4si __builtin_ia32_cmpngtss (v4sf, v4sf)
4933 v4si __builtin_ia32_cmpngess (v4sf, v4sf)
4934 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
4935 v4sf __builtin_ia32_maxps (v4sf, v4sf)
4936 v4sf __builtin_ia32_maxss (v4sf, v4sf)
4937 v4sf __builtin_ia32_minps (v4sf, v4sf)
4938 v4sf __builtin_ia32_minss (v4sf, v4sf)
4939 v4sf __builtin_ia32_andps (v4sf, v4sf)
4940 v4sf __builtin_ia32_andnps (v4sf, v4sf)
4941 v4sf __builtin_ia32_orps (v4sf, v4sf)
4942 v4sf __builtin_ia32_xorps (v4sf, v4sf)
4943 v4sf __builtin_ia32_movss (v4sf, v4sf)
4944 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
4945 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
4946 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
4947 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
4948 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
4949 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
4950 v2si __builtin_ia32_cvtps2pi (v4sf)
4951 int __builtin_ia32_cvtss2si (v4sf)
4952 v2si __builtin_ia32_cvttps2pi (v4sf)
4953 int __builtin_ia32_cvttss2si (v4sf)
4954 v4sf __builtin_ia32_rcpps (v4sf)
4955 v4sf __builtin_ia32_rsqrtps (v4sf)
4956 v4sf __builtin_ia32_sqrtps (v4sf)
4957 v4sf __builtin_ia32_rcpss (v4sf)
4958 v4sf __builtin_ia32_rsqrtss (v4sf)
4959 v4sf __builtin_ia32_sqrtss (v4sf)
4960 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
4961 void __builtin_ia32_movntps (float *, v4sf)
4962 int __builtin_ia32_movmskps (v4sf)
4965 The following built-in functions are available when @option{-msse} is used.
4968 @item v4sf __builtin_ia32_loadaps (float *)
4969 Generates the @code{movaps} machine instruction as a load from memory.
4970 @item void __builtin_ia32_storeaps (float *, v4sf)
4971 Generates the @code{movaps} machine instruction as a store to memory.
4972 @item v4sf __builtin_ia32_loadups (float *)
4973 Generates the @code{movups} machine instruction as a load from memory.
4974 @item void __builtin_ia32_storeups (float *, v4sf)
4975 Generates the @code{movups} machine instruction as a store to memory.
4976 @item v4sf __builtin_ia32_loadsss (float *)
4977 Generates the @code{movss} machine instruction as a load from memory.
4978 @item void __builtin_ia32_storess (float *, v4sf)
4979 Generates the @code{movss} machine instruction as a store to memory.
4980 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
4981 Generates the @code{movhps} machine instruction as a load from memory.
4982 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
4983 Generates the @code{movlps} machine instruction as a load from memory
4984 @item void __builtin_ia32_storehps (v4sf, v2si *)
4985 Generates the @code{movhps} machine instruction as a store to memory.
4986 @item void __builtin_ia32_storelps (v4sf, v2si *)
4987 Generates the @code{movlps} machine instruction as a store to memory.
4990 The following built-in functions are available when @option{-m3dnow} is used.
4991 All of them generate the machine instruction that is part of the name.
4994 void __builtin_ia32_femms (void)
4995 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
4996 v2si __builtin_ia32_pf2id (v2sf)
4997 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
4998 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
4999 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
5000 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
5001 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
5002 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
5003 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
5004 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
5005 v2sf __builtin_ia32_pfrcp (v2sf)
5006 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
5007 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
5008 v2sf __builtin_ia32_pfrsqrt (v2sf)
5009 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
5010 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
5011 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
5012 v2sf __builtin_ia32_pi2fd (v2si)
5013 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
5016 The following built-in functions are available when both @option{-m3dnow}
5017 and @option{-march=athlon} are used. All of them generate the machine
5018 instruction that is part of the name.
5021 v2si __builtin_ia32_pf2iw (v2sf)
5022 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
5023 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
5024 v2sf __builtin_ia32_pi2fw (v2si)
5025 v2sf __builtin_ia32_pswapdsf (v2sf)
5026 v2si __builtin_ia32_pswapdsi (v2si)
5029 @node PowerPC AltiVec Built-in Functions
5030 @subsection PowerPC AltiVec Built-in Functions
5032 These built-in functions are available for the PowerPC family
5033 of computers, depending on the command-line switches used.
5035 The following machine modes are available for use with AltiVec built-in
5036 functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four
5037 32-bit integers, @code{V4SF} for a vector of four 32-bit floating point
5038 numbers, @code{V8HI} for a vector of eight 16-bit integers, and
5039 @code{V16QI} for a vector of sixteen 8-bit integers.
5041 The following functions are made available by including
5042 @code{<altivec.h>} and using @option{-maltivec} and
5043 @option{-mabi=altivec}. The functions implement the functionality
5044 described in Motorola's AltiVec Programming Interface Manual.
5046 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
5047 Internally, GCC uses built-in functions to achieve the functionality in
5048 the aforementioned header file, but they are not supported and are
5049 subject to change without notice.
5052 vector signed char vec_abs (vector signed char, vector signed char);
5053 vector signed short vec_abs (vector signed short, vector signed short);
5054 vector signed int vec_abs (vector signed int, vector signed int);
5055 vector signed float vec_abs (vector signed float, vector signed float);
5057 vector signed char vec_abss (vector signed char, vector signed char);
5058 vector signed short vec_abss (vector signed short, vector signed short);
5060 vector signed char vec_add (vector signed char, vector signed char);
5061 vector unsigned char vec_add (vector signed char, vector unsigned char);
5063 vector unsigned char vec_add (vector unsigned char, vector signed char);
5065 vector unsigned char vec_add (vector unsigned char,
5066 vector unsigned char);
5067 vector signed short vec_add (vector signed short, vector signed short);
5068 vector unsigned short vec_add (vector signed short,
5069 vector unsigned short);
5070 vector unsigned short vec_add (vector unsigned short,
5071 vector signed short);
5072 vector unsigned short vec_add (vector unsigned short,
5073 vector unsigned short);
5074 vector signed int vec_add (vector signed int, vector signed int);
5075 vector unsigned int vec_add (vector signed int, vector unsigned int);
5076 vector unsigned int vec_add (vector unsigned int, vector signed int);
5077 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
5078 vector float vec_add (vector float, vector float);
5080 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
5082 vector unsigned char vec_adds (vector signed char,
5083 vector unsigned char);
5084 vector unsigned char vec_adds (vector unsigned char,
5085 vector signed char);
5086 vector unsigned char vec_adds (vector unsigned char,
5087 vector unsigned char);
5088 vector signed char vec_adds (vector signed char, vector signed char);
5089 vector unsigned short vec_adds (vector signed short,
5090 vector unsigned short);
5091 vector unsigned short vec_adds (vector unsigned short,
5092 vector signed short);
5093 vector unsigned short vec_adds (vector unsigned short,
5094 vector unsigned short);
5095 vector signed short vec_adds (vector signed short, vector signed short);
5097 vector unsigned int vec_adds (vector signed int, vector unsigned int);
5098 vector unsigned int vec_adds (vector unsigned int, vector signed int);
5099 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
5101 vector signed int vec_adds (vector signed int, vector signed int);
5103 vector float vec_and (vector float, vector float);
5104 vector float vec_and (vector float, vector signed int);
5105 vector float vec_and (vector signed int, vector float);
5106 vector signed int vec_and (vector signed int, vector signed int);
5107 vector unsigned int vec_and (vector signed int, vector unsigned int);
5108 vector unsigned int vec_and (vector unsigned int, vector signed int);
5109 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
5110 vector signed short vec_and (vector signed short, vector signed short);
5111 vector unsigned short vec_and (vector signed short,
5112 vector unsigned short);
5113 vector unsigned short vec_and (vector unsigned short,
5114 vector signed short);
5115 vector unsigned short vec_and (vector unsigned short,
5116 vector unsigned short);
5117 vector signed char vec_and (vector signed char, vector signed char);
5118 vector unsigned char vec_and (vector signed char, vector unsigned char);
5120 vector unsigned char vec_and (vector unsigned char, vector signed char);
5122 vector unsigned char vec_and (vector unsigned char,
5123 vector unsigned char);
5125 vector float vec_andc (vector float, vector float);
5126 vector float vec_andc (vector float, vector signed int);
5127 vector float vec_andc (vector signed int, vector float);
5128 vector signed int vec_andc (vector signed int, vector signed int);
5129 vector unsigned int vec_andc (vector signed int, vector unsigned int);
5130 vector unsigned int vec_andc (vector unsigned int, vector signed int);
5131 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
5133 vector signed short vec_andc (vector signed short, vector signed short);
5135 vector unsigned short vec_andc (vector signed short,
5136 vector unsigned short);
5137 vector unsigned short vec_andc (vector unsigned short,
5138 vector signed short);
5139 vector unsigned short vec_andc (vector unsigned short,
5140 vector unsigned short);
5141 vector signed char vec_andc (vector signed char, vector signed char);
5142 vector unsigned char vec_andc (vector signed char,
5143 vector unsigned char);
5144 vector unsigned char vec_andc (vector unsigned char,
5145 vector signed char);
5146 vector unsigned char vec_andc (vector unsigned char,
5147 vector unsigned char);
5149 vector unsigned char vec_avg (vector unsigned char,
5150 vector unsigned char);
5151 vector signed char vec_avg (vector signed char, vector signed char);
5152 vector unsigned short vec_avg (vector unsigned short,
5153 vector unsigned short);
5154 vector signed short vec_avg (vector signed short, vector signed short);
5155 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
5156 vector signed int vec_avg (vector signed int, vector signed int);
5158 vector float vec_ceil (vector float);
5160 vector signed int vec_cmpb (vector float, vector float);
5162 vector signed char vec_cmpeq (vector signed char, vector signed char);
5163 vector signed char vec_cmpeq (vector unsigned char,
5164 vector unsigned char);
5165 vector signed short vec_cmpeq (vector signed short,
5166 vector signed short);
5167 vector signed short vec_cmpeq (vector unsigned short,
5168 vector unsigned short);
5169 vector signed int vec_cmpeq (vector signed int, vector signed int);
5170 vector signed int vec_cmpeq (vector unsigned int, vector unsigned int);
5171 vector signed int vec_cmpeq (vector float, vector float);
5173 vector signed int vec_cmpge (vector float, vector float);
5175 vector signed char vec_cmpgt (vector unsigned char,
5176 vector unsigned char);
5177 vector signed char vec_cmpgt (vector signed char, vector signed char);
5178 vector signed short vec_cmpgt (vector unsigned short,
5179 vector unsigned short);
5180 vector signed short vec_cmpgt (vector signed short,
5181 vector signed short);
5182 vector signed int vec_cmpgt (vector unsigned int, vector unsigned int);
5183 vector signed int vec_cmpgt (vector signed int, vector signed int);
5184 vector signed int vec_cmpgt (vector float, vector float);
5186 vector signed int vec_cmple (vector float, vector float);
5188 vector signed char vec_cmplt (vector unsigned char,
5189 vector unsigned char);
5190 vector signed char vec_cmplt (vector signed char, vector signed char);
5191 vector signed short vec_cmplt (vector unsigned short,
5192 vector unsigned short);
5193 vector signed short vec_cmplt (vector signed short,
5194 vector signed short);
5195 vector signed int vec_cmplt (vector unsigned int, vector unsigned int);
5196 vector signed int vec_cmplt (vector signed int, vector signed int);
5197 vector signed int vec_cmplt (vector float, vector float);
5199 vector float vec_ctf (vector unsigned int, const char);
5200 vector float vec_ctf (vector signed int, const char);
5202 vector signed int vec_cts (vector float, const char);
5204 vector unsigned int vec_ctu (vector float, const char);
5206 void vec_dss (const char);
5208 void vec_dssall (void);
5210 void vec_dst (void *, int, const char);
5212 void vec_dstst (void *, int, const char);
5214 void vec_dststt (void *, int, const char);
5216 void vec_dstt (void *, int, const char);
5218 vector float vec_expte (vector float, vector float);
5220 vector float vec_floor (vector float, vector float);
5222 vector float vec_ld (int, vector float *);
5223 vector float vec_ld (int, float *):
5224 vector signed int vec_ld (int, int *);
5225 vector signed int vec_ld (int, vector signed int *);
5226 vector unsigned int vec_ld (int, vector unsigned int *);
5227 vector unsigned int vec_ld (int, unsigned int *);
5228 vector signed short vec_ld (int, short *, vector signed short *);
5229 vector unsigned short vec_ld (int, unsigned short *,
5230 vector unsigned short *);
5231 vector signed char vec_ld (int, signed char *);
5232 vector signed char vec_ld (int, vector signed char *);
5233 vector unsigned char vec_ld (int, unsigned char *);
5234 vector unsigned char vec_ld (int, vector unsigned char *);
5236 vector signed char vec_lde (int, signed char *);
5237 vector unsigned char vec_lde (int, unsigned char *);
5238 vector signed short vec_lde (int, short *);
5239 vector unsigned short vec_lde (int, unsigned short *);
5240 vector float vec_lde (int, float *);
5241 vector signed int vec_lde (int, int *);
5242 vector unsigned int vec_lde (int, unsigned int *);
5244 void float vec_ldl (int, float *);
5245 void float vec_ldl (int, vector float *);
5246 void signed int vec_ldl (int, vector signed int *);
5247 void signed int vec_ldl (int, int *);
5248 void unsigned int vec_ldl (int, unsigned int *);
5249 void unsigned int vec_ldl (int, vector unsigned int *);
5250 void signed short vec_ldl (int, vector signed short *);
5251 void signed short vec_ldl (int, short *);
5252 void unsigned short vec_ldl (int, vector unsigned short *);
5253 void unsigned short vec_ldl (int, unsigned short *);
5254 void signed char vec_ldl (int, vector signed char *);
5255 void signed char vec_ldl (int, signed char *);
5256 void unsigned char vec_ldl (int, vector unsigned char *);
5257 void unsigned char vec_ldl (int, unsigned char *);
5259 vector float vec_loge (vector float);
5261 vector unsigned char vec_lvsl (int, void *, int *);
5263 vector unsigned char vec_lvsr (int, void *, int *);
5265 vector float vec_madd (vector float, vector float, vector float);
5267 vector signed short vec_madds (vector signed short, vector signed short,
5268 vector signed short);
5270 vector unsigned char vec_max (vector signed char, vector unsigned char);
5272 vector unsigned char vec_max (vector unsigned char, vector signed char);
5274 vector unsigned char vec_max (vector unsigned char,
5275 vector unsigned char);
5276 vector signed char vec_max (vector signed char, vector signed char);
5277 vector unsigned short vec_max (vector signed short,
5278 vector unsigned short);
5279 vector unsigned short vec_max (vector unsigned short,
5280 vector signed short);
5281 vector unsigned short vec_max (vector unsigned short,
5282 vector unsigned short);
5283 vector signed short vec_max (vector signed short, vector signed short);
5284 vector unsigned int vec_max (vector signed int, vector unsigned int);
5285 vector unsigned int vec_max (vector unsigned int, vector signed int);
5286 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
5287 vector signed int vec_max (vector signed int, vector signed int);
5288 vector float vec_max (vector float, vector float);
5290 vector signed char vec_mergeh (vector signed char, vector signed char);
5291 vector unsigned char vec_mergeh (vector unsigned char,
5292 vector unsigned char);
5293 vector signed short vec_mergeh (vector signed short,
5294 vector signed short);
5295 vector unsigned short vec_mergeh (vector unsigned short,
5296 vector unsigned short);
5297 vector float vec_mergeh (vector float, vector float);
5298 vector signed int vec_mergeh (vector signed int, vector signed int);
5299 vector unsigned int vec_mergeh (vector unsigned int,
5300 vector unsigned int);
5302 vector signed char vec_mergel (vector signed char, vector signed char);
5303 vector unsigned char vec_mergel (vector unsigned char,
5304 vector unsigned char);
5305 vector signed short vec_mergel (vector signed short,
5306 vector signed short);
5307 vector unsigned short vec_mergel (vector unsigned short,
5308 vector unsigned short);
5309 vector float vec_mergel (vector float, vector float);
5310 vector signed int vec_mergel (vector signed int, vector signed int);
5311 vector unsigned int vec_mergel (vector unsigned int,
5312 vector unsigned int);
5314 vector unsigned short vec_mfvscr (void);
5316 vector unsigned char vec_min (vector signed char, vector unsigned char);
5318 vector unsigned char vec_min (vector unsigned char, vector signed char);
5320 vector unsigned char vec_min (vector unsigned char,
5321 vector unsigned char);
5322 vector signed char vec_min (vector signed char, vector signed char);
5323 vector unsigned short vec_min (vector signed short,
5324 vector unsigned short);
5325 vector unsigned short vec_min (vector unsigned short,
5326 vector signed short);
5327 vector unsigned short vec_min (vector unsigned short,
5328 vector unsigned short);
5329 vector signed short vec_min (vector signed short, vector signed short);
5330 vector unsigned int vec_min (vector signed int, vector unsigned int);
5331 vector unsigned int vec_min (vector unsigned int, vector signed int);
5332 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
5333 vector signed int vec_min (vector signed int, vector signed int);
5334 vector float vec_min (vector float, vector float);
5336 vector signed short vec_mladd (vector signed short, vector signed short,
5337 vector signed short);
5338 vector signed short vec_mladd (vector signed short,
5339 vector unsigned short,
5340 vector unsigned short);
5341 vector signed short vec_mladd (vector unsigned short,
5342 vector signed short,
5343 vector signed short);
5344 vector unsigned short vec_mladd (vector unsigned short,
5345 vector unsigned short,
5346 vector unsigned short);
5348 vector signed short vec_mradds (vector signed short,
5349 vector signed short,
5350 vector signed short);
5352 vector unsigned int vec_msum (vector unsigned char,
5353 vector unsigned char,
5354 vector unsigned int);
5355 vector signed int vec_msum (vector signed char, vector unsigned char,
5357 vector unsigned int vec_msum (vector unsigned short,
5358 vector unsigned short,
5359 vector unsigned int);
5360 vector signed int vec_msum (vector signed short, vector signed short,
5363 vector unsigned int vec_msums (vector unsigned short,
5364 vector unsigned short,
5365 vector unsigned int);
5366 vector signed int vec_msums (vector signed short, vector signed short,
5369 void vec_mtvscr (vector signed int);
5370 void vec_mtvscr (vector unsigned int);
5371 void vec_mtvscr (vector signed short);
5372 void vec_mtvscr (vector unsigned short);
5373 void vec_mtvscr (vector signed char);
5374 void vec_mtvscr (vector unsigned char);
5376 vector unsigned short vec_mule (vector unsigned char,
5377 vector unsigned char);
5378 vector signed short vec_mule (vector signed char, vector signed char);
5379 vector unsigned int vec_mule (vector unsigned short,
5380 vector unsigned short);
5381 vector signed int vec_mule (vector signed short, vector signed short);
5383 vector unsigned short vec_mulo (vector unsigned char,
5384 vector unsigned char);
5385 vector signed short vec_mulo (vector signed char, vector signed char);
5386 vector unsigned int vec_mulo (vector unsigned short,
5387 vector unsigned short);
5388 vector signed int vec_mulo (vector signed short, vector signed short);
5390 vector float vec_nmsub (vector float, vector float, vector float);
5392 vector float vec_nor (vector float, vector float);
5393 vector signed int vec_nor (vector signed int, vector signed int);
5394 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
5395 vector signed short vec_nor (vector signed short, vector signed short);
5396 vector unsigned short vec_nor (vector unsigned short,
5397 vector unsigned short);
5398 vector signed char vec_nor (vector signed char, vector signed char);
5399 vector unsigned char vec_nor (vector unsigned char,
5400 vector unsigned char);
5402 vector float vec_or (vector float, vector float);
5403 vector float vec_or (vector float, vector signed int);
5404 vector float vec_or (vector signed int, vector float);
5405 vector signed int vec_or (vector signed int, vector signed int);
5406 vector unsigned int vec_or (vector signed int, vector unsigned int);
5407 vector unsigned int vec_or (vector unsigned int, vector signed int);
5408 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
5409 vector signed short vec_or (vector signed short, vector signed short);
5410 vector unsigned short vec_or (vector signed short,
5411 vector unsigned short);
5412 vector unsigned short vec_or (vector unsigned short,
5413 vector signed short);
5414 vector unsigned short vec_or (vector unsigned short,
5415 vector unsigned short);
5416 vector signed char vec_or (vector signed char, vector signed char);
5417 vector unsigned char vec_or (vector signed char, vector unsigned char);
5418 vector unsigned char vec_or (vector unsigned char, vector signed char);
5419 vector unsigned char vec_or (vector unsigned char,
5420 vector unsigned char);
5422 vector signed char vec_pack (vector signed short, vector signed short);
5423 vector unsigned char vec_pack (vector unsigned short,
5424 vector unsigned short);
5425 vector signed short vec_pack (vector signed int, vector signed int);
5426 vector unsigned short vec_pack (vector unsigned int,
5427 vector unsigned int);
5429 vector signed short vec_packpx (vector unsigned int,
5430 vector unsigned int);
5432 vector unsigned char vec_packs (vector unsigned short,
5433 vector unsigned short);
5434 vector signed char vec_packs (vector signed short, vector signed short);
5436 vector unsigned short vec_packs (vector unsigned int,
5437 vector unsigned int);
5438 vector signed short vec_packs (vector signed int, vector signed int);
5440 vector unsigned char vec_packsu (vector unsigned short,
5441 vector unsigned short);
5442 vector unsigned char vec_packsu (vector signed short,
5443 vector signed short);
5444 vector unsigned short vec_packsu (vector unsigned int,
5445 vector unsigned int);
5446 vector unsigned short vec_packsu (vector signed int, vector signed int);
5448 vector float vec_perm (vector float, vector float,
5449 vector unsigned char);
5450 vector signed int vec_perm (vector signed int, vector signed int,
5451 vector unsigned char);
5452 vector unsigned int vec_perm (vector unsigned int, vector unsigned int,
5453 vector unsigned char);
5454 vector signed short vec_perm (vector signed short, vector signed short,
5455 vector unsigned char);
5456 vector unsigned short vec_perm (vector unsigned short,
5457 vector unsigned short,
5458 vector unsigned char);
5459 vector signed char vec_perm (vector signed char, vector signed char,
5460 vector unsigned char);
5461 vector unsigned char vec_perm (vector unsigned char,
5462 vector unsigned char,
5463 vector unsigned char);
5465 vector float vec_re (vector float);
5467 vector signed char vec_rl (vector signed char, vector unsigned char);
5468 vector unsigned char vec_rl (vector unsigned char,
5469 vector unsigned char);
5470 vector signed short vec_rl (vector signed short, vector unsigned short);
5472 vector unsigned short vec_rl (vector unsigned short,
5473 vector unsigned short);
5474 vector signed int vec_rl (vector signed int, vector unsigned int);
5475 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
5477 vector float vec_round (vector float);
5479 vector float vec_rsqrte (vector float);
5481 vector float vec_sel (vector float, vector float, vector signed int);
5482 vector float vec_sel (vector float, vector float, vector unsigned int);
5483 vector signed int vec_sel (vector signed int, vector signed int,
5485 vector signed int vec_sel (vector signed int, vector signed int,
5486 vector unsigned int);
5487 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
5489 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
5490 vector unsigned int);
5491 vector signed short vec_sel (vector signed short, vector signed short,
5492 vector signed short);
5493 vector signed short vec_sel (vector signed short, vector signed short,
5494 vector unsigned short);
5495 vector unsigned short vec_sel (vector unsigned short,
5496 vector unsigned short,
5497 vector signed short);
5498 vector unsigned short vec_sel (vector unsigned short,
5499 vector unsigned short,
5500 vector unsigned short);
5501 vector signed char vec_sel (vector signed char, vector signed char,
5502 vector signed char);
5503 vector signed char vec_sel (vector signed char, vector signed char,
5504 vector unsigned char);
5505 vector unsigned char vec_sel (vector unsigned char,
5506 vector unsigned char,
5507 vector signed char);
5508 vector unsigned char vec_sel (vector unsigned char,
5509 vector unsigned char,
5510 vector unsigned char);
5512 vector signed char vec_sl (vector signed char, vector unsigned char);
5513 vector unsigned char vec_sl (vector unsigned char,
5514 vector unsigned char);
5515 vector signed short vec_sl (vector signed short, vector unsigned short);
5517 vector unsigned short vec_sl (vector unsigned short,
5518 vector unsigned short);
5519 vector signed int vec_sl (vector signed int, vector unsigned int);
5520 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
5522 vector float vec_sld (vector float, vector float, const char);
5523 vector signed int vec_sld (vector signed int, vector signed int,
5525 vector unsigned int vec_sld (vector unsigned int, vector unsigned int,
5527 vector signed short vec_sld (vector signed short, vector signed short,
5529 vector unsigned short vec_sld (vector unsigned short,
5530 vector unsigned short, const char);
5531 vector signed char vec_sld (vector signed char, vector signed char,
5533 vector unsigned char vec_sld (vector unsigned char,
5534 vector unsigned char,
5537 vector signed int vec_sll (vector signed int, vector unsigned int);
5538 vector signed int vec_sll (vector signed int, vector unsigned short);
5539 vector signed int vec_sll (vector signed int, vector unsigned char);
5540 vector unsigned int vec_sll (vector unsigned int, vector unsigned int);
5541 vector unsigned int vec_sll (vector unsigned int,
5542 vector unsigned short);
5543 vector unsigned int vec_sll (vector unsigned int, vector unsigned char);
5545 vector signed short vec_sll (vector signed short, vector unsigned int);
5546 vector signed short vec_sll (vector signed short,
5547 vector unsigned short);
5548 vector signed short vec_sll (vector signed short, vector unsigned char);
5550 vector unsigned short vec_sll (vector unsigned short,
5551 vector unsigned int);
5552 vector unsigned short vec_sll (vector unsigned short,
5553 vector unsigned short);
5554 vector unsigned short vec_sll (vector unsigned short,
5555 vector unsigned char);
5556 vector signed char vec_sll (vector signed char, vector unsigned int);
5557 vector signed char vec_sll (vector signed char, vector unsigned short);
5558 vector signed char vec_sll (vector signed char, vector unsigned char);
5559 vector unsigned char vec_sll (vector unsigned char,
5560 vector unsigned int);
5561 vector unsigned char vec_sll (vector unsigned char,
5562 vector unsigned short);
5563 vector unsigned char vec_sll (vector unsigned char,
5564 vector unsigned char);
5566 vector float vec_slo (vector float, vector signed char);
5567 vector float vec_slo (vector float, vector unsigned char);
5568 vector signed int vec_slo (vector signed int, vector signed char);
5569 vector signed int vec_slo (vector signed int, vector unsigned char);
5570 vector unsigned int vec_slo (vector unsigned int, vector signed char);
5571 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
5573 vector signed short vec_slo (vector signed short, vector signed char);
5574 vector signed short vec_slo (vector signed short, vector unsigned char);
5576 vector unsigned short vec_slo (vector unsigned short,
5577 vector signed char);
5578 vector unsigned short vec_slo (vector unsigned short,
5579 vector unsigned char);
5580 vector signed char vec_slo (vector signed char, vector signed char);
5581 vector signed char vec_slo (vector signed char, vector unsigned char);
5582 vector unsigned char vec_slo (vector unsigned char, vector signed char);
5584 vector unsigned char vec_slo (vector unsigned char,
5585 vector unsigned char);
5587 vector signed char vec_splat (vector signed char, const char);
5588 vector unsigned char vec_splat (vector unsigned char, const char);
5589 vector signed short vec_splat (vector signed short, const char);
5590 vector unsigned short vec_splat (vector unsigned short, const char);
5591 vector float vec_splat (vector float, const char);
5592 vector signed int vec_splat (vector signed int, const char);
5593 vector unsigned int vec_splat (vector unsigned int, const char);
5595 vector signed char vec_splat_s8 (const char);
5597 vector signed short vec_splat_s16 (const char);
5599 vector signed int vec_splat_s32 (const char);
5601 vector unsigned char vec_splat_u8 (const char);
5603 vector unsigned short vec_splat_u16 (const char);
5605 vector unsigned int vec_splat_u32 (const char);
5607 vector signed char vec_sr (vector signed char, vector unsigned char);
5608 vector unsigned char vec_sr (vector unsigned char,
5609 vector unsigned char);
5610 vector signed short vec_sr (vector signed short, vector unsigned short);
5612 vector unsigned short vec_sr (vector unsigned short,
5613 vector unsigned short);
5614 vector signed int vec_sr (vector signed int, vector unsigned int);
5615 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
5617 vector signed char vec_sra (vector signed char, vector unsigned char);
5618 vector unsigned char vec_sra (vector unsigned char,
5619 vector unsigned char);
5620 vector signed short vec_sra (vector signed short,
5621 vector unsigned short);
5622 vector unsigned short vec_sra (vector unsigned short,
5623 vector unsigned short);
5624 vector signed int vec_sra (vector signed int, vector unsigned int);
5625 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
5627 vector signed int vec_srl (vector signed int, vector unsigned int);
5628 vector signed int vec_srl (vector signed int, vector unsigned short);
5629 vector signed int vec_srl (vector signed int, vector unsigned char);
5630 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
5631 vector unsigned int vec_srl (vector unsigned int,
5632 vector unsigned short);
5633 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
5635 vector signed short vec_srl (vector signed short, vector unsigned int);
5636 vector signed short vec_srl (vector signed short,
5637 vector unsigned short);
5638 vector signed short vec_srl (vector signed short, vector unsigned char);
5640 vector unsigned short vec_srl (vector unsigned short,
5641 vector unsigned int);
5642 vector unsigned short vec_srl (vector unsigned short,
5643 vector unsigned short);
5644 vector unsigned short vec_srl (vector unsigned short,
5645 vector unsigned char);
5646 vector signed char vec_srl (vector signed char, vector unsigned int);
5647 vector signed char vec_srl (vector signed char, vector unsigned short);
5648 vector signed char vec_srl (vector signed char, vector unsigned char);
5649 vector unsigned char vec_srl (vector unsigned char,
5650 vector unsigned int);
5651 vector unsigned char vec_srl (vector unsigned char,
5652 vector unsigned short);
5653 vector unsigned char vec_srl (vector unsigned char,
5654 vector unsigned char);
5656 vector float vec_sro (vector float, vector signed char);
5657 vector float vec_sro (vector float, vector unsigned char);
5658 vector signed int vec_sro (vector signed int, vector signed char);
5659 vector signed int vec_sro (vector signed int, vector unsigned char);
5660 vector unsigned int vec_sro (vector unsigned int, vector signed char);
5661 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
5663 vector signed short vec_sro (vector signed short, vector signed char);
5664 vector signed short vec_sro (vector signed short, vector unsigned char);
5666 vector unsigned short vec_sro (vector unsigned short,
5667 vector signed char);
5668 vector unsigned short vec_sro (vector unsigned short,
5669 vector unsigned char);
5670 vector signed char vec_sro (vector signed char, vector signed char);
5671 vector signed char vec_sro (vector signed char, vector unsigned char);
5672 vector unsigned char vec_sro (vector unsigned char, vector signed char);
5674 vector unsigned char vec_sro (vector unsigned char,
5675 vector unsigned char);
5677 void vec_st (vector float, int, float *);
5678 void vec_st (vector float, int, vector float *);
5679 void vec_st (vector signed int, int, int *);
5680 void vec_st (vector signed int, int, unsigned int *);
5681 void vec_st (vector unsigned int, int, unsigned int *);
5682 void vec_st (vector unsigned int, int, vector unsigned int *);
5683 void vec_st (vector signed short, int, short *);
5684 void vec_st (vector signed short, int, vector unsigned short *);
5685 void vec_st (vector signed short, int, vector signed short *);
5686 void vec_st (vector unsigned short, int, unsigned short *);
5687 void vec_st (vector unsigned short, int, vector unsigned short *);
5688 void vec_st (vector signed char, int, signed char *);
5689 void vec_st (vector signed char, int, unsigned char *);
5690 void vec_st (vector signed char, int, vector signed char *);
5691 void vec_st (vector unsigned char, int, unsigned char *);
5692 void vec_st (vector unsigned char, int, vector unsigned char *);
5694 void vec_ste (vector signed char, int, unsigned char *);
5695 void vec_ste (vector signed char, int, signed char *);
5696 void vec_ste (vector unsigned char, int, unsigned char *);
5697 void vec_ste (vector signed short, int, short *);
5698 void vec_ste (vector signed short, int, unsigned short *);
5699 void vec_ste (vector unsigned short, int, void *);
5700 void vec_ste (vector signed int, int, unsigned int *);
5701 void vec_ste (vector signed int, int, int *);
5702 void vec_ste (vector unsigned int, int, unsigned int *);
5703 void vec_ste (vector float, int, float *);
5705 void vec_stl (vector float, int, vector float *);
5706 void vec_stl (vector float, int, float *);
5707 void vec_stl (vector signed int, int, vector signed int *);
5708 void vec_stl (vector signed int, int, int *);
5709 void vec_stl (vector signed int, int, unsigned int *);
5710 void vec_stl (vector unsigned int, int, vector unsigned int *);
5711 void vec_stl (vector unsigned int, int, unsigned int *);
5712 void vec_stl (vector signed short, int, short *);
5713 void vec_stl (vector signed short, int, unsigned short *);
5714 void vec_stl (vector signed short, int, vector signed short *);
5715 void vec_stl (vector unsigned short, int, unsigned short *);
5716 void vec_stl (vector unsigned short, int, vector signed short *);
5717 void vec_stl (vector signed char, int, signed char *);
5718 void vec_stl (vector signed char, int, unsigned char *);
5719 void vec_stl (vector signed char, int, vector signed char *);
5720 void vec_stl (vector unsigned char, int, unsigned char *);
5721 void vec_stl (vector unsigned char, int, vector unsigned char *);
5723 vector signed char vec_sub (vector signed char, vector signed char);
5724 vector unsigned char vec_sub (vector signed char, vector unsigned char);
5726 vector unsigned char vec_sub (vector unsigned char, vector signed char);
5728 vector unsigned char vec_sub (vector unsigned char,
5729 vector unsigned char);
5730 vector signed short vec_sub (vector signed short, vector signed short);
5731 vector unsigned short vec_sub (vector signed short,
5732 vector unsigned short);
5733 vector unsigned short vec_sub (vector unsigned short,
5734 vector signed short);
5735 vector unsigned short vec_sub (vector unsigned short,
5736 vector unsigned short);
5737 vector signed int vec_sub (vector signed int, vector signed int);
5738 vector unsigned int vec_sub (vector signed int, vector unsigned int);
5739 vector unsigned int vec_sub (vector unsigned int, vector signed int);
5740 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
5741 vector float vec_sub (vector float, vector float);
5743 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
5745 vector unsigned char vec_subs (vector signed char,
5746 vector unsigned char);
5747 vector unsigned char vec_subs (vector unsigned char,
5748 vector signed char);
5749 vector unsigned char vec_subs (vector unsigned char,
5750 vector unsigned char);
5751 vector signed char vec_subs (vector signed char, vector signed char);
5752 vector unsigned short vec_subs (vector signed short,
5753 vector unsigned short);
5754 vector unsigned short vec_subs (vector unsigned short,
5755 vector signed short);
5756 vector unsigned short vec_subs (vector unsigned short,
5757 vector unsigned short);
5758 vector signed short vec_subs (vector signed short, vector signed short);
5760 vector unsigned int vec_subs (vector signed int, vector unsigned int);
5761 vector unsigned int vec_subs (vector unsigned int, vector signed int);
5762 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
5764 vector signed int vec_subs (vector signed int, vector signed int);
5766 vector unsigned int vec_sum4s (vector unsigned char,
5767 vector unsigned int);
5768 vector signed int vec_sum4s (vector signed char, vector signed int);
5769 vector signed int vec_sum4s (vector signed short, vector signed int);
5771 vector signed int vec_sum2s (vector signed int, vector signed int);
5773 vector signed int vec_sums (vector signed int, vector signed int);
5775 vector float vec_trunc (vector float);
5777 vector signed short vec_unpackh (vector signed char);
5778 vector unsigned int vec_unpackh (vector signed short);
5779 vector signed int vec_unpackh (vector signed short);
5781 vector signed short vec_unpackl (vector signed char);
5782 vector unsigned int vec_unpackl (vector signed short);
5783 vector signed int vec_unpackl (vector signed short);
5785 vector float vec_xor (vector float, vector float);
5786 vector float vec_xor (vector float, vector signed int);
5787 vector float vec_xor (vector signed int, vector float);
5788 vector signed int vec_xor (vector signed int, vector signed int);
5789 vector unsigned int vec_xor (vector signed int, vector unsigned int);
5790 vector unsigned int vec_xor (vector unsigned int, vector signed int);
5791 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
5792 vector signed short vec_xor (vector signed short, vector signed short);
5793 vector unsigned short vec_xor (vector signed short,
5794 vector unsigned short);
5795 vector unsigned short vec_xor (vector unsigned short,
5796 vector signed short);
5797 vector unsigned short vec_xor (vector unsigned short,
5798 vector unsigned short);
5799 vector signed char vec_xor (vector signed char, vector signed char);
5800 vector unsigned char vec_xor (vector signed char, vector unsigned char);
5802 vector unsigned char vec_xor (vector unsigned char, vector signed char);
5804 vector unsigned char vec_xor (vector unsigned char,
5805 vector unsigned char);
5807 vector signed int vec_all_eq (vector signed char, vector unsigned char);
5809 vector signed int vec_all_eq (vector signed char, vector signed char);
5810 vector signed int vec_all_eq (vector unsigned char, vector signed char);
5812 vector signed int vec_all_eq (vector unsigned char,
5813 vector unsigned char);
5814 vector signed int vec_all_eq (vector signed short,
5815 vector unsigned short);
5816 vector signed int vec_all_eq (vector signed short, vector signed short);
5818 vector signed int vec_all_eq (vector unsigned short,
5819 vector signed short);
5820 vector signed int vec_all_eq (vector unsigned short,
5821 vector unsigned short);
5822 vector signed int vec_all_eq (vector signed int, vector unsigned int);
5823 vector signed int vec_all_eq (vector signed int, vector signed int);
5824 vector signed int vec_all_eq (vector unsigned int, vector signed int);
5825 vector signed int vec_all_eq (vector unsigned int, vector unsigned int);
5827 vector signed int vec_all_eq (vector float, vector float);
5829 vector signed int vec_all_ge (vector signed char, vector unsigned char);
5831 vector signed int vec_all_ge (vector unsigned char, vector signed char);
5833 vector signed int vec_all_ge (vector unsigned char,
5834 vector unsigned char);
5835 vector signed int vec_all_ge (vector signed char, vector signed char);
5836 vector signed int vec_all_ge (vector signed short,
5837 vector unsigned short);
5838 vector signed int vec_all_ge (vector unsigned short,
5839 vector signed short);
5840 vector signed int vec_all_ge (vector unsigned short,
5841 vector unsigned short);
5842 vector signed int vec_all_ge (vector signed short, vector signed short);
5844 vector signed int vec_all_ge (vector signed int, vector unsigned int);
5845 vector signed int vec_all_ge (vector unsigned int, vector signed int);
5846 vector signed int vec_all_ge (vector unsigned int, vector unsigned int);
5848 vector signed int vec_all_ge (vector signed int, vector signed int);
5849 vector signed int vec_all_ge (vector float, vector float);
5851 vector signed int vec_all_gt (vector signed char, vector unsigned char);
5853 vector signed int vec_all_gt (vector unsigned char, vector signed char);
5855 vector signed int vec_all_gt (vector unsigned char,
5856 vector unsigned char);
5857 vector signed int vec_all_gt (vector signed char, vector signed char);
5858 vector signed int vec_all_gt (vector signed short,
5859 vector unsigned short);
5860 vector signed int vec_all_gt (vector unsigned short,
5861 vector signed short);
5862 vector signed int vec_all_gt (vector unsigned short,
5863 vector unsigned short);
5864 vector signed int vec_all_gt (vector signed short, vector signed short);
5866 vector signed int vec_all_gt (vector signed int, vector unsigned int);
5867 vector signed int vec_all_gt (vector unsigned int, vector signed int);
5868 vector signed int vec_all_gt (vector unsigned int, vector unsigned int);
5870 vector signed int vec_all_gt (vector signed int, vector signed int);
5871 vector signed int vec_all_gt (vector float, vector float);
5873 vector signed int vec_all_in (vector float, vector float);
5875 vector signed int vec_all_le (vector signed char, vector unsigned char);
5877 vector signed int vec_all_le (vector unsigned char, vector signed char);
5879 vector signed int vec_all_le (vector unsigned char,
5880 vector unsigned char);
5881 vector signed int vec_all_le (vector signed char, vector signed char);
5882 vector signed int vec_all_le (vector signed short,
5883 vector unsigned short);
5884 vector signed int vec_all_le (vector unsigned short,
5885 vector signed short);
5886 vector signed int vec_all_le (vector unsigned short,
5887 vector unsigned short);
5888 vector signed int vec_all_le (vector signed short, vector signed short);
5890 vector signed int vec_all_le (vector signed int, vector unsigned int);
5891 vector signed int vec_all_le (vector unsigned int, vector signed int);
5892 vector signed int vec_all_le (vector unsigned int, vector unsigned int);
5894 vector signed int vec_all_le (vector signed int, vector signed int);
5895 vector signed int vec_all_le (vector float, vector float);
5897 vector signed int vec_all_lt (vector signed char, vector unsigned char);
5899 vector signed int vec_all_lt (vector unsigned char, vector signed char);
5901 vector signed int vec_all_lt (vector unsigned char,
5902 vector unsigned char);
5903 vector signed int vec_all_lt (vector signed char, vector signed char);
5904 vector signed int vec_all_lt (vector signed short,
5905 vector unsigned short);
5906 vector signed int vec_all_lt (vector unsigned short,
5907 vector signed short);
5908 vector signed int vec_all_lt (vector unsigned short,
5909 vector unsigned short);
5910 vector signed int vec_all_lt (vector signed short, vector signed short);
5912 vector signed int vec_all_lt (vector signed int, vector unsigned int);
5913 vector signed int vec_all_lt (vector unsigned int, vector signed int);
5914 vector signed int vec_all_lt (vector unsigned int, vector unsigned int);
5916 vector signed int vec_all_lt (vector signed int, vector signed int);
5917 vector signed int vec_all_lt (vector float, vector float);
5919 vector signed int vec_all_nan (vector float);
5921 vector signed int vec_all_ne (vector signed char, vector unsigned char);
5923 vector signed int vec_all_ne (vector signed char, vector signed char);
5924 vector signed int vec_all_ne (vector unsigned char, vector signed char);
5926 vector signed int vec_all_ne (vector unsigned char,
5927 vector unsigned char);
5928 vector signed int vec_all_ne (vector signed short,
5929 vector unsigned short);
5930 vector signed int vec_all_ne (vector signed short, vector signed short);
5932 vector signed int vec_all_ne (vector unsigned short,
5933 vector signed short);
5934 vector signed int vec_all_ne (vector unsigned short,
5935 vector unsigned short);
5936 vector signed int vec_all_ne (vector signed int, vector unsigned int);
5937 vector signed int vec_all_ne (vector signed int, vector signed int);
5938 vector signed int vec_all_ne (vector unsigned int, vector signed int);
5939 vector signed int vec_all_ne (vector unsigned int, vector unsigned int);
5941 vector signed int vec_all_ne (vector float, vector float);
5943 vector signed int vec_all_nge (vector float, vector float);
5945 vector signed int vec_all_ngt (vector float, vector float);
5947 vector signed int vec_all_nle (vector float, vector float);
5949 vector signed int vec_all_nlt (vector float, vector float);
5951 vector signed int vec_all_numeric (vector float);
5953 vector signed int vec_any_eq (vector signed char, vector unsigned char);
5955 vector signed int vec_any_eq (vector signed char, vector signed char);
5956 vector signed int vec_any_eq (vector unsigned char, vector signed char);
5958 vector signed int vec_any_eq (vector unsigned char,
5959 vector unsigned char);
5960 vector signed int vec_any_eq (vector signed short,
5961 vector unsigned short);
5962 vector signed int vec_any_eq (vector signed short, vector signed short);
5964 vector signed int vec_any_eq (vector unsigned short,
5965 vector signed short);
5966 vector signed int vec_any_eq (vector unsigned short,
5967 vector unsigned short);
5968 vector signed int vec_any_eq (vector signed int, vector unsigned int);
5969 vector signed int vec_any_eq (vector signed int, vector signed int);
5970 vector signed int vec_any_eq (vector unsigned int, vector signed int);
5971 vector signed int vec_any_eq (vector unsigned int, vector unsigned int);
5973 vector signed int vec_any_eq (vector float, vector float);
5975 vector signed int vec_any_ge (vector signed char, vector unsigned char);
5977 vector signed int vec_any_ge (vector unsigned char, vector signed char);
5979 vector signed int vec_any_ge (vector unsigned char,
5980 vector unsigned char);
5981 vector signed int vec_any_ge (vector signed char, vector signed char);
5982 vector signed int vec_any_ge (vector signed short,
5983 vector unsigned short);
5984 vector signed int vec_any_ge (vector unsigned short,
5985 vector signed short);
5986 vector signed int vec_any_ge (vector unsigned short,
5987 vector unsigned short);
5988 vector signed int vec_any_ge (vector signed short, vector signed short);
5990 vector signed int vec_any_ge (vector signed int, vector unsigned int);
5991 vector signed int vec_any_ge (vector unsigned int, vector signed int);
5992 vector signed int vec_any_ge (vector unsigned int, vector unsigned int);
5994 vector signed int vec_any_ge (vector signed int, vector signed int);
5995 vector signed int vec_any_ge (vector float, vector float);
5997 vector signed int vec_any_gt (vector signed char, vector unsigned char);
5999 vector signed int vec_any_gt (vector unsigned char, vector signed char);
6001 vector signed int vec_any_gt (vector unsigned char,
6002 vector unsigned char);
6003 vector signed int vec_any_gt (vector signed char, vector signed char);
6004 vector signed int vec_any_gt (vector signed short,
6005 vector unsigned short);
6006 vector signed int vec_any_gt (vector unsigned short,
6007 vector signed short);
6008 vector signed int vec_any_gt (vector unsigned short,
6009 vector unsigned short);
6010 vector signed int vec_any_gt (vector signed short, vector signed short);
6012 vector signed int vec_any_gt (vector signed int, vector unsigned int);
6013 vector signed int vec_any_gt (vector unsigned int, vector signed int);
6014 vector signed int vec_any_gt (vector unsigned int, vector unsigned int);
6016 vector signed int vec_any_gt (vector signed int, vector signed int);
6017 vector signed int vec_any_gt (vector float, vector float);
6019 vector signed int vec_any_le (vector signed char, vector unsigned char);
6021 vector signed int vec_any_le (vector unsigned char, vector signed char);
6023 vector signed int vec_any_le (vector unsigned char,
6024 vector unsigned char);
6025 vector signed int vec_any_le (vector signed char, vector signed char);
6026 vector signed int vec_any_le (vector signed short,
6027 vector unsigned short);
6028 vector signed int vec_any_le (vector unsigned short,
6029 vector signed short);
6030 vector signed int vec_any_le (vector unsigned short,
6031 vector unsigned short);
6032 vector signed int vec_any_le (vector signed short, vector signed short);
6034 vector signed int vec_any_le (vector signed int, vector unsigned int);
6035 vector signed int vec_any_le (vector unsigned int, vector signed int);
6036 vector signed int vec_any_le (vector unsigned int, vector unsigned int);
6038 vector signed int vec_any_le (vector signed int, vector signed int);
6039 vector signed int vec_any_le (vector float, vector float);
6041 vector signed int vec_any_lt (vector signed char, vector unsigned char);
6043 vector signed int vec_any_lt (vector unsigned char, vector signed char);
6045 vector signed int vec_any_lt (vector unsigned char,
6046 vector unsigned char);
6047 vector signed int vec_any_lt (vector signed char, vector signed char);
6048 vector signed int vec_any_lt (vector signed short,
6049 vector unsigned short);
6050 vector signed int vec_any_lt (vector unsigned short,
6051 vector signed short);
6052 vector signed int vec_any_lt (vector unsigned short,
6053 vector unsigned short);
6054 vector signed int vec_any_lt (vector signed short, vector signed short);
6056 vector signed int vec_any_lt (vector signed int, vector unsigned int);
6057 vector signed int vec_any_lt (vector unsigned int, vector signed int);
6058 vector signed int vec_any_lt (vector unsigned int, vector unsigned int);
6060 vector signed int vec_any_lt (vector signed int, vector signed int);
6061 vector signed int vec_any_lt (vector float, vector float);
6063 vector signed int vec_any_nan (vector float);
6065 vector signed int vec_any_ne (vector signed char, vector unsigned char);
6067 vector signed int vec_any_ne (vector signed char, vector signed char);
6068 vector signed int vec_any_ne (vector unsigned char, vector signed char);
6070 vector signed int vec_any_ne (vector unsigned char,
6071 vector unsigned char);
6072 vector signed int vec_any_ne (vector signed short,
6073 vector unsigned short);
6074 vector signed int vec_any_ne (vector signed short, vector signed short);
6076 vector signed int vec_any_ne (vector unsigned short,
6077 vector signed short);
6078 vector signed int vec_any_ne (vector unsigned short,
6079 vector unsigned short);
6080 vector signed int vec_any_ne (vector signed int, vector unsigned int);
6081 vector signed int vec_any_ne (vector signed int, vector signed int);
6082 vector signed int vec_any_ne (vector unsigned int, vector signed int);
6083 vector signed int vec_any_ne (vector unsigned int, vector unsigned int);
6085 vector signed int vec_any_ne (vector float, vector float);
6087 vector signed int vec_any_nge (vector float, vector float);
6089 vector signed int vec_any_ngt (vector float, vector float);
6091 vector signed int vec_any_nle (vector float, vector float);
6093 vector signed int vec_any_nlt (vector float, vector float);
6095 vector signed int vec_any_numeric (vector float);
6097 vector signed int vec_any_out (vector float, vector float);
6101 @section Pragmas Accepted by GCC
6105 GCC supports several types of pragmas, primarily in order to compile
6106 code originally written for other compilers. Note that in general
6107 we do not recommend the use of pragmas; @xref{Function Attributes},
6108 for further explanation.
6112 * RS/6000 and PowerPC Pragmas::
6119 @subsection ARM Pragmas
6121 The ARM target defines pragmas for controlling the default addition of
6122 @code{long_call} and @code{short_call} attributes to functions.
6123 @xref{Function Attributes}, for information about the effects of these
6128 @cindex pragma, long_calls
6129 Set all subsequent functions to have the @code{long_call} attribute.
6132 @cindex pragma, no_long_calls
6133 Set all subsequent functions to have the @code{short_call} attribute.
6135 @item long_calls_off
6136 @cindex pragma, long_calls_off
6137 Do not affect the @code{long_call} or @code{short_call} attributes of
6138 subsequent functions.
6141 @node RS/6000 and PowerPC Pragmas
6142 @subsection RS/6000 and PowerPC Pragmas
6144 The RS/6000 and PowerPC targets define one pragma for controlling
6145 whether or not the @code{longcall} attribute is added to function
6146 declarations by default. This pragma overrides the @option{-mlongcall}
6147 option, but not the @code{longcall} and @code{shortcall} attributes.
6148 @xref{RS/6000 and PowerPC Options}, for more information about when long
6149 calls are and are not necessary.
6153 @cindex pragma, longcall
6154 Apply the @code{longcall} attribute to all subsequent function
6158 Do not apply the @code{longcall} attribute to subsequent function
6162 @c Describe c4x pragmas here.
6163 @c Describe h8300 pragmas here.
6164 @c Describe i370 pragmas here.
6165 @c Describe i960 pragmas here.
6166 @c Describe sh pragmas here.
6167 @c Describe v850 pragmas here.
6169 @node Darwin Pragmas
6170 @subsection Darwin Pragmas
6172 The following pragmas are available for all architectures running the
6173 Darwin operating system. These are useful for compatibility with other
6177 @item mark @var{tokens}@dots{}
6178 @cindex pragma, mark
6179 This pragma is accepted, but has no effect.
6181 @item options align=@var{alignment}
6182 @cindex pragma, options align
6183 This pragma sets the alignment of fields in structures. The values of
6184 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
6185 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
6186 properly; to restore the previous setting, use @code{reset} for the
6189 @item segment @var{tokens}@dots{}
6190 @cindex pragma, segment
6191 This pragma is accepted, but has no effect.
6193 @item unused (@var{var} [, @var{var}]@dots{})
6194 @cindex pragma, unused
6195 This pragma declares variables to be possibly unused. GCC will not
6196 produce warnings for the listed variables. The effect is similar to
6197 that of the @code{unused} attribute, except that this pragma may appear
6198 anywhere within the variables' scopes.
6201 @node Solaris Pragmas
6202 @subsection Solaris Pragmas
6204 For compatibility with the SunPRO compiler, the following pragma
6208 @item redefine_extname @var{oldname} @var{newname}
6209 @cindex pragma, redefine_extname
6211 This pragma gives the C function @var{oldname} the assembler label
6212 @var{newname}. The pragma must appear before the function declaration.
6213 This pragma is equivalent to the asm labels extension (@pxref{Asm
6214 Labels}). The preprocessor defines @code{__PRAGMA_REDEFINE_EXTNAME}
6215 if the pragma is available.
6219 @subsection Tru64 Pragmas
6221 For compatibility with the Compaq C compiler, the following pragma
6225 @item extern_prefix @var{string}
6226 @cindex pragma, extern_prefix
6228 This pragma renames all subsequent function and variable declarations
6229 such that @var{string} is prepended to the name. This effect may be
6230 terminated by using another @code{extern_prefix} pragma with the
6233 This pragma is similar in intent to to the asm labels extension
6234 (@pxref{Asm Labels}) in that the system programmer wants to change
6235 the assembly-level ABI without changing the source-level API. The
6236 preprocessor defines @code{__EXTERN_PREFIX} if the pragma is available.
6239 @node Unnamed Fields
6240 @section Unnamed struct/union fields within structs/unions.
6244 For compatibility with other compilers, GCC allows you to define
6245 a structure or union that contains, as fields, structures and unions
6246 without names. For example:
6259 In this example, the user would be able to access members of the unnamed
6260 union with code like @samp{foo.b}. Note that only unnamed structs and
6261 unions are allowed, you may not have, for example, an unnamed
6264 You must never create such structures that cause ambiguous field definitions.
6265 For example, this structure:
6276 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
6277 Such constructs are not supported and must be avoided. In the future,
6278 such constructs may be detected and treated as compilation errors.
6281 @section Thread-Local Storage
6282 @cindex Thread-Local Storage
6283 @cindex @acronym{TLS}
6286 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
6287 are allocated such that there is one instance of the variable per extant
6288 thread. The run-time model GCC uses to implement this originates
6289 in the IA-64 processor-specific ABI, but has since been migrated
6290 to other processors as well. It requires significant support from
6291 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
6292 system libraries (@file{libc.so} and @file{libpthread.so}), so it
6293 is not available everywhere.
6295 At the user level, the extension is visible with a new storage
6296 class keyword: @code{__thread}. For example:
6300 extern __thread struct state s;
6301 static __thread char *p;
6304 The @code{__thread} specifier may be used alone, with the @code{extern}
6305 or @code{static} specifiers, but with no other storage class specifier.
6306 When used with @code{extern} or @code{static}, @code{__thread} must appear
6307 immediately after the other storage class specifier.
6309 The @code{__thread} specifier may be applied to any global, file-scoped
6310 static, function-scoped static, or static data member of a class. It may
6311 not be applied to block-scoped automatic or non-static data member.
6313 When the address-of operator is applied to a thread-local variable, it is
6314 evaluated at run-time and returns the address of the current thread's
6315 instance of that variable. An address so obtained may be used by any
6316 thread. When a thread terminates, any pointers to thread-local variables
6317 in that thread become invalid.
6319 No static initialization may refer to the address of a thread-local variable.
6321 In C++, if an initializer is present for a thread-local variable, it must
6322 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
6325 See @uref{http://people.redhat.com/drepper/tls.pdf,
6326 ELF Handling For Thread-Local Storage} for a detailed explanation of
6327 the four thread-local storage addressing models, and how the run-time
6328 is expected to function.
6331 * C99 Thread-Local Edits::
6332 * C++98 Thread-Local Edits::
6335 @node C99 Thread-Local Edits
6336 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
6338 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
6339 that document the exact semantics of the language extension.
6343 @cite{5.1.2 Execution environments}
6345 Add new text after paragraph 1
6348 Within either execution environment, a @dfn{thread} is a flow of
6349 control within a program. It is implementation defined whether
6350 or not there may be more than one thread associated with a program.
6351 It is implementation defined how threads beyond the first are
6352 created, the name and type of the function called at thread
6353 startup, and how threads may be terminated. However, objects
6354 with thread storage duration shall be initialized before thread
6359 @cite{6.2.4 Storage durations of objects}
6361 Add new text before paragraph 3
6364 An object whose identifier is declared with the storage-class
6365 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
6366 Its lifetime is the entire execution of the thread, and its
6367 stored value is initialized only once, prior to thread startup.
6371 @cite{6.4.1 Keywords}
6373 Add @code{__thread}.
6376 @cite{6.7.1 Storage-class specifiers}
6378 Add @code{__thread} to the list of storage class specifiers in
6381 Change paragraph 2 to
6384 With the exception of @code{__thread}, at most one storage-class
6385 specifier may be given [@dots{}]. The @code{__thread} specifier may
6386 be used alone, or immediately following @code{extern} or
6390 Add new text after paragraph 6
6393 The declaration of an identifier for a variable that has
6394 block scope that specifies @code{__thread} shall also
6395 specify either @code{extern} or @code{static}.
6397 The @code{__thread} specifier shall be used only with
6402 @node C++98 Thread-Local Edits
6403 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
6405 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
6406 that document the exact semantics of the language extension.
6409 @b{[intro.execution]}
6411 New text after paragraph 4
6414 A @dfn{thread} is a flow of control within the abstract machine.
6415 It is implementation defined whether or not there may be more than
6419 New text after paragraph 7
6422 It is unspecified whether additional action must be taken to
6423 ensure when and whether side effects are visible to other threads.
6429 Add @code{__thread}.
6432 @b{[basic.start.main]}
6434 Add after paragraph 5
6437 The thread that begins execution at the @code{main} function is called
6438 the @dfn{main thread}. It is implementation defined how functions
6439 beginning threads other than the main thread are designated or typed.
6440 A function so designated, as well as the @code{main} function, is called
6441 a @dfn{thread startup function}. It is implementation defined what
6442 happens if a thread startup function returns. It is implementation
6443 defined what happens to other threads when any thread calls @code{exit}.
6447 @b{[basic.start.init]}
6449 Add after paragraph 4
6452 The storage for an object of thread storage duration shall be
6453 staticly initialized before the first statement of the thread startup
6454 function. An object of thread storage duration shall not require
6455 dynamic initialization.
6459 @b{[basic.start.term]}
6461 Add after paragraph 3
6464 The type of an object with thread storage duration shall not have a
6465 non-trivial destructor, nor shall it be an array type whose elements
6466 (directly or indirectly) have non-trivial destructors.
6472 Add ``thread storage duration'' to the list in paragraph 1.
6477 Thread, static, and automatic storage durations are associated with
6478 objects introduced by declarations [@dots{}].
6481 Add @code{__thread} to the list of specifiers in paragraph 3.
6484 @b{[basic.stc.thread]}
6486 New section before @b{[basic.stc.static]}
6489 The keyword @code{__thread} applied to an non-local object gives the
6490 object thread storage duration.
6492 A local variable or class data member declared both @code{static}
6493 and @code{__thread} gives the variable or member thread storage
6498 @b{[basic.stc.static]}
6503 All objects which have neither thread storage duration, dynamic
6504 storage duration nor are local [@dots{}].
6510 Add @code{__thread} to the list in paragraph 1.
6515 With the exception of @code{__thread}, at most one
6516 @var{storage-class-specifier} shall appear in a given
6517 @var{decl-specifier-seq}. The @code{__thread} specifier may
6518 be used alone, or immediately following the @code{extern} or
6519 @code{static} specifiers. [@dots{}]
6522 Add after paragraph 5
6525 The @code{__thread} specifier can be applied only to the names of objects
6526 and to anonymous unions.
6532 Add after paragraph 6
6535 Non-@code{static} members shall not be @code{__thread}.
6539 @node C++ Extensions
6540 @chapter Extensions to the C++ Language
6541 @cindex extensions, C++ language
6542 @cindex C++ language extensions
6544 The GNU compiler provides these extensions to the C++ language (and you
6545 can also use most of the C language extensions in your C++ programs). If you
6546 want to write code that checks whether these features are available, you can
6547 test for the GNU compiler the same way as for C programs: check for a
6548 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
6549 test specifically for GNU C++ (@pxref{Standard Predefined,,Standard
6550 Predefined Macros,cpp.info,The C Preprocessor}).
6553 * Min and Max:: C++ Minimum and maximum operators.
6554 * Volatiles:: What constitutes an access to a volatile object.
6555 * Restricted Pointers:: C99 restricted pointers and references.
6556 * Vague Linkage:: Where G++ puts inlines, vtables and such.
6557 * C++ Interface:: You can use a single C++ header file for both
6558 declarations and definitions.
6559 * Template Instantiation:: Methods for ensuring that exactly one copy of
6560 each needed template instantiation is emitted.
6561 * Bound member functions:: You can extract a function pointer to the
6562 method denoted by a @samp{->*} or @samp{.*} expression.
6563 * C++ Attributes:: Variable, function, and type attributes for C++ only.
6564 * Java Exceptions:: Tweaking exception handling to work with Java.
6565 * Deprecated Features:: Things might disappear from g++.
6566 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
6570 @section Minimum and Maximum Operators in C++
6572 It is very convenient to have operators which return the ``minimum'' or the
6573 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
6576 @item @var{a} <? @var{b}
6578 @cindex minimum operator
6579 is the @dfn{minimum}, returning the smaller of the numeric values
6580 @var{a} and @var{b};
6582 @item @var{a} >? @var{b}
6584 @cindex maximum operator
6585 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
6589 These operations are not primitive in ordinary C++, since you can
6590 use a macro to return the minimum of two things in C++, as in the
6594 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
6598 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
6599 the minimum value of variables @var{i} and @var{j}.
6601 However, side effects in @code{X} or @code{Y} may cause unintended
6602 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
6603 the smaller counter twice. A GNU C extension allows you to write safe
6604 macros that avoid this kind of problem (@pxref{Naming Types,,Naming an
6605 Expression's Type}). However, writing @code{MIN} and @code{MAX} as
6606 macros also forces you to use function-call notation for a
6607 fundamental arithmetic operation. Using GNU C++ extensions, you can
6608 write @w{@samp{int min = i <? j;}} instead.
6610 Since @code{<?} and @code{>?} are built into the compiler, they properly
6611 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
6615 @section When is a Volatile Object Accessed?
6616 @cindex accessing volatiles
6617 @cindex volatile read
6618 @cindex volatile write
6619 @cindex volatile access
6621 Both the C and C++ standard have the concept of volatile objects. These
6622 are normally accessed by pointers and used for accessing hardware. The
6623 standards encourage compilers to refrain from optimizations
6624 concerning accesses to volatile objects that it might perform on
6625 non-volatile objects. The C standard leaves it implementation defined
6626 as to what constitutes a volatile access. The C++ standard omits to
6627 specify this, except to say that C++ should behave in a similar manner
6628 to C with respect to volatiles, where possible. The minimum either
6629 standard specifies is that at a sequence point all previous accesses to
6630 volatile objects have stabilized and no subsequent accesses have
6631 occurred. Thus an implementation is free to reorder and combine
6632 volatile accesses which occur between sequence points, but cannot do so
6633 for accesses across a sequence point. The use of volatiles does not
6634 allow you to violate the restriction on updating objects multiple times
6635 within a sequence point.
6637 In most expressions, it is intuitively obvious what is a read and what is
6638 a write. For instance
6641 volatile int *dst = @var{somevalue};
6642 volatile int *src = @var{someothervalue};
6647 will cause a read of the volatile object pointed to by @var{src} and stores the
6648 value into the volatile object pointed to by @var{dst}. There is no
6649 guarantee that these reads and writes are atomic, especially for objects
6650 larger than @code{int}.
6652 Less obvious expressions are where something which looks like an access
6653 is used in a void context. An example would be,
6656 volatile int *src = @var{somevalue};
6660 With C, such expressions are rvalues, and as rvalues cause a read of
6661 the object, GCC interprets this as a read of the volatile being pointed
6662 to. The C++ standard specifies that such expressions do not undergo
6663 lvalue to rvalue conversion, and that the type of the dereferenced
6664 object may be incomplete. The C++ standard does not specify explicitly
6665 that it is this lvalue to rvalue conversion which is responsible for
6666 causing an access. However, there is reason to believe that it is,
6667 because otherwise certain simple expressions become undefined. However,
6668 because it would surprise most programmers, G++ treats dereferencing a
6669 pointer to volatile object of complete type in a void context as a read
6670 of the object. When the object has incomplete type, G++ issues a
6675 struct T @{int m;@};
6676 volatile S *ptr1 = @var{somevalue};
6677 volatile T *ptr2 = @var{somevalue};
6682 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
6683 causes a read of the object pointed to. If you wish to force an error on
6684 the first case, you must force a conversion to rvalue with, for instance
6685 a static cast, @code{static_cast<S>(*ptr1)}.
6687 When using a reference to volatile, G++ does not treat equivalent
6688 expressions as accesses to volatiles, but instead issues a warning that
6689 no volatile is accessed. The rationale for this is that otherwise it
6690 becomes difficult to determine where volatile access occur, and not
6691 possible to ignore the return value from functions returning volatile
6692 references. Again, if you wish to force a read, cast the reference to
6695 @node Restricted Pointers
6696 @section Restricting Pointer Aliasing
6697 @cindex restricted pointers
6698 @cindex restricted references
6699 @cindex restricted this pointer
6701 As with gcc, g++ understands the C99 feature of restricted pointers,
6702 specified with the @code{__restrict__}, or @code{__restrict} type
6703 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
6704 language flag, @code{restrict} is not a keyword in C++.
6706 In addition to allowing restricted pointers, you can specify restricted
6707 references, which indicate that the reference is not aliased in the local
6711 void fn (int *__restrict__ rptr, int &__restrict__ rref)
6718 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
6719 @var{rref} refers to a (different) unaliased integer.
6721 You may also specify whether a member function's @var{this} pointer is
6722 unaliased by using @code{__restrict__} as a member function qualifier.
6725 void T::fn () __restrict__
6732 Within the body of @code{T::fn}, @var{this} will have the effective
6733 definition @code{T *__restrict__ const this}. Notice that the
6734 interpretation of a @code{__restrict__} member function qualifier is
6735 different to that of @code{const} or @code{volatile} qualifier, in that it
6736 is applied to the pointer rather than the object. This is consistent with
6737 other compilers which implement restricted pointers.
6739 As with all outermost parameter qualifiers, @code{__restrict__} is
6740 ignored in function definition matching. This means you only need to
6741 specify @code{__restrict__} in a function definition, rather than
6742 in a function prototype as well.
6745 @section Vague Linkage
6746 @cindex vague linkage
6748 There are several constructs in C++ which require space in the object
6749 file but are not clearly tied to a single translation unit. We say that
6750 these constructs have ``vague linkage''. Typically such constructs are
6751 emitted wherever they are needed, though sometimes we can be more
6755 @item Inline Functions
6756 Inline functions are typically defined in a header file which can be
6757 included in many different compilations. Hopefully they can usually be
6758 inlined, but sometimes an out-of-line copy is necessary, if the address
6759 of the function is taken or if inlining fails. In general, we emit an
6760 out-of-line copy in all translation units where one is needed. As an
6761 exception, we only emit inline virtual functions with the vtable, since
6762 it will always require a copy.
6764 Local static variables and string constants used in an inline function
6765 are also considered to have vague linkage, since they must be shared
6766 between all inlined and out-of-line instances of the function.
6770 C++ virtual functions are implemented in most compilers using a lookup
6771 table, known as a vtable. The vtable contains pointers to the virtual
6772 functions provided by a class, and each object of the class contains a
6773 pointer to its vtable (or vtables, in some multiple-inheritance
6774 situations). If the class declares any non-inline, non-pure virtual
6775 functions, the first one is chosen as the ``key method'' for the class,
6776 and the vtable is only emitted in the translation unit where the key
6779 @emph{Note:} If the chosen key method is later defined as inline, the
6780 vtable will still be emitted in every translation unit which defines it.
6781 Make sure that any inline virtuals are declared inline in the class
6782 body, even if they are not defined there.
6784 @item type_info objects
6787 C++ requires information about types to be written out in order to
6788 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
6789 For polymorphic classes (classes with virtual functions), the type_info
6790 object is written out along with the vtable so that @samp{dynamic_cast}
6791 can determine the dynamic type of a class object at runtime. For all
6792 other types, we write out the type_info object when it is used: when
6793 applying @samp{typeid} to an expression, throwing an object, or
6794 referring to a type in a catch clause or exception specification.
6796 @item Template Instantiations
6797 Most everything in this section also applies to template instantiations,
6798 but there are other options as well.
6799 @xref{Template Instantiation,,Where's the Template?}.
6803 When used with GNU ld version 2.8 or later on an ELF system such as
6804 Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of
6805 these constructs will be discarded at link time. This is known as
6808 On targets that don't support COMDAT, but do support weak symbols, GCC
6809 will use them. This way one copy will override all the others, but
6810 the unused copies will still take up space in the executable.
6812 For targets which do not support either COMDAT or weak symbols,
6813 most entities with vague linkage will be emitted as local symbols to
6814 avoid duplicate definition errors from the linker. This will not happen
6815 for local statics in inlines, however, as having multiple copies will
6816 almost certainly break things.
6818 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
6819 another way to control placement of these constructs.
6822 @section Declarations and Definitions in One Header
6824 @cindex interface and implementation headers, C++
6825 @cindex C++ interface and implementation headers
6826 C++ object definitions can be quite complex. In principle, your source
6827 code will need two kinds of things for each object that you use across
6828 more than one source file. First, you need an @dfn{interface}
6829 specification, describing its structure with type declarations and
6830 function prototypes. Second, you need the @dfn{implementation} itself.
6831 It can be tedious to maintain a separate interface description in a
6832 header file, in parallel to the actual implementation. It is also
6833 dangerous, since separate interface and implementation definitions may
6834 not remain parallel.
6836 @cindex pragmas, interface and implementation
6837 With GNU C++, you can use a single header file for both purposes.
6840 @emph{Warning:} The mechanism to specify this is in transition. For the
6841 nonce, you must use one of two @code{#pragma} commands; in a future
6842 release of GNU C++, an alternative mechanism will make these
6843 @code{#pragma} commands unnecessary.
6846 The header file contains the full definitions, but is marked with
6847 @samp{#pragma interface} in the source code. This allows the compiler
6848 to use the header file only as an interface specification when ordinary
6849 source files incorporate it with @code{#include}. In the single source
6850 file where the full implementation belongs, you can use either a naming
6851 convention or @samp{#pragma implementation} to indicate this alternate
6852 use of the header file.
6855 @item #pragma interface
6856 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
6857 @kindex #pragma interface
6858 Use this directive in @emph{header files} that define object classes, to save
6859 space in most of the object files that use those classes. Normally,
6860 local copies of certain information (backup copies of inline member
6861 functions, debugging information, and the internal tables that implement
6862 virtual functions) must be kept in each object file that includes class
6863 definitions. You can use this pragma to avoid such duplication. When a
6864 header file containing @samp{#pragma interface} is included in a
6865 compilation, this auxiliary information will not be generated (unless
6866 the main input source file itself uses @samp{#pragma implementation}).
6867 Instead, the object files will contain references to be resolved at link
6870 The second form of this directive is useful for the case where you have
6871 multiple headers with the same name in different directories. If you
6872 use this form, you must specify the same string to @samp{#pragma
6875 @item #pragma implementation
6876 @itemx #pragma implementation "@var{objects}.h"
6877 @kindex #pragma implementation
6878 Use this pragma in a @emph{main input file}, when you want full output from
6879 included header files to be generated (and made globally visible). The
6880 included header file, in turn, should use @samp{#pragma interface}.
6881 Backup copies of inline member functions, debugging information, and the
6882 internal tables used to implement virtual functions are all generated in
6883 implementation files.
6885 @cindex implied @code{#pragma implementation}
6886 @cindex @code{#pragma implementation}, implied
6887 @cindex naming convention, implementation headers
6888 If you use @samp{#pragma implementation} with no argument, it applies to
6889 an include file with the same basename@footnote{A file's @dfn{basename}
6890 was the name stripped of all leading path information and of trailing
6891 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
6892 file. For example, in @file{allclass.cc}, giving just
6893 @samp{#pragma implementation}
6894 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
6896 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
6897 an implementation file whenever you would include it from
6898 @file{allclass.cc} even if you never specified @samp{#pragma
6899 implementation}. This was deemed to be more trouble than it was worth,
6900 however, and disabled.
6902 If you use an explicit @samp{#pragma implementation}, it must appear in
6903 your source file @emph{before} you include the affected header files.
6905 Use the string argument if you want a single implementation file to
6906 include code from multiple header files. (You must also use
6907 @samp{#include} to include the header file; @samp{#pragma
6908 implementation} only specifies how to use the file---it doesn't actually
6911 There is no way to split up the contents of a single header file into
6912 multiple implementation files.
6915 @cindex inlining and C++ pragmas
6916 @cindex C++ pragmas, effect on inlining
6917 @cindex pragmas in C++, effect on inlining
6918 @samp{#pragma implementation} and @samp{#pragma interface} also have an
6919 effect on function inlining.
6921 If you define a class in a header file marked with @samp{#pragma
6922 interface}, the effect on a function defined in that class is similar to
6923 an explicit @code{extern} declaration---the compiler emits no code at
6924 all to define an independent version of the function. Its definition
6925 is used only for inlining with its callers.
6927 @opindex fno-implement-inlines
6928 Conversely, when you include the same header file in a main source file
6929 that declares it as @samp{#pragma implementation}, the compiler emits
6930 code for the function itself; this defines a version of the function
6931 that can be found via pointers (or by callers compiled without
6932 inlining). If all calls to the function can be inlined, you can avoid
6933 emitting the function by compiling with @option{-fno-implement-inlines}.
6934 If any calls were not inlined, you will get linker errors.
6936 @node Template Instantiation
6937 @section Where's the Template?
6939 @cindex template instantiation
6941 C++ templates are the first language feature to require more
6942 intelligence from the environment than one usually finds on a UNIX
6943 system. Somehow the compiler and linker have to make sure that each
6944 template instance occurs exactly once in the executable if it is needed,
6945 and not at all otherwise. There are two basic approaches to this
6946 problem, which I will refer to as the Borland model and the Cfront model.
6950 Borland C++ solved the template instantiation problem by adding the code
6951 equivalent of common blocks to their linker; the compiler emits template
6952 instances in each translation unit that uses them, and the linker
6953 collapses them together. The advantage of this model is that the linker
6954 only has to consider the object files themselves; there is no external
6955 complexity to worry about. This disadvantage is that compilation time
6956 is increased because the template code is being compiled repeatedly.
6957 Code written for this model tends to include definitions of all
6958 templates in the header file, since they must be seen to be
6962 The AT&T C++ translator, Cfront, solved the template instantiation
6963 problem by creating the notion of a template repository, an
6964 automatically maintained place where template instances are stored. A
6965 more modern version of the repository works as follows: As individual
6966 object files are built, the compiler places any template definitions and
6967 instantiations encountered in the repository. At link time, the link
6968 wrapper adds in the objects in the repository and compiles any needed
6969 instances that were not previously emitted. The advantages of this
6970 model are more optimal compilation speed and the ability to use the
6971 system linker; to implement the Borland model a compiler vendor also
6972 needs to replace the linker. The disadvantages are vastly increased
6973 complexity, and thus potential for error; for some code this can be
6974 just as transparent, but in practice it can been very difficult to build
6975 multiple programs in one directory and one program in multiple
6976 directories. Code written for this model tends to separate definitions
6977 of non-inline member templates into a separate file, which should be
6978 compiled separately.
6981 When used with GNU ld version 2.8 or later on an ELF system such as
6982 Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the
6983 Borland model. On other systems, g++ implements neither automatic
6986 A future version of g++ will support a hybrid model whereby the compiler
6987 will emit any instantiations for which the template definition is
6988 included in the compile, and store template definitions and
6989 instantiation context information into the object file for the rest.
6990 The link wrapper will extract that information as necessary and invoke
6991 the compiler to produce the remaining instantiations. The linker will
6992 then combine duplicate instantiations.
6994 In the mean time, you have the following options for dealing with
6995 template instantiations:
7000 Compile your template-using code with @option{-frepo}. The compiler will
7001 generate files with the extension @samp{.rpo} listing all of the
7002 template instantiations used in the corresponding object files which
7003 could be instantiated there; the link wrapper, @samp{collect2}, will
7004 then update the @samp{.rpo} files to tell the compiler where to place
7005 those instantiations and rebuild any affected object files. The
7006 link-time overhead is negligible after the first pass, as the compiler
7007 will continue to place the instantiations in the same files.
7009 This is your best option for application code written for the Borland
7010 model, as it will just work. Code written for the Cfront model will
7011 need to be modified so that the template definitions are available at
7012 one or more points of instantiation; usually this is as simple as adding
7013 @code{#include <tmethods.cc>} to the end of each template header.
7015 For library code, if you want the library to provide all of the template
7016 instantiations it needs, just try to link all of its object files
7017 together; the link will fail, but cause the instantiations to be
7018 generated as a side effect. Be warned, however, that this may cause
7019 conflicts if multiple libraries try to provide the same instantiations.
7020 For greater control, use explicit instantiation as described in the next
7024 @opindex fno-implicit-templates
7025 Compile your code with @option{-fno-implicit-templates} to disable the
7026 implicit generation of template instances, and explicitly instantiate
7027 all the ones you use. This approach requires more knowledge of exactly
7028 which instances you need than do the others, but it's less
7029 mysterious and allows greater control. You can scatter the explicit
7030 instantiations throughout your program, perhaps putting them in the
7031 translation units where the instances are used or the translation units
7032 that define the templates themselves; you can put all of the explicit
7033 instantiations you need into one big file; or you can create small files
7040 template class Foo<int>;
7041 template ostream& operator <<
7042 (ostream&, const Foo<int>&);
7045 for each of the instances you need, and create a template instantiation
7048 If you are using Cfront-model code, you can probably get away with not
7049 using @option{-fno-implicit-templates} when compiling files that don't
7050 @samp{#include} the member template definitions.
7052 If you use one big file to do the instantiations, you may want to
7053 compile it without @option{-fno-implicit-templates} so you get all of the
7054 instances required by your explicit instantiations (but not by any
7055 other files) without having to specify them as well.
7057 g++ has extended the template instantiation syntax outlined in the
7058 Working Paper to allow forward declaration of explicit instantiations
7059 (with @code{extern}), instantiation of the compiler support data for a
7060 template class (i.e.@: the vtable) without instantiating any of its
7061 members (with @code{inline}), and instantiation of only the static data
7062 members of a template class, without the support data or member
7063 functions (with (@code{static}):
7066 extern template int max (int, int);
7067 inline template class Foo<int>;
7068 static template class Foo<int>;
7072 Do nothing. Pretend g++ does implement automatic instantiation
7073 management. Code written for the Borland model will work fine, but
7074 each translation unit will contain instances of each of the templates it
7075 uses. In a large program, this can lead to an unacceptable amount of code
7079 @opindex fexternal-templates
7080 Add @samp{#pragma interface} to all files containing template
7081 definitions. For each of these files, add @samp{#pragma implementation
7082 "@var{filename}"} to the top of some @samp{.C} file which
7083 @samp{#include}s it. Then compile everything with
7084 @option{-fexternal-templates}. The templates will then only be expanded
7085 in the translation unit which implements them (i.e.@: has a @samp{#pragma
7086 implementation} line for the file where they live); all other files will
7087 use external references. If you're lucky, everything should work
7088 properly. If you get undefined symbol errors, you need to make sure
7089 that each template instance which is used in the program is used in the
7090 file which implements that template. If you don't have any use for a
7091 particular instance in that file, you can just instantiate it
7092 explicitly, using the syntax from the latest C++ working paper:
7095 template class A<int>;
7096 template ostream& operator << (ostream&, const A<int>&);
7099 This strategy will work with code written for either model. If you are
7100 using code written for the Cfront model, the file containing a class
7101 template and the file containing its member templates should be
7102 implemented in the same translation unit.
7105 @opindex falt-external-templates
7106 A slight variation on this approach is to use the flag
7107 @option{-falt-external-templates} instead. This flag causes template
7108 instances to be emitted in the translation unit that implements the
7109 header where they are first instantiated, rather than the one which
7110 implements the file where the templates are defined. This header must
7111 be the same in all translation units, or things are likely to break.
7113 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
7114 more discussion of these pragmas.
7117 @node Bound member functions
7118 @section Extracting the function pointer from a bound pointer to member function
7121 @cindex pointer to member function
7122 @cindex bound pointer to member function
7124 In C++, pointer to member functions (PMFs) are implemented using a wide
7125 pointer of sorts to handle all the possible call mechanisms; the PMF
7126 needs to store information about how to adjust the @samp{this} pointer,
7127 and if the function pointed to is virtual, where to find the vtable, and
7128 where in the vtable to look for the member function. If you are using
7129 PMFs in an inner loop, you should really reconsider that decision. If
7130 that is not an option, you can extract the pointer to the function that
7131 would be called for a given object/PMF pair and call it directly inside
7132 the inner loop, to save a bit of time.
7134 Note that you will still be paying the penalty for the call through a
7135 function pointer; on most modern architectures, such a call defeats the
7136 branch prediction features of the CPU@. This is also true of normal
7137 virtual function calls.
7139 The syntax for this extension is
7143 extern int (A::*fp)();
7144 typedef int (*fptr)(A *);
7146 fptr p = (fptr)(a.*fp);
7149 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
7150 no object is needed to obtain the address of the function. They can be
7151 converted to function pointers directly:
7154 fptr p1 = (fptr)(&A::foo);
7157 @opindex Wno-pmf-conversions
7158 You must specify @option{-Wno-pmf-conversions} to use this extension.
7160 @node C++ Attributes
7161 @section C++-Specific Variable, Function, and Type Attributes
7163 Some attributes only make sense for C++ programs.
7166 @item init_priority (@var{priority})
7167 @cindex init_priority attribute
7170 In Standard C++, objects defined at namespace scope are guaranteed to be
7171 initialized in an order in strict accordance with that of their definitions
7172 @emph{in a given translation unit}. No guarantee is made for initializations
7173 across translation units. However, GNU C++ allows users to control the
7174 order of initialization of objects defined at namespace scope with the
7175 @code{init_priority} attribute by specifying a relative @var{priority},
7176 a constant integral expression currently bounded between 101 and 65535
7177 inclusive. Lower numbers indicate a higher priority.
7179 In the following example, @code{A} would normally be created before
7180 @code{B}, but the @code{init_priority} attribute has reversed that order:
7183 Some_Class A __attribute__ ((init_priority (2000)));
7184 Some_Class B __attribute__ ((init_priority (543)));
7188 Note that the particular values of @var{priority} do not matter; only their
7191 @item java_interface
7192 @cindex java_interface attribute
7194 This type attribute informs C++ that the class is a Java interface. It may
7195 only be applied to classes declared within an @code{extern "Java"} block.
7196 Calls to methods declared in this interface will be dispatched using GCJ's
7197 interface table mechanism, instead of regular virtual table dispatch.
7201 @node Java Exceptions
7202 @section Java Exceptions
7204 The Java language uses a slightly different exception handling model
7205 from C++. Normally, GNU C++ will automatically detect when you are
7206 writing C++ code that uses Java exceptions, and handle them
7207 appropriately. However, if C++ code only needs to execute destructors
7208 when Java exceptions are thrown through it, GCC will guess incorrectly.
7209 Sample problematic code is:
7212 struct S @{ ~S(); @};
7213 extern void bar(); // is written in Java, and may throw exceptions
7222 The usual effect of an incorrect guess is a link failure, complaining of
7223 a missing routine called @samp{__gxx_personality_v0}.
7225 You can inform the compiler that Java exceptions are to be used in a
7226 translation unit, irrespective of what it might think, by writing
7227 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
7228 @samp{#pragma} must appear before any functions that throw or catch
7229 exceptions, or run destructors when exceptions are thrown through them.
7231 You cannot mix Java and C++ exceptions in the same translation unit. It
7232 is believed to be safe to throw a C++ exception from one file through
7233 another file compiled for the Java exception model, or vice versa, but
7234 there may be bugs in this area.
7236 @node Deprecated Features
7237 @section Deprecated Features
7239 In the past, the GNU C++ compiler was extended to experiment with new
7240 features, at a time when the C++ language was still evolving. Now that
7241 the C++ standard is complete, some of those features are superseded by
7242 superior alternatives. Using the old features might cause a warning in
7243 some cases that the feature will be dropped in the future. In other
7244 cases, the feature might be gone already.
7246 While the list below is not exhaustive, it documents some of the options
7247 that are now deprecated:
7250 @item -fexternal-templates
7251 @itemx -falt-external-templates
7252 These are two of the many ways for g++ to implement template
7253 instantiation. @xref{Template Instantiation}. The C++ standard clearly
7254 defines how template definitions have to be organized across
7255 implementation units. g++ has an implicit instantiation mechanism that
7256 should work just fine for standard-conforming code.
7258 @item -fstrict-prototype
7259 @itemx -fno-strict-prototype
7260 Previously it was possible to use an empty prototype parameter list to
7261 indicate an unspecified number of parameters (like C), rather than no
7262 parameters, as C++ demands. This feature has been removed, except where
7263 it is required for backwards compatibility @xref{Backwards Compatibility}.
7266 The named return value extension has been deprecated, and is now
7269 The use of initializer lists with new expressions has been deprecated,
7270 and is now removed from g++.
7272 Floating and complex non-type template parameters have been deprecated,
7273 and are now removed from g++.
7275 The implicit typename extension has been deprecated and will be removed
7276 from g++ at some point. In some cases g++ determines that a dependant
7277 type such as @code{TPL<T>::X} is a type without needing a
7278 @code{typename} keyword, contrary to the standard.
7280 @node Backwards Compatibility
7281 @section Backwards Compatibility
7282 @cindex Backwards Compatibility
7283 @cindex ARM [Annotated C++ Reference Manual]
7285 Now that there is a definitive ISO standard C++, G++ has a specification
7286 to adhere to. The C++ language evolved over time, and features that
7287 used to be acceptable in previous drafts of the standard, such as the ARM
7288 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
7289 compilation of C++ written to such drafts, G++ contains some backwards
7290 compatibilities. @emph{All such backwards compatibility features are
7291 liable to disappear in future versions of G++.} They should be considered
7292 deprecated @xref{Deprecated Features}.
7296 If a variable is declared at for scope, it used to remain in scope until
7297 the end of the scope which contained the for statement (rather than just
7298 within the for scope). G++ retains this, but issues a warning, if such a
7299 variable is accessed outside the for scope.
7301 @item Implicit C language
7302 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
7303 scope to set the language. On such systems, all header files are
7304 implicitly scoped inside a C language scope. Also, an empty prototype
7305 @code{()} will be treated as an unspecified number of arguments, rather
7306 than no arguments, as C++ demands.