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).}
38 Diagnostics consist of all the output sent to stderr by GCC.
41 @cite{Whether each nonempty sequence of white-space characters other than
42 new-line is retained or replaced by one space character in translation
46 @node Environment implementation
49 The behavior of these points are dependent on the implementation
50 of the C library, and are not defined by GCC itself.
52 @node Identifiers implementation
57 @cite{Which additional multibyte characters may appear in identifiers
58 and their correspondence to universal character names (6.4.2).}
61 @cite{The number of significant initial characters in an identifier
64 For internal names, all characters are significant. For external names,
65 the number of significant characters are defined by the linker; for
66 almost all targets, all characters are significant.
70 @node Characters implementation
75 @cite{The number of bits in a byte (3.6).}
78 @cite{The values of the members of the execution character set (5.2.1).}
81 @cite{The unique value of the member of the execution character set produced
82 for each of the standard alphabetic escape sequences (5.2.2).}
85 @cite{The value of a @code{char} object into which has been stored any
86 character other than a member of the basic execution character set (6.2.5).}
89 @cite{Which of @code{signed char} or @code{unsigned char} has the same range,
90 representation, and behavior as ``plain'' @code{char} (6.2.5, 6.3.1.1).}
93 @cite{The mapping of members of the source character set (in character
94 constants and string literals) to members of the execution character
95 set (6.4.4.4, 5.1.1.2).}
98 @cite{The value of an integer character constant containing more than one
99 character or containing a character or escape sequence that does not map
100 to a single-byte execution character (6.4.4.4).}
103 @cite{The value of a wide character constant containing more than one
104 multibyte character, or containing a multibyte character or escape
105 sequence not represented in the extended execution character set (6.4.4.4).}
108 @cite{The current locale used to convert a wide character constant consisting
109 of a single multibyte character that maps to a member of the extended
110 execution character set into a corresponding wide character code (6.4.4.4).}
113 @cite{The current locale used to convert a wide string literal into
114 corresponding wide character codes (6.4.5).}
117 @cite{The value of a string literal containing a multibyte character or escape
118 sequence not represented in the execution character set (6.4.5).}
121 @node Integers implementation
126 @cite{Any extended integer types that exist in the implementation (6.2.5).}
129 @cite{Whether signed integer types are represented using sign and magnitude,
130 two's complement, or one's complement, and whether the extraordinary value
131 is a trap representation or an ordinary value (6.2.6.2).}
133 GCC supports only two's complement integer types, and all bit patterns
137 @cite{The rank of any extended integer type relative to another extended
138 integer type with the same precision (6.3.1.1).}
141 @cite{The result of, or the signal raised by, converting an integer to a
142 signed integer type when the value cannot be represented in an object of
143 that type (6.3.1.3).}
146 @cite{The results of some bitwise operations on signed integers (6.5).}
149 @node Floating point implementation
150 @section Floating point
154 @cite{The accuracy of the floating-point operations and of the library
155 functions in @code{<math.h>} and @code{<complex.h>} that return floating-point
156 results (5.2.4.2.2).}
159 @cite{The rounding behaviors characterized by non-standard values
160 of @code{FLT_ROUNDS} @gol
164 @cite{The evaluation methods characterized by non-standard negative
165 values of @code{FLT_EVAL_METHOD} (5.2.4.2.2).}
168 @cite{The direction of rounding when an integer is converted to a
169 floating-point number that cannot exactly represent the original
173 @cite{The direction of rounding when a floating-point number is
174 converted to a narrower floating-point number (6.3.1.5).}
177 @cite{How the nearest representable value or the larger or smaller
178 representable value immediately adjacent to the nearest representable
179 value is chosen for certain floating constants (6.4.4.2).}
182 @cite{Whether and how floating expressions are contracted when not
183 disallowed by the @code{FP_CONTRACT} pragma (6.5).}
186 @cite{The default state for the @code{FENV_ACCESS} pragma (7.6.1).}
189 @cite{Additional floating-point exceptions, rounding modes, environments,
190 and classifications, and their macro names (7.6, 7.12).}
193 @cite{The default state for the @code{FP_CONTRACT} pragma (7.12.2).}
196 @cite{Whether the ``inexact'' floating-point exception can be raised
197 when the rounded result actually does equal the mathematical result
198 in an IEC 60559 conformant implementation (F.9).}
201 @cite{Whether the ``underflow'' (and ``inexact'') floating-point
202 exception can be raised when a result is tiny but not inexact in an
203 IEC 60559 conformant implementation (F.9).}
207 @node Arrays and pointers implementation
208 @section Arrays and pointers
212 @cite{The result of converting a pointer to an integer or
213 vice versa (6.3.2.3).}
215 A cast from pointer to integer discards most-significant bits if the
216 pointer representation is larger than the integer type,
217 sign-extends@footnote{Future versions of GCC may zero-extend, or use
218 a target-defined @code{ptr_extend} pattern. Do not rely on sign extension.}
219 if the pointer representation is smaller than the integer type, otherwise
220 the bits are unchanged.
221 @c ??? We've always claimed that pointers were unsigned entities.
222 @c Shouldn't we therefore be doing zero-extension? If so, the bug
223 @c is in convert_to_integer, where we call type_for_size and request
224 @c a signed integral type. On the other hand, it might be most useful
225 @c for the target if we extend according to POINTERS_EXTEND_UNSIGNED.
227 A cast from integer to pointer discards most-significant bits if the
228 pointer representation is smaller than the integer type, extends according
229 to the signedness of the integer type if the pointer representation
230 is larger than the integer type, otherwise the bits are unchanged.
232 When casting from pointer to integer and back again, the resulting
233 pointer must reference the same object as the original pointer, otherwise
234 the behavior is undefined. That is, one may not use integer arithmetic to
235 avoid the undefined behavior of pointer arithmetic as proscribed in 6.5.6/8.
238 @cite{The size of the result of subtracting two pointers to elements
239 of the same array (6.5.6).}
243 @node Hints implementation
248 @cite{The extent to which suggestions made by using the @code{register}
249 storage-class specifier are effective (6.7.1).}
251 The @code{register} specifier affects code generation only in these ways:
255 When used as part of the register variable extension, see
256 @ref{Explicit Reg Vars}.
259 When @option{-O0} is in use, the compiler allocates distinct stack
260 memory for all variables that do not have the @code{register}
261 storage-class specifier; if @code{register} is specified, the variable
262 may have a shorter lifespan than the code would indicate and may never
266 On some rare x86 targets, @code{setjmp} doesn't save the registers in
267 all circumstances. In those cases, GCC doesn't allocate any variables
268 in registers unless they are marked @code{register}.
273 @cite{The extent to which suggestions made by using the inline function
274 specifier are effective (6.7.4).}
276 GCC will not inline any functions if the @option{-fno-inline} option is
277 used or if @option{-O0} is used. Otherwise, GCC may still be unable to
278 inline a function for many reasons; the @option{-Winline} option may be
279 used to determine if a function has not been inlined and why not.
283 @node Structures unions enumerations and bit-fields implementation
284 @section Structures, unions, enumerations, and bit-fields
288 @cite{Whether a ``plain'' int bit-field is treated as a @code{signed int}
289 bit-field or as an @code{unsigned int} bit-field (6.7.2, 6.7.2.1).}
292 @cite{Allowable bit-field types other than @code{_Bool}, @code{signed int},
293 and @code{unsigned int} (6.7.2.1).}
296 @cite{Whether a bit-field can straddle a storage-unit boundary (6.7.2.1).}
299 @cite{The order of allocation of bit-fields within a unit (6.7.2.1).}
302 @cite{The alignment of non-bit-field members of structures (6.7.2.1).}
305 @cite{The integer type compatible with each enumerated type (6.7.2.2).}
309 @node Qualifiers implementation
314 @cite{What constitutes an access to an object that has volatile-qualified
319 @node Preprocessing directives implementation
320 @section Preprocessing directives
324 @cite{How sequences in both forms of header names are mapped to headers
325 or external source file names (6.4.7).}
328 @cite{Whether the value of a character constant in a constant expression
329 that controls conditional inclusion matches the value of the same character
330 constant in the execution character set (6.10.1).}
333 @cite{Whether the value of a single-character character constant in a
334 constant expression that controls conditional inclusion may have a
335 negative value (6.10.1).}
338 @cite{The places that are searched for an included @samp{<>} delimited
339 header, and how the places are specified or the header is
340 identified (6.10.2).}
343 @cite{How the named source file is searched for in an included @samp{""}
344 delimited header (6.10.2).}
347 @cite{The method by which preprocessing tokens (possibly resulting from
348 macro expansion) in a @code{#include} directive are combined into a header
352 @cite{The nesting limit for @code{#include} processing (6.10.2).}
354 GCC imposes a limit of 200 nested @code{#include}s.
357 @cite{Whether the @samp{#} operator inserts a @samp{\} character before
358 the @samp{\} character that begins a universal character name in a
359 character constant or string literal (6.10.3.2).}
362 @cite{The behavior on each recognized non-@code{STDC #pragma}
366 @cite{The definitions for @code{__DATE__} and @code{__TIME__} when
367 respectively, the date and time of translation are not available (6.10.8).}
369 If the date and time are not available, @code{__DATE__} expands to
370 @code{@w{"??? ?? ????"}} and @code{__TIME__} expands to
375 @node Library functions implementation
376 @section Library functions
378 The behavior of these points are dependent on the implementation
379 of the C library, and are not defined by GCC itself.
381 @node Architecture implementation
382 @section Architecture
386 @cite{The values or expressions assigned to the macros specified in the
387 headers @code{<float.h>}, @code{<limits.h>}, and @code{<stdint.h>}
388 (5.2.4.2, 7.18.2, 7.18.3).}
391 @cite{The number, order, and encoding of bytes in any object
392 (when not explicitly specified in this International Standard) (6.2.6.1).}
395 @cite{The value of the result of the sizeof operator (6.5.3.4).}
399 @node Locale-specific behavior implementation
400 @section Locale-specific behavior
402 The behavior of these points are dependent on the implementation
403 of the C library, and are not defined by GCC itself.
406 @chapter Extensions to the C Language Family
407 @cindex extensions, C language
408 @cindex C language extensions
411 GNU C provides several language features not found in ISO standard C@.
412 (The @option{-pedantic} option directs GCC to print a warning message if
413 any of these features is used.) To test for the availability of these
414 features in conditional compilation, check for a predefined macro
415 @code{__GNUC__}, which is always defined under GCC@.
417 These extensions are available in C and Objective-C@. Most of them are
418 also available in C++. @xref{C++ Extensions,,Extensions to the
419 C++ Language}, for extensions that apply @emph{only} to C++.
421 Some features that are in ISO C99 but not C89 or C++ are also, as
422 extensions, accepted by GCC in C89 mode and in C++.
425 * Statement Exprs:: Putting statements and declarations inside expressions.
426 * Local Labels:: Labels local to a statement-expression.
427 * Labels as Values:: Getting pointers to labels, and computed gotos.
428 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
429 * Constructing Calls:: Dispatching a call to another function.
430 * Typeof:: @code{typeof}: referring to the type of an expression.
431 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
432 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
433 * Long Long:: Double-word integers---@code{long long int}.
434 * Complex:: Data types for complex numbers.
435 * Hex Floats:: Hexadecimal floating-point constants.
436 * Zero Length:: Zero-length arrays.
437 * Variable Length:: Arrays whose length is computed at run time.
438 * Variadic Macros:: Macros with a variable number of arguments.
439 * Escaped Newlines:: Slightly looser rules for escaped newlines.
440 * Multi-line Strings:: String literals with embedded newlines.
441 * Subscripting:: Any array can be subscripted, even if not an lvalue.
442 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
443 * Initializers:: Non-constant initializers.
444 * Compound Literals:: Compound literals give structures, unions
446 * Designated Inits:: Labeling elements of initializers.
447 * Cast to Union:: Casting to union type from any member of the union.
448 * Case Ranges:: `case 1 ... 9' and such.
449 * Mixed Declarations:: Mixing declarations and code.
450 * Function Attributes:: Declaring that functions have no side effects,
451 or that they can never return.
452 * Attribute Syntax:: Formal syntax for attributes.
453 * Function Prototypes:: Prototype declarations and old-style definitions.
454 * C++ Comments:: C++ comments are recognized.
455 * Dollar Signs:: Dollar sign is allowed in identifiers.
456 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
457 * Variable Attributes:: Specifying attributes of variables.
458 * Type Attributes:: Specifying attributes of types.
459 * Alignment:: Inquiring about the alignment of a type or variable.
460 * Inline:: Defining inline functions (as fast as macros).
461 * Extended Asm:: Assembler instructions with C expressions as operands.
462 (With them you can define ``built-in'' functions.)
463 * Constraints:: Constraints for asm operands
464 * Asm Labels:: Specifying the assembler name to use for a C symbol.
465 * Explicit Reg Vars:: Defining variables residing in specified registers.
466 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
467 * Incomplete Enums:: @code{enum foo;}, with details to follow.
468 * Function Names:: Printable strings which are the name of the current
470 * Return Address:: Getting the return or frame address of a function.
471 * Vector Extensions:: Using vector instructions through built-in functions.
472 * Other Builtins:: Other built-in functions.
473 * Target Builtins:: Built-in functions specific to particular targets.
474 * Pragmas:: Pragmas accepted by GCC.
475 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
476 * Thread-Local:: Per-thread variables.
479 @node Statement Exprs
480 @section Statements and Declarations in Expressions
481 @cindex statements inside expressions
482 @cindex declarations inside expressions
483 @cindex expressions containing statements
484 @cindex macros, statements in expressions
486 @c the above section title wrapped and causes an underfull hbox.. i
487 @c changed it from "within" to "in". --mew 4feb93
489 A compound statement enclosed in parentheses may appear as an expression
490 in GNU C@. This allows you to use loops, switches, and local variables
491 within an expression.
493 Recall that a compound statement is a sequence of statements surrounded
494 by braces; in this construct, parentheses go around the braces. For
498 (@{ int y = foo (); int z;
505 is a valid (though slightly more complex than necessary) expression
506 for the absolute value of @code{foo ()}.
508 The last thing in the compound statement should be an expression
509 followed by a semicolon; the value of this subexpression serves as the
510 value of the entire construct. (If you use some other kind of statement
511 last within the braces, the construct has type @code{void}, and thus
512 effectively no value.)
514 This feature is especially useful in making macro definitions ``safe'' (so
515 that they evaluate each operand exactly once). For example, the
516 ``maximum'' function is commonly defined as a macro in standard C as
520 #define max(a,b) ((a) > (b) ? (a) : (b))
524 @cindex side effects, macro argument
525 But this definition computes either @var{a} or @var{b} twice, with bad
526 results if the operand has side effects. In GNU C, if you know the
527 type of the operands (here let's assume @code{int}), you can define
528 the macro safely as follows:
531 #define maxint(a,b) \
532 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
535 Embedded statements are not allowed in constant expressions, such as
536 the value of an enumeration constant, the width of a bit-field, or
537 the initial value of a static variable.
539 If you don't know the type of the operand, you can still do this, but you
540 must use @code{typeof} (@pxref{Typeof}).
542 Statement expressions are not supported fully in G++, and their fate
543 there is unclear. (It is possible that they will become fully supported
544 at some point, or that they will be deprecated, or that the bugs that
545 are present will continue to exist indefinitely.) Presently, statement
546 expressions do not work well as default arguments.
548 In addition, there are semantic issues with statement-expressions in
549 C++. If you try to use statement-expressions instead of inline
550 functions in C++, you may be surprised at the way object destruction is
551 handled. For example:
554 #define foo(a) (@{int b = (a); b + 3; @})
558 does not work the same way as:
561 inline int foo(int a) @{ int b = a; return b + 3; @}
565 In particular, if the expression passed into @code{foo} involves the
566 creation of temporaries, the destructors for those temporaries will be
567 run earlier in the case of the macro than in the case of the function.
569 These considerations mean that it is probably a bad idea to use
570 statement-expressions of this form in header files that are designed to
571 work with C++. (Note that some versions of the GNU C Library contained
572 header files using statement-expression that lead to precisely this
576 @section Locally Declared Labels
578 @cindex macros, local labels
580 Each statement expression is a scope in which @dfn{local labels} can be
581 declared. A local label is simply an identifier; you can jump to it
582 with an ordinary @code{goto} statement, but only from within the
583 statement expression it belongs to.
585 A local label declaration looks like this:
588 __label__ @var{label};
595 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
598 Local label declarations must come at the beginning of the statement
599 expression, right after the @samp{(@{}, before any ordinary
602 The label declaration defines the label @emph{name}, but does not define
603 the label itself. You must do this in the usual way, with
604 @code{@var{label}:}, within the statements of the statement expression.
606 The local label feature is useful because statement expressions are
607 often used in macros. If the macro contains nested loops, a @code{goto}
608 can be useful for breaking out of them. However, an ordinary label
609 whose scope is the whole function cannot be used: if the macro can be
610 expanded several times in one function, the label will be multiply
611 defined in that function. A local label avoids this problem. For
615 #define SEARCH(array, target) \
618 typeof (target) _SEARCH_target = (target); \
619 typeof (*(array)) *_SEARCH_array = (array); \
622 for (i = 0; i < max; i++) \
623 for (j = 0; j < max; j++) \
624 if (_SEARCH_array[i][j] == _SEARCH_target) \
625 @{ value = i; goto found; @} \
632 @node Labels as Values
633 @section Labels as Values
634 @cindex labels as values
635 @cindex computed gotos
636 @cindex goto with computed label
637 @cindex address of a label
639 You can get the address of a label defined in the current function
640 (or a containing function) with the unary operator @samp{&&}. The
641 value has type @code{void *}. This value is a constant and can be used
642 wherever a constant of that type is valid. For example:
650 To use these values, you need to be able to jump to one. This is done
651 with the computed goto statement@footnote{The analogous feature in
652 Fortran is called an assigned goto, but that name seems inappropriate in
653 C, where one can do more than simply store label addresses in label
654 variables.}, @code{goto *@var{exp};}. For example,
661 Any expression of type @code{void *} is allowed.
663 One way of using these constants is in initializing a static array that
664 will serve as a jump table:
667 static void *array[] = @{ &&foo, &&bar, &&hack @};
670 Then you can select a label with indexing, like this:
677 Note that this does not check whether the subscript is in bounds---array
678 indexing in C never does that.
680 Such an array of label values serves a purpose much like that of the
681 @code{switch} statement. The @code{switch} statement is cleaner, so
682 use that rather than an array unless the problem does not fit a
683 @code{switch} statement very well.
685 Another use of label values is in an interpreter for threaded code.
686 The labels within the interpreter function can be stored in the
687 threaded code for super-fast dispatching.
689 You may not use this mechanism to jump to code in a different function.
690 If you do that, totally unpredictable things will happen. The best way to
691 avoid this is to store the label address only in automatic variables and
692 never pass it as an argument.
694 An alternate way to write the above example is
697 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
699 goto *(&&foo + array[i]);
703 This is more friendly to code living in shared libraries, as it reduces
704 the number of dynamic relocations that are needed, and by consequence,
705 allows the data to be read-only.
707 @node Nested Functions
708 @section Nested Functions
709 @cindex nested functions
710 @cindex downward funargs
713 A @dfn{nested function} is a function defined inside another function.
714 (Nested functions are not supported for GNU C++.) The nested function's
715 name is local to the block where it is defined. For example, here we
716 define a nested function named @code{square}, and call it twice:
720 foo (double a, double b)
722 double square (double z) @{ return z * z; @}
724 return square (a) + square (b);
729 The nested function can access all the variables of the containing
730 function that are visible at the point of its definition. This is
731 called @dfn{lexical scoping}. For example, here we show a nested
732 function which uses an inherited variable named @code{offset}:
736 bar (int *array, int offset, int size)
738 int access (int *array, int index)
739 @{ return array[index + offset]; @}
742 for (i = 0; i < size; i++)
743 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
748 Nested function definitions are permitted within functions in the places
749 where variable definitions are allowed; that is, in any block, before
750 the first statement in the block.
752 It is possible to call the nested function from outside the scope of its
753 name by storing its address or passing the address to another function:
756 hack (int *array, int size)
758 void store (int index, int value)
759 @{ array[index] = value; @}
761 intermediate (store, size);
765 Here, the function @code{intermediate} receives the address of
766 @code{store} as an argument. If @code{intermediate} calls @code{store},
767 the arguments given to @code{store} are used to store into @code{array}.
768 But this technique works only so long as the containing function
769 (@code{hack}, in this example) does not exit.
771 If you try to call the nested function through its address after the
772 containing function has exited, all hell will break loose. If you try
773 to call it after a containing scope level has exited, and if it refers
774 to some of the variables that are no longer in scope, you may be lucky,
775 but it's not wise to take the risk. If, however, the nested function
776 does not refer to anything that has gone out of scope, you should be
779 GCC implements taking the address of a nested function using a technique
780 called @dfn{trampolines}. A paper describing them is available as
783 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
785 A nested function can jump to a label inherited from a containing
786 function, provided the label was explicitly declared in the containing
787 function (@pxref{Local Labels}). Such a jump returns instantly to the
788 containing function, exiting the nested function which did the
789 @code{goto} and any intermediate functions as well. Here is an example:
793 bar (int *array, int offset, int size)
796 int access (int *array, int index)
800 return array[index + offset];
804 for (i = 0; i < size; i++)
805 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
809 /* @r{Control comes here from @code{access}
810 if it detects an error.} */
817 A nested function always has internal linkage. Declaring one with
818 @code{extern} is erroneous. If you need to declare the nested function
819 before its definition, use @code{auto} (which is otherwise meaningless
820 for function declarations).
823 bar (int *array, int offset, int size)
826 auto int access (int *, int);
828 int access (int *array, int index)
832 return array[index + offset];
838 @node Constructing Calls
839 @section Constructing Function Calls
840 @cindex constructing calls
841 @cindex forwarding calls
843 Using the built-in functions described below, you can record
844 the arguments a function received, and call another function
845 with the same arguments, without knowing the number or types
848 You can also record the return value of that function call,
849 and later return that value, without knowing what data type
850 the function tried to return (as long as your caller expects
853 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
854 This built-in function returns a pointer to data
855 describing how to perform a call with the same arguments as were passed
856 to the current function.
858 The function saves the arg pointer register, structure value address,
859 and all registers that might be used to pass arguments to a function
860 into a block of memory allocated on the stack. Then it returns the
861 address of that block.
864 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
865 This built-in function invokes @var{function}
866 with a copy of the parameters described by @var{arguments}
869 The value of @var{arguments} should be the value returned by
870 @code{__builtin_apply_args}. The argument @var{size} specifies the size
871 of the stack argument data, in bytes.
873 This function returns a pointer to data describing
874 how to return whatever value was returned by @var{function}. The data
875 is saved in a block of memory allocated on the stack.
877 It is not always simple to compute the proper value for @var{size}. The
878 value is used by @code{__builtin_apply} to compute the amount of data
879 that should be pushed on the stack and copied from the incoming argument
883 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
884 This built-in function returns the value described by @var{result} from
885 the containing function. You should specify, for @var{result}, a value
886 returned by @code{__builtin_apply}.
889 @cindex underscores in variables in macros
890 @cindex @samp{_} in variables in macros
891 @cindex local variables in macros
892 @cindex variables, local, in macros
893 @cindex macros, local variables in
895 The reason for using names that start with underscores for the local
896 variables is to avoid conflicts with variable names that occur within the
897 expressions that are substituted for @code{a} and @code{b}. Eventually we
898 hope to design a new form of declaration syntax that allows you to declare
899 variables whose scopes start only after their initializers; this will be a
900 more reliable way to prevent such conflicts.
903 @section Referring to a Type with @code{typeof}
906 @cindex macros, types of arguments
908 Another way to refer to the type of an expression is with @code{typeof}.
909 The syntax of using of this keyword looks like @code{sizeof}, but the
910 construct acts semantically like a type name defined with @code{typedef}.
912 There are two ways of writing the argument to @code{typeof}: with an
913 expression or with a type. Here is an example with an expression:
920 This assumes that @code{x} is an array of pointers to functions;
921 the type described is that of the values of the functions.
923 Here is an example with a typename as the argument:
930 Here the type described is that of pointers to @code{int}.
932 If you are writing a header file that must work when included in ISO C
933 programs, write @code{__typeof__} instead of @code{typeof}.
934 @xref{Alternate Keywords}.
936 A @code{typeof}-construct can be used anywhere a typedef name could be
937 used. For example, you can use it in a declaration, in a cast, or inside
938 of @code{sizeof} or @code{typeof}.
940 @code{typeof} is often useful in conjunction with the
941 statements-within-expressions feature. Here is how the two together can
942 be used to define a safe ``maximum'' macro that operates on any
943 arithmetic type and evaluates each of its arguments exactly once:
947 (@{ typeof (a) _a = (a); \
948 typeof (b) _b = (b); \
949 _a > _b ? _a : _b; @})
953 Some more examples of the use of @code{typeof}:
957 This declares @code{y} with the type of what @code{x} points to.
964 This declares @code{y} as an array of such values.
971 This declares @code{y} as an array of pointers to characters:
974 typeof (typeof (char *)[4]) y;
978 It is equivalent to the following traditional C declaration:
984 To see the meaning of the declaration using @code{typeof}, and why it
985 might be a useful way to write, let's rewrite it with these macros:
988 #define pointer(T) typeof(T *)
989 #define array(T, N) typeof(T [N])
993 Now the declaration can be rewritten this way:
996 array (pointer (char), 4) y;
1000 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1001 pointers to @code{char}.
1004 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
1005 a more limited extension which permitted one to write
1008 typedef @var{T} = @var{expr};
1012 with the effect of declaring @var{T} to have the type of the expression
1013 @var{expr}. This extension does not work with GCC 3 (versions between
1014 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
1015 relies on it should be rewritten to use @code{typeof}:
1018 typedef typeof(@var{expr}) @var{T};
1022 This will work with all versions of GCC@.
1025 @section Generalized Lvalues
1026 @cindex compound expressions as lvalues
1027 @cindex expressions, compound, as lvalues
1028 @cindex conditional expressions as lvalues
1029 @cindex expressions, conditional, as lvalues
1030 @cindex casts as lvalues
1031 @cindex generalized lvalues
1032 @cindex lvalues, generalized
1033 @cindex extensions, @code{?:}
1034 @cindex @code{?:} extensions
1035 Compound expressions, conditional expressions and casts are allowed as
1036 lvalues provided their operands are lvalues. This means that you can take
1037 their addresses or store values into them.
1039 Standard C++ allows compound expressions and conditional expressions as
1040 lvalues, and permits casts to reference type, so use of this extension
1041 is deprecated for C++ code.
1043 For example, a compound expression can be assigned, provided the last
1044 expression in the sequence is an lvalue. These two expressions are
1052 Similarly, the address of the compound expression can be taken. These two
1053 expressions are equivalent:
1060 A conditional expression is a valid lvalue if its type is not void and the
1061 true and false branches are both valid lvalues. For example, these two
1062 expressions are equivalent:
1066 (a ? b = 5 : (c = 5))
1069 A cast is a valid lvalue if its operand is an lvalue. A simple
1070 assignment whose left-hand side is a cast works by converting the
1071 right-hand side first to the specified type, then to the type of the
1072 inner left-hand side expression. After this is stored, the value is
1073 converted back to the specified type to become the value of the
1074 assignment. Thus, if @code{a} has type @code{char *}, the following two
1075 expressions are equivalent:
1079 (int)(a = (char *)(int)5)
1082 An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1083 performs the arithmetic using the type resulting from the cast, and then
1084 continues as in the previous case. Therefore, these two expressions are
1089 (int)(a = (char *)(int) ((int)a + 5))
1092 You cannot take the address of an lvalue cast, because the use of its
1093 address would not work out coherently. Suppose that @code{&(int)f} were
1094 permitted, where @code{f} has type @code{float}. Then the following
1095 statement would try to store an integer bit-pattern where a floating
1096 point number belongs:
1102 This is quite different from what @code{(int)f = 1} would do---that
1103 would convert 1 to floating point and store it. Rather than cause this
1104 inconsistency, we think it is better to prohibit use of @samp{&} on a cast.
1106 If you really do want an @code{int *} pointer with the address of
1107 @code{f}, you can simply write @code{(int *)&f}.
1110 @section Conditionals with Omitted Operands
1111 @cindex conditional expressions, extensions
1112 @cindex omitted middle-operands
1113 @cindex middle-operands, omitted
1114 @cindex extensions, @code{?:}
1115 @cindex @code{?:} extensions
1117 The middle operand in a conditional expression may be omitted. Then
1118 if the first operand is nonzero, its value is the value of the conditional
1121 Therefore, the expression
1128 has the value of @code{x} if that is nonzero; otherwise, the value of
1131 This example is perfectly equivalent to
1137 @cindex side effect in ?:
1138 @cindex ?: side effect
1140 In this simple case, the ability to omit the middle operand is not
1141 especially useful. When it becomes useful is when the first operand does,
1142 or may (if it is a macro argument), contain a side effect. Then repeating
1143 the operand in the middle would perform the side effect twice. Omitting
1144 the middle operand uses the value already computed without the undesirable
1145 effects of recomputing it.
1148 @section Double-Word Integers
1149 @cindex @code{long long} data types
1150 @cindex double-word arithmetic
1151 @cindex multiprecision arithmetic
1152 @cindex @code{LL} integer suffix
1153 @cindex @code{ULL} integer suffix
1155 ISO C99 supports data types for integers that are at least 64 bits wide,
1156 and as an extension GCC supports them in C89 mode and in C++.
1157 Simply write @code{long long int} for a signed integer, or
1158 @code{unsigned long long int} for an unsigned integer. To make an
1159 integer constant of type @code{long long int}, add the suffix @samp{LL}
1160 to the integer. To make an integer constant of type @code{unsigned long
1161 long int}, add the suffix @samp{ULL} to the integer.
1163 You can use these types in arithmetic like any other integer types.
1164 Addition, subtraction, and bitwise boolean operations on these types
1165 are open-coded on all types of machines. Multiplication is open-coded
1166 if the machine supports fullword-to-doubleword a widening multiply
1167 instruction. Division and shifts are open-coded only on machines that
1168 provide special support. The operations that are not open-coded use
1169 special library routines that come with GCC@.
1171 There may be pitfalls when you use @code{long long} types for function
1172 arguments, unless you declare function prototypes. If a function
1173 expects type @code{int} for its argument, and you pass a value of type
1174 @code{long long int}, confusion will result because the caller and the
1175 subroutine will disagree about the number of bytes for the argument.
1176 Likewise, if the function expects @code{long long int} and you pass
1177 @code{int}. The best way to avoid such problems is to use prototypes.
1180 @section Complex Numbers
1181 @cindex complex numbers
1182 @cindex @code{_Complex} keyword
1183 @cindex @code{__complex__} keyword
1185 ISO C99 supports complex floating data types, and as an extension GCC
1186 supports them in C89 mode and in C++, and supports complex integer data
1187 types which are not part of ISO C99. You can declare complex types
1188 using the keyword @code{_Complex}. As an extension, the older GNU
1189 keyword @code{__complex__} is also supported.
1191 For example, @samp{_Complex double x;} declares @code{x} as a
1192 variable whose real part and imaginary part are both of type
1193 @code{double}. @samp{_Complex short int y;} declares @code{y} to
1194 have real and imaginary parts of type @code{short int}; this is not
1195 likely to be useful, but it shows that the set of complex types is
1198 To write a constant with a complex data type, use the suffix @samp{i} or
1199 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
1200 has type @code{_Complex float} and @code{3i} has type
1201 @code{_Complex int}. Such a constant always has a pure imaginary
1202 value, but you can form any complex value you like by adding one to a
1203 real constant. This is a GNU extension; if you have an ISO C99
1204 conforming C library (such as GNU libc), and want to construct complex
1205 constants of floating type, you should include @code{<complex.h>} and
1206 use the macros @code{I} or @code{_Complex_I} instead.
1208 @cindex @code{__real__} keyword
1209 @cindex @code{__imag__} keyword
1210 To extract the real part of a complex-valued expression @var{exp}, write
1211 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
1212 extract the imaginary part. This is a GNU extension; for values of
1213 floating type, you should use the ISO C99 functions @code{crealf},
1214 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
1215 @code{cimagl}, declared in @code{<complex.h>} and also provided as
1216 built-in functions by GCC@.
1218 @cindex complex conjugation
1219 The operator @samp{~} performs complex conjugation when used on a value
1220 with a complex type. This is a GNU extension; for values of
1221 floating type, you should use the ISO C99 functions @code{conjf},
1222 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
1223 provided as built-in functions by GCC@.
1225 GCC can allocate complex automatic variables in a noncontiguous
1226 fashion; it's even possible for the real part to be in a register while
1227 the imaginary part is on the stack (or vice-versa). Only the DWARF2
1228 debug info format can represent this, so use of DWARF2 is recommended.
1229 If you are using the stabs debug info format, GCC describes a noncontiguous
1230 complex variable as if it were two separate variables of noncomplex type.
1231 If the variable's actual name is @code{foo}, the two fictitious
1232 variables are named @code{foo$real} and @code{foo$imag}. You can
1233 examine and set these two fictitious variables with your debugger.
1239 ISO C99 supports floating-point numbers written not only in the usual
1240 decimal notation, such as @code{1.55e1}, but also numbers such as
1241 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1242 supports this in C89 mode (except in some cases when strictly
1243 conforming) and in C++. In that format the
1244 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1245 mandatory. The exponent is a decimal number that indicates the power of
1246 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1253 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1254 is the same as @code{1.55e1}.
1256 Unlike for floating-point numbers in the decimal notation the exponent
1257 is always required in the hexadecimal notation. Otherwise the compiler
1258 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1259 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1260 extension for floating-point constants of type @code{float}.
1263 @section Arrays of Length Zero
1264 @cindex arrays of length zero
1265 @cindex zero-length arrays
1266 @cindex length-zero arrays
1267 @cindex flexible array members
1269 Zero-length arrays are allowed in GNU C@. They are very useful as the
1270 last element of a structure which is really a header for a variable-length
1279 struct line *thisline = (struct line *)
1280 malloc (sizeof (struct line) + this_length);
1281 thisline->length = this_length;
1284 In ISO C90, you would have to give @code{contents} a length of 1, which
1285 means either you waste space or complicate the argument to @code{malloc}.
1287 In ISO C99, you would use a @dfn{flexible array member}, which is
1288 slightly different in syntax and semantics:
1292 Flexible array members are written as @code{contents[]} without
1296 Flexible array members have incomplete type, and so the @code{sizeof}
1297 operator may not be applied. As a quirk of the original implementation
1298 of zero-length arrays, @code{sizeof} evaluates to zero.
1301 Flexible array members may only appear as the last member of a
1302 @code{struct} that is otherwise non-empty.
1305 A structure containing a flexible array member, or a union containing
1306 such a structure (possibly recursively), may not be a member of a
1307 structure or an element of an array. (However, these uses are
1308 permitted by GCC as extensions.)
1311 GCC versions before 3.0 allowed zero-length arrays to be statically
1312 initialized, as if they were flexible arrays. In addition to those
1313 cases that were useful, it also allowed initializations in situations
1314 that would corrupt later data. Non-empty initialization of zero-length
1315 arrays is now treated like any case where there are more initializer
1316 elements than the array holds, in that a suitable warning about "excess
1317 elements in array" is given, and the excess elements (all of them, in
1318 this case) are ignored.
1320 Instead GCC allows static initialization of flexible array members.
1321 This is equivalent to defining a new structure containing the original
1322 structure followed by an array of sufficient size to contain the data.
1323 I.e.@: in the following, @code{f1} is constructed as if it were declared
1329 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1332 struct f1 f1; int data[3];
1333 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1337 The convenience of this extension is that @code{f1} has the desired
1338 type, eliminating the need to consistently refer to @code{f2.f1}.
1340 This has symmetry with normal static arrays, in that an array of
1341 unknown size is also written with @code{[]}.
1343 Of course, this extension only makes sense if the extra data comes at
1344 the end of a top-level object, as otherwise we would be overwriting
1345 data at subsequent offsets. To avoid undue complication and confusion
1346 with initialization of deeply nested arrays, we simply disallow any
1347 non-empty initialization except when the structure is the top-level
1348 object. For example:
1351 struct foo @{ int x; int y[]; @};
1352 struct bar @{ struct foo z; @};
1354 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1355 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1356 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1357 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1360 @node Variable Length
1361 @section Arrays of Variable Length
1362 @cindex variable-length arrays
1363 @cindex arrays of variable length
1366 Variable-length automatic arrays are allowed in ISO C99, and as an
1367 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1368 implementation of variable-length arrays does not yet conform in detail
1369 to the ISO C99 standard.) These arrays are
1370 declared like any other automatic arrays, but with a length that is not
1371 a constant expression. The storage is allocated at the point of
1372 declaration and deallocated when the brace-level is exited. For
1377 concat_fopen (char *s1, char *s2, char *mode)
1379 char str[strlen (s1) + strlen (s2) + 1];
1382 return fopen (str, mode);
1386 @cindex scope of a variable length array
1387 @cindex variable-length array scope
1388 @cindex deallocating variable length arrays
1389 Jumping or breaking out of the scope of the array name deallocates the
1390 storage. Jumping into the scope is not allowed; you get an error
1393 @cindex @code{alloca} vs variable-length arrays
1394 You can use the function @code{alloca} to get an effect much like
1395 variable-length arrays. The function @code{alloca} is available in
1396 many other C implementations (but not in all). On the other hand,
1397 variable-length arrays are more elegant.
1399 There are other differences between these two methods. Space allocated
1400 with @code{alloca} exists until the containing @emph{function} returns.
1401 The space for a variable-length array is deallocated as soon as the array
1402 name's scope ends. (If you use both variable-length arrays and
1403 @code{alloca} in the same function, deallocation of a variable-length array
1404 will also deallocate anything more recently allocated with @code{alloca}.)
1406 You can also use variable-length arrays as arguments to functions:
1410 tester (int len, char data[len][len])
1416 The length of an array is computed once when the storage is allocated
1417 and is remembered for the scope of the array in case you access it with
1420 If you want to pass the array first and the length afterward, you can
1421 use a forward declaration in the parameter list---another GNU extension.
1425 tester (int len; char data[len][len], int len)
1431 @cindex parameter forward declaration
1432 The @samp{int len} before the semicolon is a @dfn{parameter forward
1433 declaration}, and it serves the purpose of making the name @code{len}
1434 known when the declaration of @code{data} is parsed.
1436 You can write any number of such parameter forward declarations in the
1437 parameter list. They can be separated by commas or semicolons, but the
1438 last one must end with a semicolon, which is followed by the ``real''
1439 parameter declarations. Each forward declaration must match a ``real''
1440 declaration in parameter name and data type. ISO C99 does not support
1441 parameter forward declarations.
1443 @node Variadic Macros
1444 @section Macros with a Variable Number of Arguments.
1445 @cindex variable number of arguments
1446 @cindex macro with variable arguments
1447 @cindex rest argument (in macro)
1448 @cindex variadic macros
1450 In the ISO C standard of 1999, a macro can be declared to accept a
1451 variable number of arguments much as a function can. The syntax for
1452 defining the macro is similar to that of a function. Here is an
1456 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1459 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1460 such a macro, it represents the zero or more tokens until the closing
1461 parenthesis that ends the invocation, including any commas. This set of
1462 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1463 wherever it appears. See the CPP manual for more information.
1465 GCC has long supported variadic macros, and used a different syntax that
1466 allowed you to give a name to the variable arguments just like any other
1467 argument. Here is an example:
1470 #define debug(format, args...) fprintf (stderr, format, args)
1473 This is in all ways equivalent to the ISO C example above, but arguably
1474 more readable and descriptive.
1476 GNU CPP has two further variadic macro extensions, and permits them to
1477 be used with either of the above forms of macro definition.
1479 In standard C, you are not allowed to leave the variable argument out
1480 entirely; but you are allowed to pass an empty argument. For example,
1481 this invocation is invalid in ISO C, because there is no comma after
1488 GNU CPP permits you to completely omit the variable arguments in this
1489 way. In the above examples, the compiler would complain, though since
1490 the expansion of the macro still has the extra comma after the format
1493 To help solve this problem, CPP behaves specially for variable arguments
1494 used with the token paste operator, @samp{##}. If instead you write
1497 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1500 and if the variable arguments are omitted or empty, the @samp{##}
1501 operator causes the preprocessor to remove the comma before it. If you
1502 do provide some variable arguments in your macro invocation, GNU CPP
1503 does not complain about the paste operation and instead places the
1504 variable arguments after the comma. Just like any other pasted macro
1505 argument, these arguments are not macro expanded.
1507 @node Escaped Newlines
1508 @section Slightly Looser Rules for Escaped Newlines
1509 @cindex escaped newlines
1510 @cindex newlines (escaped)
1512 Recently, the preprocessor has relaxed its treatment of escaped
1513 newlines. Previously, the newline had to immediately follow a
1514 backslash. The current implementation allows whitespace in the form of
1515 spaces, horizontal and vertical tabs, and form feeds between the
1516 backslash and the subsequent newline. The preprocessor issues a
1517 warning, but treats it as a valid escaped newline and combines the two
1518 lines to form a single logical line. This works within comments and
1519 tokens, including multi-line strings, as well as between tokens.
1520 Comments are @emph{not} treated as whitespace for the purposes of this
1521 relaxation, since they have not yet been replaced with spaces.
1523 @node Multi-line Strings
1524 @section String Literals with Embedded Newlines
1525 @cindex multi-line string literals
1527 As an extension, GNU CPP permits string literals to cross multiple lines
1528 without escaping the embedded newlines. Each embedded newline is
1529 replaced with a single @samp{\n} character in the resulting string
1530 literal, regardless of what form the newline took originally.
1532 CPP currently allows such strings in directives as well (other than the
1533 @samp{#include} family). This is deprecated and will eventually be
1537 @section Non-Lvalue Arrays May Have Subscripts
1538 @cindex subscripting
1539 @cindex arrays, non-lvalue
1541 @cindex subscripting and function values
1542 In ISO C99, arrays that are not lvalues still decay to pointers, and
1543 may be subscripted, although they may not be modified or used after
1544 the next sequence point and the unary @samp{&} operator may not be
1545 applied to them. As an extension, GCC allows such arrays to be
1546 subscripted in C89 mode, though otherwise they do not decay to
1547 pointers outside C99 mode. For example,
1548 this is valid in GNU C though not valid in C89:
1552 struct foo @{int a[4];@};
1558 return f().a[index];
1564 @section Arithmetic on @code{void}- and Function-Pointers
1565 @cindex void pointers, arithmetic
1566 @cindex void, size of pointer to
1567 @cindex function pointers, arithmetic
1568 @cindex function, size of pointer to
1570 In GNU C, addition and subtraction operations are supported on pointers to
1571 @code{void} and on pointers to functions. This is done by treating the
1572 size of a @code{void} or of a function as 1.
1574 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1575 and on function types, and returns 1.
1577 @opindex Wpointer-arith
1578 The option @option{-Wpointer-arith} requests a warning if these extensions
1582 @section Non-Constant Initializers
1583 @cindex initializers, non-constant
1584 @cindex non-constant initializers
1586 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1587 automatic variable are not required to be constant expressions in GNU C@.
1588 Here is an example of an initializer with run-time varying elements:
1591 foo (float f, float g)
1593 float beat_freqs[2] = @{ f-g, f+g @};
1598 @node Compound Literals
1599 @section Compound Literals
1600 @cindex constructor expressions
1601 @cindex initializations in expressions
1602 @cindex structures, constructor expression
1603 @cindex expressions, constructor
1604 @cindex compound literals
1605 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1607 ISO C99 supports compound literals. A compound literal looks like
1608 a cast containing an initializer. Its value is an object of the
1609 type specified in the cast, containing the elements specified in
1610 the initializer; it is an lvalue. As an extension, GCC supports
1611 compound literals in C89 mode and in C++.
1613 Usually, the specified type is a structure. Assume that
1614 @code{struct foo} and @code{structure} are declared as shown:
1617 struct foo @{int a; char b[2];@} structure;
1621 Here is an example of constructing a @code{struct foo} with a compound literal:
1624 structure = ((struct foo) @{x + y, 'a', 0@});
1628 This is equivalent to writing the following:
1632 struct foo temp = @{x + y, 'a', 0@};
1637 You can also construct an array. If all the elements of the compound literal
1638 are (made up of) simple constant expressions, suitable for use in
1639 initializers of objects of static storage duration, then the compound
1640 literal can be coerced to a pointer to its first element and used in
1641 such an initializer, as shown here:
1644 char **foo = (char *[]) @{ "x", "y", "z" @};
1647 Compound literals for scalar types and union types are is
1648 also allowed, but then the compound literal is equivalent
1651 As a GNU extension, GCC allows initialization of objects with static storage
1652 duration by compound literals (which is not possible in ISO C99, because
1653 the initializer is not a constant).
1654 It is handled as if the object was initialized only with the bracket
1655 enclosed list if compound literal's and object types match.
1656 The initializer list of the compound literal must be constant.
1657 If the object being initialized has array type of unknown size, the size is
1658 determined by compound literal size.
1661 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1662 static int y[] = (int []) @{1, 2, 3@};
1663 static int z[] = (int [3]) @{1@};
1667 The above lines are equivalent to the following:
1669 static struct foo x = @{1, 'a', 'b'@};
1670 static int y[] = @{1, 2, 3@};
1671 static int z[] = @{1, 0, 0@};
1674 @node Designated Inits
1675 @section Designated Initializers
1676 @cindex initializers with labeled elements
1677 @cindex labeled elements in initializers
1678 @cindex case labels in initializers
1679 @cindex designated initializers
1681 Standard C89 requires the elements of an initializer to appear in a fixed
1682 order, the same as the order of the elements in the array or structure
1685 In ISO C99 you can give the elements in any order, specifying the array
1686 indices or structure field names they apply to, and GNU C allows this as
1687 an extension in C89 mode as well. This extension is not
1688 implemented in GNU C++.
1690 To specify an array index, write
1691 @samp{[@var{index}] =} before the element value. For example,
1694 int a[6] = @{ [4] = 29, [2] = 15 @};
1701 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1705 The index values must be constant expressions, even if the array being
1706 initialized is automatic.
1708 An alternative syntax for this which has been obsolete since GCC 2.5 but
1709 GCC still accepts is to write @samp{[@var{index}]} before the element
1710 value, with no @samp{=}.
1712 To initialize a range of elements to the same value, write
1713 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1714 extension. For example,
1717 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1721 If the value in it has side-effects, the side-effects will happen only once,
1722 not for each initialized field by the range initializer.
1725 Note that the length of the array is the highest value specified
1728 In a structure initializer, specify the name of a field to initialize
1729 with @samp{.@var{fieldname} =} before the element value. For example,
1730 given the following structure,
1733 struct point @{ int x, y; @};
1737 the following initialization
1740 struct point p = @{ .y = yvalue, .x = xvalue @};
1747 struct point p = @{ xvalue, yvalue @};
1750 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1751 @samp{@var{fieldname}:}, as shown here:
1754 struct point p = @{ y: yvalue, x: xvalue @};
1758 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1759 @dfn{designator}. You can also use a designator (or the obsolete colon
1760 syntax) when initializing a union, to specify which element of the union
1761 should be used. For example,
1764 union foo @{ int i; double d; @};
1766 union foo f = @{ .d = 4 @};
1770 will convert 4 to a @code{double} to store it in the union using
1771 the second element. By contrast, casting 4 to type @code{union foo}
1772 would store it into the union as the integer @code{i}, since it is
1773 an integer. (@xref{Cast to Union}.)
1775 You can combine this technique of naming elements with ordinary C
1776 initialization of successive elements. Each initializer element that
1777 does not have a designator applies to the next consecutive element of the
1778 array or structure. For example,
1781 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1788 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1791 Labeling the elements of an array initializer is especially useful
1792 when the indices are characters or belong to an @code{enum} type.
1797 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1798 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1801 @cindex designator lists
1802 You can also write a series of @samp{.@var{fieldname}} and
1803 @samp{[@var{index}]} designators before an @samp{=} to specify a
1804 nested subobject to initialize; the list is taken relative to the
1805 subobject corresponding to the closest surrounding brace pair. For
1806 example, with the @samp{struct point} declaration above:
1809 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1813 If the same field is initialized multiple times, it will have value from
1814 the last initialization. If any such overridden initialization has
1815 side-effect, it is unspecified whether the side-effect happens or not.
1816 Currently, gcc will discard them and issue a warning.
1819 @section Case Ranges
1821 @cindex ranges in case statements
1823 You can specify a range of consecutive values in a single @code{case} label,
1827 case @var{low} ... @var{high}:
1831 This has the same effect as the proper number of individual @code{case}
1832 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1834 This feature is especially useful for ranges of ASCII character codes:
1840 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1841 it may be parsed wrong when you use it with integer values. For example,
1856 @section Cast to a Union Type
1857 @cindex cast to a union
1858 @cindex union, casting to a
1860 A cast to union type is similar to other casts, except that the type
1861 specified is a union type. You can specify the type either with
1862 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1863 a constructor though, not a cast, and hence does not yield an lvalue like
1864 normal casts. (@xref{Compound Literals}.)
1866 The types that may be cast to the union type are those of the members
1867 of the union. Thus, given the following union and variables:
1870 union foo @{ int i; double d; @};
1876 both @code{x} and @code{y} can be cast to type @code{union foo}.
1878 Using the cast as the right-hand side of an assignment to a variable of
1879 union type is equivalent to storing in a member of the union:
1884 u = (union foo) x @equiv{} u.i = x
1885 u = (union foo) y @equiv{} u.d = y
1888 You can also use the union cast as a function argument:
1891 void hack (union foo);
1893 hack ((union foo) x);
1896 @node Mixed Declarations
1897 @section Mixed Declarations and Code
1898 @cindex mixed declarations and code
1899 @cindex declarations, mixed with code
1900 @cindex code, mixed with declarations
1902 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1903 within compound statements. As an extension, GCC also allows this in
1904 C89 mode. For example, you could do:
1913 Each identifier is visible from where it is declared until the end of
1914 the enclosing block.
1916 @node Function Attributes
1917 @section Declaring Attributes of Functions
1918 @cindex function attributes
1919 @cindex declaring attributes of functions
1920 @cindex functions that never return
1921 @cindex functions that have no side effects
1922 @cindex functions in arbitrary sections
1923 @cindex functions that behave like malloc
1924 @cindex @code{volatile} applied to function
1925 @cindex @code{const} applied to function
1926 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1927 @cindex functions with non-null pointer arguments
1928 @cindex functions that are passed arguments in registers on the 386
1929 @cindex functions that pop the argument stack on the 386
1930 @cindex functions that do not pop the argument stack on the 386
1932 In GNU C, you declare certain things about functions called in your program
1933 which help the compiler optimize function calls and check your code more
1936 The keyword @code{__attribute__} allows you to specify special
1937 attributes when making a declaration. This keyword is followed by an
1938 attribute specification inside double parentheses. The following
1939 attributes are currently defined for functions on all targets:
1940 @code{noreturn}, @code{noinline}, @code{always_inline},
1941 @code{pure}, @code{const}, @code{nothrow},
1942 @code{format}, @code{format_arg}, @code{no_instrument_function},
1943 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1944 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1945 @code{alias}, and @code{nonnull}. Several other attributes are defined
1946 for functions on particular target systems. Other attributes, including
1947 @code{section} are supported for variables declarations
1948 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1950 You may also specify attributes with @samp{__} preceding and following
1951 each keyword. This allows you to use them in header files without
1952 being concerned about a possible macro of the same name. For example,
1953 you may use @code{__noreturn__} instead of @code{noreturn}.
1955 @xref{Attribute Syntax}, for details of the exact syntax for using
1959 @cindex @code{noreturn} function attribute
1961 A few standard library functions, such as @code{abort} and @code{exit},
1962 cannot return. GCC knows this automatically. Some programs define
1963 their own functions that never return. You can declare them
1964 @code{noreturn} to tell the compiler this fact. For example,
1968 void fatal () __attribute__ ((noreturn));
1971 fatal (/* @r{@dots{}} */)
1973 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
1979 The @code{noreturn} keyword tells the compiler to assume that
1980 @code{fatal} cannot return. It can then optimize without regard to what
1981 would happen if @code{fatal} ever did return. This makes slightly
1982 better code. More importantly, it helps avoid spurious warnings of
1983 uninitialized variables.
1985 Do not assume that registers saved by the calling function are
1986 restored before calling the @code{noreturn} function.
1988 It does not make sense for a @code{noreturn} function to have a return
1989 type other than @code{void}.
1991 The attribute @code{noreturn} is not implemented in GCC versions
1992 earlier than 2.5. An alternative way to declare that a function does
1993 not return, which works in the current version and in some older
1994 versions, is as follows:
1997 typedef void voidfn ();
1999 volatile voidfn fatal;
2002 @cindex @code{noinline} function attribute
2004 This function attribute prevents a function from being considered for
2007 @cindex @code{always_inline} function attribute
2009 Generally, functions are not inlined unless optimization is specified.
2010 For functions declared inline, this attribute inlines the function even
2011 if no optimization level was specified.
2013 @cindex @code{pure} function attribute
2015 Many functions have no effects except the return value and their
2016 return value depends only on the parameters and/or global variables.
2017 Such a function can be subject
2018 to common subexpression elimination and loop optimization just as an
2019 arithmetic operator would be. These functions should be declared
2020 with the attribute @code{pure}. For example,
2023 int square (int) __attribute__ ((pure));
2027 says that the hypothetical function @code{square} is safe to call
2028 fewer times than the program says.
2030 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2031 Interesting non-pure functions are functions with infinite loops or those
2032 depending on volatile memory or other system resource, that may change between
2033 two consecutive calls (such as @code{feof} in a multithreading environment).
2035 The attribute @code{pure} is not implemented in GCC versions earlier
2037 @cindex @code{const} function attribute
2039 Many functions do not examine any values except their arguments, and
2040 have no effects except the return value. Basically this is just slightly
2041 more strict class than the @code{pure} attribute above, since function is not
2042 allowed to read global memory.
2044 @cindex pointer arguments
2045 Note that a function that has pointer arguments and examines the data
2046 pointed to must @emph{not} be declared @code{const}. Likewise, a
2047 function that calls a non-@code{const} function usually must not be
2048 @code{const}. It does not make sense for a @code{const} function to
2051 The attribute @code{const} is not implemented in GCC versions earlier
2052 than 2.5. An alternative way to declare that a function has no side
2053 effects, which works in the current version and in some older versions,
2057 typedef int intfn ();
2059 extern const intfn square;
2062 This approach does not work in GNU C++ from 2.6.0 on, since the language
2063 specifies that the @samp{const} must be attached to the return value.
2065 @cindex @code{nothrow} function attribute
2067 The @code{nothrow} attribute is used to inform the compiler that a
2068 function cannot throw an exception. For example, most functions in
2069 the standard C library can be guaranteed not to throw an exception
2070 with the notable exceptions of @code{qsort} and @code{bsearch} that
2071 take function pointer arguments. The @code{nothrow} attribute is not
2072 implemented in GCC versions earlier than 3.2.
2074 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2075 @cindex @code{format} function attribute
2077 The @code{format} attribute specifies that a function takes @code{printf},
2078 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2079 should be type-checked against a format string. For example, the
2084 my_printf (void *my_object, const char *my_format, ...)
2085 __attribute__ ((format (printf, 2, 3)));
2089 causes the compiler to check the arguments in calls to @code{my_printf}
2090 for consistency with the @code{printf} style format string argument
2093 The parameter @var{archetype} determines how the format string is
2094 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
2095 or @code{strfmon}. (You can also use @code{__printf__},
2096 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
2097 parameter @var{string-index} specifies which argument is the format
2098 string argument (starting from 1), while @var{first-to-check} is the
2099 number of the first argument to check against the format string. For
2100 functions where the arguments are not available to be checked (such as
2101 @code{vprintf}), specify the third parameter as zero. In this case the
2102 compiler only checks the format string for consistency. For
2103 @code{strftime} formats, the third parameter is required to be zero.
2105 In the example above, the format string (@code{my_format}) is the second
2106 argument of the function @code{my_print}, and the arguments to check
2107 start with the third argument, so the correct parameters for the format
2108 attribute are 2 and 3.
2110 @opindex ffreestanding
2111 The @code{format} attribute allows you to identify your own functions
2112 which take format strings as arguments, so that GCC can check the
2113 calls to these functions for errors. The compiler always (unless
2114 @option{-ffreestanding} is used) checks formats
2115 for the standard library functions @code{printf}, @code{fprintf},
2116 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2117 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2118 warnings are requested (using @option{-Wformat}), so there is no need to
2119 modify the header file @file{stdio.h}. In C99 mode, the functions
2120 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2121 @code{vsscanf} are also checked. Except in strictly conforming C
2122 standard modes, the X/Open function @code{strfmon} is also checked as
2123 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2124 @xref{C Dialect Options,,Options Controlling C Dialect}.
2126 @item format_arg (@var{string-index})
2127 @cindex @code{format_arg} function attribute
2128 @opindex Wformat-nonliteral
2129 The @code{format_arg} attribute specifies that a function takes a format
2130 string for a @code{printf}, @code{scanf}, @code{strftime} or
2131 @code{strfmon} style function and modifies it (for example, to translate
2132 it into another language), so the result can be passed to a
2133 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2134 function (with the remaining arguments to the format function the same
2135 as they would have been for the unmodified string). For example, the
2140 my_dgettext (char *my_domain, const char *my_format)
2141 __attribute__ ((format_arg (2)));
2145 causes the compiler to check the arguments in calls to a @code{printf},
2146 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2147 format string argument is a call to the @code{my_dgettext} function, for
2148 consistency with the format string argument @code{my_format}. If the
2149 @code{format_arg} attribute had not been specified, all the compiler
2150 could tell in such calls to format functions would be that the format
2151 string argument is not constant; this would generate a warning when
2152 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2153 without the attribute.
2155 The parameter @var{string-index} specifies which argument is the format
2156 string argument (starting from 1).
2158 The @code{format-arg} attribute allows you to identify your own
2159 functions which modify format strings, so that GCC can check the
2160 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2161 type function whose operands are a call to one of your own function.
2162 The compiler always treats @code{gettext}, @code{dgettext}, and
2163 @code{dcgettext} in this manner except when strict ISO C support is
2164 requested by @option{-ansi} or an appropriate @option{-std} option, or
2165 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
2166 Controlling C Dialect}.
2168 @item nonnull (@var{arg-index}, @dots{})
2169 @cindex @code{nonnull} function attribute
2170 The @code{nonnull} attribute specifies that some function parameters should
2171 be non-null pointers. For instance, the declaration:
2175 my_memcpy (void *dest, const void *src, size_t len)
2176 __attribute__((nonnull (1, 2)));
2180 causes the compiler to check that, in calls to @code{my_memcpy},
2181 arguments @var{dest} and @var{src} are non-null. If the compiler
2182 determines that a null pointer is passed in an argument slot marked
2183 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2184 is issued. The compiler may also choose to make optimizations based
2185 on the knowledge that certain function arguments will not be null.
2187 If no argument index list is given to the @code{nonnull} attribute,
2188 all pointer arguments are marked as non-null. To illustrate, the
2189 following declaration is equivalent to the previous example:
2193 my_memcpy (void *dest, const void *src, size_t len)
2194 __attribute__((nonnull));
2197 @item no_instrument_function
2198 @cindex @code{no_instrument_function} function attribute
2199 @opindex finstrument-functions
2200 If @option{-finstrument-functions} is given, profiling function calls will
2201 be generated at entry and exit of most user-compiled functions.
2202 Functions with this attribute will not be so instrumented.
2204 @item section ("@var{section-name}")
2205 @cindex @code{section} function attribute
2206 Normally, the compiler places the code it generates in the @code{text} section.
2207 Sometimes, however, you need additional sections, or you need certain
2208 particular functions to appear in special sections. The @code{section}
2209 attribute specifies that a function lives in a particular section.
2210 For example, the declaration:
2213 extern void foobar (void) __attribute__ ((section ("bar")));
2217 puts the function @code{foobar} in the @code{bar} section.
2219 Some file formats do not support arbitrary sections so the @code{section}
2220 attribute is not available on all platforms.
2221 If you need to map the entire contents of a module to a particular
2222 section, consider using the facilities of the linker instead.
2226 @cindex @code{constructor} function attribute
2227 @cindex @code{destructor} function attribute
2228 The @code{constructor} attribute causes the function to be called
2229 automatically before execution enters @code{main ()}. Similarly, the
2230 @code{destructor} attribute causes the function to be called
2231 automatically after @code{main ()} has completed or @code{exit ()} has
2232 been called. Functions with these attributes are useful for
2233 initializing data that will be used implicitly during the execution of
2236 These attributes are not currently implemented for Objective-C@.
2238 @cindex @code{unused} attribute.
2240 This attribute, attached to a function, means that the function is meant
2241 to be possibly unused. GCC will not produce a warning for this
2242 function. GNU C++ does not currently support this attribute as
2243 definitions without parameters are valid in C++.
2245 @cindex @code{used} attribute.
2247 This attribute, attached to a function, means that code must be emitted
2248 for the function even if it appears that the function is not referenced.
2249 This is useful, for example, when the function is referenced only in
2252 @cindex @code{deprecated} attribute.
2254 The @code{deprecated} attribute results in a warning if the function
2255 is used anywhere in the source file. This is useful when identifying
2256 functions that are expected to be removed in a future version of a
2257 program. The warning also includes the location of the declaration
2258 of the deprecated function, to enable users to easily find further
2259 information about why the function is deprecated, or what they should
2260 do instead. Note that the warnings only occurs for uses:
2263 int old_fn () __attribute__ ((deprecated));
2265 int (*fn_ptr)() = old_fn;
2268 results in a warning on line 3 but not line 2.
2270 The @code{deprecated} attribute can also be used for variables and
2271 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2274 @cindex @code{weak} attribute
2275 The @code{weak} attribute causes the declaration to be emitted as a weak
2276 symbol rather than a global. This is primarily useful in defining
2277 library functions which can be overridden in user code, though it can
2278 also be used with non-function declarations. Weak symbols are supported
2279 for ELF targets, and also for a.out targets when using the GNU assembler
2283 @cindex @code{malloc} attribute
2284 The @code{malloc} attribute is used to tell the compiler that a function
2285 may be treated as if it were the malloc function. The compiler assumes
2286 that calls to malloc result in a pointers that cannot alias anything.
2287 This will often improve optimization.
2289 @item alias ("@var{target}")
2290 @cindex @code{alias} attribute
2291 The @code{alias} attribute causes the declaration to be emitted as an
2292 alias for another symbol, which must be specified. For instance,
2295 void __f () @{ /* @r{Do something.} */; @}
2296 void f () __attribute__ ((weak, alias ("__f")));
2299 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
2300 mangled name for the target must be used.
2302 Not all target machines support this attribute.
2304 @item visibility ("@var{visibility_type}")
2305 @cindex @code{visibility} attribute
2306 The @code{visibility} attribute on ELF targets causes the declaration
2307 to be emitted with default, hidden, protected or internal visibility.
2310 void __attribute__ ((visibility ("protected")))
2311 f () @{ /* @r{Do something.} */; @}
2312 int i __attribute__ ((visibility ("hidden")));
2315 See the ELF gABI for complete details, but the short story is
2319 Default visibility is the normal case for ELF. This value is
2320 available for the visibiliy attribute to override other options
2321 that may change the assumed visibility of symbols.
2324 Hidden visibility indicates that the symbol will not be placed into
2325 the dynamic symbol table, so no other @dfn{module} (executable or
2326 shared library) can reference it directly.
2329 Protected visibility indicates that the symbol will be placed in the
2330 dynamic symbol table, but that references within the defining module
2331 will bind to the local symbol. That is, the symbol cannot be overridden
2335 Internal visibility is like hidden visibility, but with additional
2336 processor specific semantics. Unless otherwise specified by the psABI,
2337 gcc defines internal visibility to mean that the function is @emph{never}
2338 called from another module. Note that hidden symbols, while then cannot
2339 be referenced directly by other modules, can be referenced indirectly via
2340 function pointers. By indicating that a symbol cannot be called from
2341 outside the module, gcc may for instance omit the load of a PIC register
2342 since it is known that the calling function loaded the correct value.
2345 Not all ELF targets support this attribute.
2347 @item tls_model ("@var{tls_model}")
2348 @cindex @code{tls_model} attribute
2349 The @code{tls_model} attribute sets thread-local storage model
2350 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
2351 overriding @code{-ftls-model=} command line switch on a per-variable
2353 The @var{tls_model} argument should be one of @code{global-dynamic},
2354 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
2356 @item regparm (@var{number})
2357 @cindex functions that are passed arguments in registers on the 386
2358 On the Intel 386, the @code{regparm} attribute causes the compiler to
2359 pass up to @var{number} integer arguments in registers EAX,
2360 EDX, and ECX instead of on the stack. Functions that take a
2361 variable number of arguments will continue to be passed all of their
2362 arguments on the stack.
2365 @cindex functions that pop the argument stack on the 386
2366 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2367 assume that the called function will pop off the stack space used to
2368 pass arguments, unless it takes a variable number of arguments.
2370 The PowerPC compiler for Windows NT currently ignores the @code{stdcall}
2374 @cindex functions that pop the argument stack on the 386
2375 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2376 pass the first two arguments in the registers ECX and EDX. Subsequent
2377 arguments are passed on the stack. The called function will pop the
2378 arguments off the stack. If the number of arguments is variable all
2379 arguments are pushed on the stack.
2382 @cindex functions that do pop the argument stack on the 386
2384 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2385 assume that the calling function will pop off the stack space used to
2386 pass arguments. This is
2387 useful to override the effects of the @option{-mrtd} switch.
2389 The PowerPC compiler for Windows NT currently ignores the @code{cdecl}
2392 @item longcall/shortcall
2393 @cindex functions called via pointer on the RS/6000 and PowerPC
2394 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
2395 compiler to always call this function via a pointer, just as it would if
2396 the @option{-mlongcall} option had been specified. The @code{shortcall}
2397 attribute causes the compiler not to do this. These attributes override
2398 both the @option{-mlongcall} switch and the @code{#pragma longcall}
2401 @xref{RS/6000 and PowerPC Options}, for more information on when long
2402 calls are and are not necessary.
2404 @item long_call/short_call
2405 @cindex indirect calls on ARM
2406 This attribute allows to specify how to call a particular function on
2407 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2408 command line switch and @code{#pragma long_calls} settings. The
2409 @code{long_call} attribute causes the compiler to always call the
2410 function by first loading its address into a register and then using the
2411 contents of that register. The @code{short_call} attribute always places
2412 the offset to the function from the call site into the @samp{BL}
2413 instruction directly.
2416 @cindex functions which are imported from a dll on PowerPC Windows NT
2417 On the PowerPC running Windows NT, the @code{dllimport} attribute causes
2418 the compiler to call the function via a global pointer to the function
2419 pointer that is set up by the Windows NT dll library. The pointer name
2420 is formed by combining @code{__imp_} and the function name.
2423 @cindex functions which are exported from a dll on PowerPC Windows NT
2424 On the PowerPC running Windows NT, the @code{dllexport} attribute causes
2425 the compiler to provide a global pointer to the function pointer, so
2426 that it can be called with the @code{dllimport} attribute. The pointer
2427 name is formed by combining @code{__imp_} and the function name.
2429 @item exception (@var{except-func} [, @var{except-arg}])
2430 @cindex functions which specify exception handling on PowerPC Windows NT
2431 On the PowerPC running Windows NT, the @code{exception} attribute causes
2432 the compiler to modify the structured exception table entry it emits for
2433 the declared function. The string or identifier @var{except-func} is
2434 placed in the third entry of the structured exception table. It
2435 represents a function, which is called by the exception handling
2436 mechanism if an exception occurs. If it was specified, the string or
2437 identifier @var{except-arg} is placed in the fourth entry of the
2438 structured exception table.
2440 @item function_vector
2441 @cindex calling functions through the function vector on the H8/300 processors
2442 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2443 function should be called through the function vector. Calling a
2444 function through the function vector will reduce code size, however;
2445 the function vector has a limited size (maximum 128 entries on the H8/300
2446 and 64 entries on the H8/300H) and shares space with the interrupt vector.
2448 You must use GAS and GLD from GNU binutils version 2.7 or later for
2449 this attribute to work correctly.
2452 @cindex interrupt handler functions
2453 Use this attribute on the ARM, AVR, C4x, M32R/D and Xstormy16 ports to indicate
2454 that the specified function is an interrupt handler. The compiler will
2455 generate function entry and exit sequences suitable for use in an
2456 interrupt handler when this attribute is present.
2458 Note, interrupt handlers for the H8/300, H8/300H and SH processors can
2459 be specified via the @code{interrupt_handler} attribute.
2461 Note, on the AVR interrupts will be enabled inside the function.
2463 Note, for the ARM you can specify the kind of interrupt to be handled by
2464 adding an optional parameter to the interrupt attribute like this:
2467 void f () __attribute__ ((interrupt ("IRQ")));
2470 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2472 @item interrupt_handler
2473 @cindex interrupt handler functions on the H8/300 and SH processors
2474 Use this attribute on the H8/300, H8/300H and SH to indicate that the
2475 specified function is an interrupt handler. The compiler will generate
2476 function entry and exit sequences suitable for use in an interrupt
2477 handler when this attribute is present.
2480 Use this attribute on the SH to indicate an @code{interrupt_handler}
2481 function should switch to an alternate stack. It expects a string
2482 argument that names a global variable holding the address of the
2487 void f () __attribute__ ((interrupt_handler,
2488 sp_switch ("alt_stack")));
2492 Use this attribute on the SH for an @code{interrupt_handle} to return using
2493 @code{trapa} instead of @code{rte}. This attribute expects an integer
2494 argument specifying the trap number to be used.
2497 @cindex eight bit data on the H8/300 and H8/300H
2498 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2499 variable should be placed into the eight bit data section.
2500 The compiler will generate more efficient code for certain operations
2501 on data in the eight bit data area. Note the eight bit data area is limited to
2504 You must use GAS and GLD from GNU binutils version 2.7 or later for
2505 this attribute to work correctly.
2508 @cindex tiny data section on the H8/300H
2509 Use this attribute on the H8/300H to indicate that the specified
2510 variable should be placed into the tiny data section.
2511 The compiler will generate more efficient code for loads and stores
2512 on data in the tiny data section. Note the tiny data area is limited to
2513 slightly under 32kbytes of data.
2516 @cindex signal handler functions on the AVR processors
2517 Use this attribute on the AVR to indicate that the specified
2518 function is an signal handler. The compiler will generate function
2519 entry and exit sequences suitable for use in an signal handler when this
2520 attribute is present. Interrupts will be disabled inside function.
2523 @cindex function without a prologue/epilogue code
2524 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate that the
2525 specified function do not need prologue/epilogue sequences generated by
2526 the compiler. It is up to the programmer to provide these sequences.
2528 @item model (@var{model-name})
2529 @cindex function addressability on the M32R/D
2530 Use this attribute on the M32R/D to set the addressability of an object,
2531 and the code generated for a function.
2532 The identifier @var{model-name} is one of @code{small}, @code{medium},
2533 or @code{large}, representing each of the code models.
2535 Small model objects live in the lower 16MB of memory (so that their
2536 addresses can be loaded with the @code{ld24} instruction), and are
2537 callable with the @code{bl} instruction.
2539 Medium model objects may live anywhere in the 32-bit address space (the
2540 compiler will generate @code{seth/add3} instructions to load their addresses),
2541 and are callable with the @code{bl} instruction.
2543 Large model objects may live anywhere in the 32-bit address space (the
2544 compiler will generate @code{seth/add3} instructions to load their addresses),
2545 and may not be reachable with the @code{bl} instruction (the compiler will
2546 generate the much slower @code{seth/add3/jl} instruction sequence).
2550 You can specify multiple attributes in a declaration by separating them
2551 by commas within the double parentheses or by immediately following an
2552 attribute declaration with another attribute declaration.
2554 @cindex @code{#pragma}, reason for not using
2555 @cindex pragma, reason for not using
2556 Some people object to the @code{__attribute__} feature, suggesting that
2557 ISO C's @code{#pragma} should be used instead. At the time
2558 @code{__attribute__} was designed, there were two reasons for not doing
2563 It is impossible to generate @code{#pragma} commands from a macro.
2566 There is no telling what the same @code{#pragma} might mean in another
2570 These two reasons applied to almost any application that might have been
2571 proposed for @code{#pragma}. It was basically a mistake to use
2572 @code{#pragma} for @emph{anything}.
2574 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2575 to be generated from macros. In addition, a @code{#pragma GCC}
2576 namespace is now in use for GCC-specific pragmas. However, it has been
2577 found convenient to use @code{__attribute__} to achieve a natural
2578 attachment of attributes to their corresponding declarations, whereas
2579 @code{#pragma GCC} is of use for constructs that do not naturally form
2580 part of the grammar. @xref{Other Directives,,Miscellaneous
2581 Preprocessing Directives, cpp, The C Preprocessor}.
2583 @node Attribute Syntax
2584 @section Attribute Syntax
2585 @cindex attribute syntax
2587 This section describes the syntax with which @code{__attribute__} may be
2588 used, and the constructs to which attribute specifiers bind, for the C
2589 language. Some details may vary for C++ and Objective-C@. Because of
2590 infelicities in the grammar for attributes, some forms described here
2591 may not be successfully parsed in all cases.
2593 There are some problems with the semantics of attributes in C++. For
2594 example, there are no manglings for attributes, although they may affect
2595 code generation, so problems may arise when attributed types are used in
2596 conjunction with templates or overloading. Similarly, @code{typeid}
2597 does not distinguish between types with different attributes. Support
2598 for attributes in C++ may be restricted in future to attributes on
2599 declarations only, but not on nested declarators.
2601 @xref{Function Attributes}, for details of the semantics of attributes
2602 applying to functions. @xref{Variable Attributes}, for details of the
2603 semantics of attributes applying to variables. @xref{Type Attributes},
2604 for details of the semantics of attributes applying to structure, union
2605 and enumerated types.
2607 An @dfn{attribute specifier} is of the form
2608 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2609 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2610 each attribute is one of the following:
2614 Empty. Empty attributes are ignored.
2617 A word (which may be an identifier such as @code{unused}, or a reserved
2618 word such as @code{const}).
2621 A word, followed by, in parentheses, parameters for the attribute.
2622 These parameters take one of the following forms:
2626 An identifier. For example, @code{mode} attributes use this form.
2629 An identifier followed by a comma and a non-empty comma-separated list
2630 of expressions. For example, @code{format} attributes use this form.
2633 A possibly empty comma-separated list of expressions. For example,
2634 @code{format_arg} attributes use this form with the list being a single
2635 integer constant expression, and @code{alias} attributes use this form
2636 with the list being a single string constant.
2640 An @dfn{attribute specifier list} is a sequence of one or more attribute
2641 specifiers, not separated by any other tokens.
2643 An attribute specifier list may appear after the colon following a
2644 label, other than a @code{case} or @code{default} label. The only
2645 attribute it makes sense to use after a label is @code{unused}. This
2646 feature is intended for code generated by programs which contains labels
2647 that may be unused but which is compiled with @option{-Wall}. It would
2648 not normally be appropriate to use in it human-written code, though it
2649 could be useful in cases where the code that jumps to the label is
2650 contained within an @code{#ifdef} conditional.
2652 An attribute specifier list may appear as part of a @code{struct},
2653 @code{union} or @code{enum} specifier. It may go either immediately
2654 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2655 the closing brace. It is ignored if the content of the structure, union
2656 or enumerated type is not defined in the specifier in which the
2657 attribute specifier list is used---that is, in usages such as
2658 @code{struct __attribute__((foo)) bar} with no following opening brace.
2659 Where attribute specifiers follow the closing brace, they are considered
2660 to relate to the structure, union or enumerated type defined, not to any
2661 enclosing declaration the type specifier appears in, and the type
2662 defined is not complete until after the attribute specifiers.
2663 @c Otherwise, there would be the following problems: a shift/reduce
2664 @c conflict between attributes binding the struct/union/enum and
2665 @c binding to the list of specifiers/qualifiers; and "aligned"
2666 @c attributes could use sizeof for the structure, but the size could be
2667 @c changed later by "packed" attributes.
2669 Otherwise, an attribute specifier appears as part of a declaration,
2670 counting declarations of unnamed parameters and type names, and relates
2671 to that declaration (which may be nested in another declaration, for
2672 example in the case of a parameter declaration), or to a particular declarator
2673 within a declaration. Where an
2674 attribute specifier is applied to a parameter declared as a function or
2675 an array, it should apply to the function or array rather than the
2676 pointer to which the parameter is implicitly converted, but this is not
2677 yet correctly implemented.
2679 Any list of specifiers and qualifiers at the start of a declaration may
2680 contain attribute specifiers, whether or not such a list may in that
2681 context contain storage class specifiers. (Some attributes, however,
2682 are essentially in the nature of storage class specifiers, and only make
2683 sense where storage class specifiers may be used; for example,
2684 @code{section}.) There is one necessary limitation to this syntax: the
2685 first old-style parameter declaration in a function definition cannot
2686 begin with an attribute specifier, because such an attribute applies to
2687 the function instead by syntax described below (which, however, is not
2688 yet implemented in this case). In some other cases, attribute
2689 specifiers are permitted by this grammar but not yet supported by the
2690 compiler. All attribute specifiers in this place relate to the
2691 declaration as a whole. In the obsolescent usage where a type of
2692 @code{int} is implied by the absence of type specifiers, such a list of
2693 specifiers and qualifiers may be an attribute specifier list with no
2694 other specifiers or qualifiers.
2696 An attribute specifier list may appear immediately before a declarator
2697 (other than the first) in a comma-separated list of declarators in a
2698 declaration of more than one identifier using a single list of
2699 specifiers and qualifiers. Such attribute specifiers apply
2700 only to the identifier before whose declarator they appear. For
2704 __attribute__((noreturn)) void d0 (void),
2705 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2710 the @code{noreturn} attribute applies to all the functions
2711 declared; the @code{format} attribute only applies to @code{d1}.
2713 An attribute specifier list may appear immediately before the comma,
2714 @code{=} or semicolon terminating the declaration of an identifier other
2715 than a function definition. At present, such attribute specifiers apply
2716 to the declared object or function, but in future they may attach to the
2717 outermost adjacent declarator. In simple cases there is no difference,
2718 but, for example, in
2721 void (****f)(void) __attribute__((noreturn));
2725 at present the @code{noreturn} attribute applies to @code{f}, which
2726 causes a warning since @code{f} is not a function, but in future it may
2727 apply to the function @code{****f}. The precise semantics of what
2728 attributes in such cases will apply to are not yet specified. Where an
2729 assembler name for an object or function is specified (@pxref{Asm
2730 Labels}), at present the attribute must follow the @code{asm}
2731 specification; in future, attributes before the @code{asm} specification
2732 may apply to the adjacent declarator, and those after it to the declared
2735 An attribute specifier list may, in future, be permitted to appear after
2736 the declarator in a function definition (before any old-style parameter
2737 declarations or the function body).
2739 Attribute specifiers may be mixed with type qualifiers appearing inside
2740 the @code{[]} of a parameter array declarator, in the C99 construct by
2741 which such qualifiers are applied to the pointer to which the array is
2742 implicitly converted. Such attribute specifiers apply to the pointer,
2743 not to the array, but at present this is not implemented and they are
2746 An attribute specifier list may appear at the start of a nested
2747 declarator. At present, there are some limitations in this usage: the
2748 attributes correctly apply to the declarator, but for most individual
2749 attributes the semantics this implies are not implemented.
2750 When attribute specifiers follow the @code{*} of a pointer
2751 declarator, they may be mixed with any type qualifiers present.
2752 The following describes the formal semantics of this syntax. It will make the
2753 most sense if you are familiar with the formal specification of
2754 declarators in the ISO C standard.
2756 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2757 D1}, where @code{T} contains declaration specifiers that specify a type
2758 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2759 contains an identifier @var{ident}. The type specified for @var{ident}
2760 for derived declarators whose type does not include an attribute
2761 specifier is as in the ISO C standard.
2763 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2764 and the declaration @code{T D} specifies the type
2765 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2766 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2767 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2769 If @code{D1} has the form @code{*
2770 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2771 declaration @code{T D} specifies the type
2772 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2773 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2774 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2780 void (__attribute__((noreturn)) ****f) (void);
2784 specifies the type ``pointer to pointer to pointer to pointer to
2785 non-returning function returning @code{void}''. As another example,
2788 char *__attribute__((aligned(8))) *f;
2792 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2793 Note again that this does not work with most attributes; for example,
2794 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2795 is not yet supported.
2797 For compatibility with existing code written for compiler versions that
2798 did not implement attributes on nested declarators, some laxity is
2799 allowed in the placing of attributes. If an attribute that only applies
2800 to types is applied to a declaration, it will be treated as applying to
2801 the type of that declaration. If an attribute that only applies to
2802 declarations is applied to the type of a declaration, it will be treated
2803 as applying to that declaration; and, for compatibility with code
2804 placing the attributes immediately before the identifier declared, such
2805 an attribute applied to a function return type will be treated as
2806 applying to the function type, and such an attribute applied to an array
2807 element type will be treated as applying to the array type. If an
2808 attribute that only applies to function types is applied to a
2809 pointer-to-function type, it will be treated as applying to the pointer
2810 target type; if such an attribute is applied to a function return type
2811 that is not a pointer-to-function type, it will be treated as applying
2812 to the function type.
2814 @node Function Prototypes
2815 @section Prototypes and Old-Style Function Definitions
2816 @cindex function prototype declarations
2817 @cindex old-style function definitions
2818 @cindex promotion of formal parameters
2820 GNU C extends ISO C to allow a function prototype to override a later
2821 old-style non-prototype definition. Consider the following example:
2824 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2831 /* @r{Prototype function declaration.} */
2832 int isroot P((uid_t));
2834 /* @r{Old-style function definition.} */
2836 isroot (x) /* ??? lossage here ??? */
2843 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2844 not allow this example, because subword arguments in old-style
2845 non-prototype definitions are promoted. Therefore in this example the
2846 function definition's argument is really an @code{int}, which does not
2847 match the prototype argument type of @code{short}.
2849 This restriction of ISO C makes it hard to write code that is portable
2850 to traditional C compilers, because the programmer does not know
2851 whether the @code{uid_t} type is @code{short}, @code{int}, or
2852 @code{long}. Therefore, in cases like these GNU C allows a prototype
2853 to override a later old-style definition. More precisely, in GNU C, a
2854 function prototype argument type overrides the argument type specified
2855 by a later old-style definition if the former type is the same as the
2856 latter type before promotion. Thus in GNU C the above example is
2857 equivalent to the following:
2870 GNU C++ does not support old-style function definitions, so this
2871 extension is irrelevant.
2874 @section C++ Style Comments
2876 @cindex C++ comments
2877 @cindex comments, C++ style
2879 In GNU C, you may use C++ style comments, which start with @samp{//} and
2880 continue until the end of the line. Many other C implementations allow
2881 such comments, and they are included in the 1999 C standard. However,
2882 C++ style comments are not recognized if you specify an @option{-std}
2883 option specifying a version of ISO C before C99, or @option{-ansi}
2884 (equivalent to @option{-std=c89}).
2887 @section Dollar Signs in Identifier Names
2889 @cindex dollar signs in identifier names
2890 @cindex identifier names, dollar signs in
2892 In GNU C, you may normally use dollar signs in identifier names.
2893 This is because many traditional C implementations allow such identifiers.
2894 However, dollar signs in identifiers are not supported on a few target
2895 machines, typically because the target assembler does not allow them.
2897 @node Character Escapes
2898 @section The Character @key{ESC} in Constants
2900 You can use the sequence @samp{\e} in a string or character constant to
2901 stand for the ASCII character @key{ESC}.
2904 @section Inquiring on Alignment of Types or Variables
2906 @cindex type alignment
2907 @cindex variable alignment
2909 The keyword @code{__alignof__} allows you to inquire about how an object
2910 is aligned, or the minimum alignment usually required by a type. Its
2911 syntax is just like @code{sizeof}.
2913 For example, if the target machine requires a @code{double} value to be
2914 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2915 This is true on many RISC machines. On more traditional machine
2916 designs, @code{__alignof__ (double)} is 4 or even 2.
2918 Some machines never actually require alignment; they allow reference to any
2919 data type even at an odd addresses. For these machines, @code{__alignof__}
2920 reports the @emph{recommended} alignment of a type.
2922 If the operand of @code{__alignof__} is an lvalue rather than a type,
2923 its value is the required alignment for its type, taking into account
2924 any minimum alignment specified with GCC's @code{__attribute__}
2925 extension (@pxref{Variable Attributes}). For example, after this
2929 struct foo @{ int x; char y; @} foo1;
2933 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
2934 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
2936 It is an error to ask for the alignment of an incomplete type.
2938 @node Variable Attributes
2939 @section Specifying Attributes of Variables
2940 @cindex attribute of variables
2941 @cindex variable attributes
2943 The keyword @code{__attribute__} allows you to specify special
2944 attributes of variables or structure fields. This keyword is followed
2945 by an attribute specification inside double parentheses. Ten
2946 attributes are currently defined for variables: @code{aligned},
2947 @code{mode}, @code{nocommon}, @code{packed}, @code{section},
2948 @code{transparent_union}, @code{unused}, @code{deprecated},
2949 @code{vector_size}, and @code{weak}. Some other attributes are defined
2950 for variables on particular target systems. Other attributes are
2951 available for functions (@pxref{Function Attributes}) and for types
2952 (@pxref{Type Attributes}). Other front ends might define more
2953 attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2955 You may also specify attributes with @samp{__} preceding and following
2956 each keyword. This allows you to use them in header files without
2957 being concerned about a possible macro of the same name. For example,
2958 you may use @code{__aligned__} instead of @code{aligned}.
2960 @xref{Attribute Syntax}, for details of the exact syntax for using
2964 @cindex @code{aligned} attribute
2965 @item aligned (@var{alignment})
2966 This attribute specifies a minimum alignment for the variable or
2967 structure field, measured in bytes. For example, the declaration:
2970 int x __attribute__ ((aligned (16))) = 0;
2974 causes the compiler to allocate the global variable @code{x} on a
2975 16-byte boundary. On a 68040, this could be used in conjunction with
2976 an @code{asm} expression to access the @code{move16} instruction which
2977 requires 16-byte aligned operands.
2979 You can also specify the alignment of structure fields. For example, to
2980 create a double-word aligned @code{int} pair, you could write:
2983 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2987 This is an alternative to creating a union with a @code{double} member
2988 that forces the union to be double-word aligned.
2990 As in the preceding examples, you can explicitly specify the alignment
2991 (in bytes) that you wish the compiler to use for a given variable or
2992 structure field. Alternatively, you can leave out the alignment factor
2993 and just ask the compiler to align a variable or field to the maximum
2994 useful alignment for the target machine you are compiling for. For
2995 example, you could write:
2998 short array[3] __attribute__ ((aligned));
3001 Whenever you leave out the alignment factor in an @code{aligned} attribute
3002 specification, the compiler automatically sets the alignment for the declared
3003 variable or field to the largest alignment which is ever used for any data
3004 type on the target machine you are compiling for. Doing this can often make
3005 copy operations more efficient, because the compiler can use whatever
3006 instructions copy the biggest chunks of memory when performing copies to
3007 or from the variables or fields that you have aligned this way.
3009 The @code{aligned} attribute can only increase the alignment; but you
3010 can decrease it by specifying @code{packed} as well. See below.
3012 Note that the effectiveness of @code{aligned} attributes may be limited
3013 by inherent limitations in your linker. On many systems, the linker is
3014 only able to arrange for variables to be aligned up to a certain maximum
3015 alignment. (For some linkers, the maximum supported alignment may
3016 be very very small.) If your linker is only able to align variables
3017 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3018 in an @code{__attribute__} will still only provide you with 8 byte
3019 alignment. See your linker documentation for further information.
3021 @item mode (@var{mode})
3022 @cindex @code{mode} attribute
3023 This attribute specifies the data type for the declaration---whichever
3024 type corresponds to the mode @var{mode}. This in effect lets you
3025 request an integer or floating point type according to its width.
3027 You may also specify a mode of @samp{byte} or @samp{__byte__} to
3028 indicate the mode corresponding to a one-byte integer, @samp{word} or
3029 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
3030 or @samp{__pointer__} for the mode used to represent pointers.
3033 @cindex @code{nocommon} attribute
3035 This attribute specifies requests GCC not to place a variable
3036 ``common'' but instead to allocate space for it directly. If you
3037 specify the @option{-fno-common} flag, GCC will do this for all
3040 Specifying the @code{nocommon} attribute for a variable provides an
3041 initialization of zeros. A variable may only be initialized in one
3045 @cindex @code{packed} attribute
3046 The @code{packed} attribute specifies that a variable or structure field
3047 should have the smallest possible alignment---one byte for a variable,
3048 and one bit for a field, unless you specify a larger value with the
3049 @code{aligned} attribute.
3051 Here is a structure in which the field @code{x} is packed, so that it
3052 immediately follows @code{a}:
3058 int x[2] __attribute__ ((packed));
3062 @item section ("@var{section-name}")
3063 @cindex @code{section} variable attribute
3064 Normally, the compiler places the objects it generates in sections like
3065 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
3066 or you need certain particular variables to appear in special sections,
3067 for example to map to special hardware. The @code{section}
3068 attribute specifies that a variable (or function) lives in a particular
3069 section. For example, this small program uses several specific section names:
3072 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
3073 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
3074 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
3075 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3079 /* Initialize stack pointer */
3080 init_sp (stack + sizeof (stack));
3082 /* Initialize initialized data */
3083 memcpy (&init_data, &data, &edata - &data);
3085 /* Turn on the serial ports */
3092 Use the @code{section} attribute with an @emph{initialized} definition
3093 of a @emph{global} variable, as shown in the example. GCC issues
3094 a warning and otherwise ignores the @code{section} attribute in
3095 uninitialized variable declarations.
3097 You may only use the @code{section} attribute with a fully initialized
3098 global definition because of the way linkers work. The linker requires
3099 each object be defined once, with the exception that uninitialized
3100 variables tentatively go in the @code{common} (or @code{bss}) section
3101 and can be multiply ``defined''. You can force a variable to be
3102 initialized with the @option{-fno-common} flag or the @code{nocommon}
3105 Some file formats do not support arbitrary sections so the @code{section}
3106 attribute is not available on all platforms.
3107 If you need to map the entire contents of a module to a particular
3108 section, consider using the facilities of the linker instead.
3111 @cindex @code{shared} variable attribute
3112 On Windows NT, in addition to putting variable definitions in a named
3113 section, the section can also be shared among all running copies of an
3114 executable or DLL@. For example, this small program defines shared data
3115 by putting it in a named section @code{shared} and marking the section
3119 int foo __attribute__((section ("shared"), shared)) = 0;
3124 /* Read and write foo. All running
3125 copies see the same value. */
3131 You may only use the @code{shared} attribute along with @code{section}
3132 attribute with a fully initialized global definition because of the way
3133 linkers work. See @code{section} attribute for more information.
3135 The @code{shared} attribute is only available on Windows NT@.
3137 @item transparent_union
3138 This attribute, attached to a function parameter which is a union, means
3139 that the corresponding argument may have the type of any union member,
3140 but the argument is passed as if its type were that of the first union
3141 member. For more details see @xref{Type Attributes}. You can also use
3142 this attribute on a @code{typedef} for a union data type; then it
3143 applies to all function parameters with that type.
3146 This attribute, attached to a variable, means that the variable is meant
3147 to be possibly unused. GCC will not produce a warning for this
3151 The @code{deprecated} attribute results in a warning if the variable
3152 is used anywhere in the source file. This is useful when identifying
3153 variables that are expected to be removed in a future version of a
3154 program. The warning also includes the location of the declaration
3155 of the deprecated variable, to enable users to easily find further
3156 information about why the variable is deprecated, or what they should
3157 do instead. Note that the warnings only occurs for uses:
3160 extern int old_var __attribute__ ((deprecated));
3162 int new_fn () @{ return old_var; @}
3165 results in a warning on line 3 but not line 2.
3167 The @code{deprecated} attribute can also be used for functions and
3168 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
3170 @item vector_size (@var{bytes})
3171 This attribute specifies the vector size for the variable, measured in
3172 bytes. For example, the declaration:
3175 int foo __attribute__ ((vector_size (16)));
3179 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3180 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3181 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3183 This attribute is only applicable to integral and float scalars,
3184 although arrays, pointers, and function return values are allowed in
3185 conjunction with this construct.
3187 Aggregates with this attribute are invalid, even if they are of the same
3188 size as a corresponding scalar. For example, the declaration:
3191 struct S @{ int a; @};
3192 struct S __attribute__ ((vector_size (16))) foo;
3196 is invalid even if the size of the structure is the same as the size of
3200 The @code{weak} attribute is described in @xref{Function Attributes}.
3202 @item model (@var{model-name})
3203 @cindex variable addressability on the M32R/D
3204 Use this attribute on the M32R/D to set the addressability of an object.
3205 The identifier @var{model-name} is one of @code{small}, @code{medium},
3206 or @code{large}, representing each of the code models.
3208 Small model objects live in the lower 16MB of memory (so that their
3209 addresses can be loaded with the @code{ld24} instruction).
3211 Medium and large model objects may live anywhere in the 32-bit address space
3212 (the compiler will generate @code{seth/add3} instructions to load their
3217 To specify multiple attributes, separate them by commas within the
3218 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3221 @node Type Attributes
3222 @section Specifying Attributes of Types
3223 @cindex attribute of types
3224 @cindex type attributes
3226 The keyword @code{__attribute__} allows you to specify special
3227 attributes of @code{struct} and @code{union} types when you define such
3228 types. This keyword is followed by an attribute specification inside
3229 double parentheses. Six attributes are currently defined for types:
3230 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
3231 @code{deprecated} and @code{may_alias}. Other attributes are defined for
3232 functions (@pxref{Function Attributes}) and for variables
3233 (@pxref{Variable Attributes}).
3235 You may also specify any one of these attributes with @samp{__}
3236 preceding and following its keyword. This allows you to use these
3237 attributes in header files without being concerned about a possible
3238 macro of the same name. For example, you may use @code{__aligned__}
3239 instead of @code{aligned}.
3241 You may specify the @code{aligned} and @code{transparent_union}
3242 attributes either in a @code{typedef} declaration or just past the
3243 closing curly brace of a complete enum, struct or union type
3244 @emph{definition} and the @code{packed} attribute only past the closing
3245 brace of a definition.
3247 You may also specify attributes between the enum, struct or union
3248 tag and the name of the type rather than after the closing brace.
3250 @xref{Attribute Syntax}, for details of the exact syntax for using
3254 @cindex @code{aligned} attribute
3255 @item aligned (@var{alignment})
3256 This attribute specifies a minimum alignment (in bytes) for variables
3257 of the specified type. For example, the declarations:
3260 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3261 typedef int more_aligned_int __attribute__ ((aligned (8)));
3265 force the compiler to insure (as far as it can) that each variable whose
3266 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3267 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
3268 variables of type @code{struct S} aligned to 8-byte boundaries allows
3269 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3270 store) instructions when copying one variable of type @code{struct S} to
3271 another, thus improving run-time efficiency.
3273 Note that the alignment of any given @code{struct} or @code{union} type
3274 is required by the ISO C standard to be at least a perfect multiple of
3275 the lowest common multiple of the alignments of all of the members of
3276 the @code{struct} or @code{union} in question. This means that you @emph{can}
3277 effectively adjust the alignment of a @code{struct} or @code{union}
3278 type by attaching an @code{aligned} attribute to any one of the members
3279 of such a type, but the notation illustrated in the example above is a
3280 more obvious, intuitive, and readable way to request the compiler to
3281 adjust the alignment of an entire @code{struct} or @code{union} type.
3283 As in the preceding example, you can explicitly specify the alignment
3284 (in bytes) that you wish the compiler to use for a given @code{struct}
3285 or @code{union} type. Alternatively, you can leave out the alignment factor
3286 and just ask the compiler to align a type to the maximum
3287 useful alignment for the target machine you are compiling for. For
3288 example, you could write:
3291 struct S @{ short f[3]; @} __attribute__ ((aligned));
3294 Whenever you leave out the alignment factor in an @code{aligned}
3295 attribute specification, the compiler automatically sets the alignment
3296 for the type to the largest alignment which is ever used for any data
3297 type on the target machine you are compiling for. Doing this can often
3298 make copy operations more efficient, because the compiler can use
3299 whatever instructions copy the biggest chunks of memory when performing
3300 copies to or from the variables which have types that you have aligned
3303 In the example above, if the size of each @code{short} is 2 bytes, then
3304 the size of the entire @code{struct S} type is 6 bytes. The smallest
3305 power of two which is greater than or equal to that is 8, so the
3306 compiler sets the alignment for the entire @code{struct S} type to 8
3309 Note that although you can ask the compiler to select a time-efficient
3310 alignment for a given type and then declare only individual stand-alone
3311 objects of that type, the compiler's ability to select a time-efficient
3312 alignment is primarily useful only when you plan to create arrays of
3313 variables having the relevant (efficiently aligned) type. If you
3314 declare or use arrays of variables of an efficiently-aligned type, then
3315 it is likely that your program will also be doing pointer arithmetic (or
3316 subscripting, which amounts to the same thing) on pointers to the
3317 relevant type, and the code that the compiler generates for these
3318 pointer arithmetic operations will often be more efficient for
3319 efficiently-aligned types than for other types.
3321 The @code{aligned} attribute can only increase the alignment; but you
3322 can decrease it by specifying @code{packed} as well. See below.
3324 Note that the effectiveness of @code{aligned} attributes may be limited
3325 by inherent limitations in your linker. On many systems, the linker is
3326 only able to arrange for variables to be aligned up to a certain maximum
3327 alignment. (For some linkers, the maximum supported alignment may
3328 be very very small.) If your linker is only able to align variables
3329 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3330 in an @code{__attribute__} will still only provide you with 8 byte
3331 alignment. See your linker documentation for further information.
3334 This attribute, attached to an @code{enum}, @code{struct}, or
3335 @code{union} type definition, specified that the minimum required memory
3336 be used to represent the type.
3338 @opindex fshort-enums
3339 Specifying this attribute for @code{struct} and @code{union} types is
3340 equivalent to specifying the @code{packed} attribute on each of the
3341 structure or union members. Specifying the @option{-fshort-enums}
3342 flag on the line is equivalent to specifying the @code{packed}
3343 attribute on all @code{enum} definitions.
3345 You may only specify this attribute after a closing curly brace on an
3346 @code{enum} definition, not in a @code{typedef} declaration, unless that
3347 declaration also contains the definition of the @code{enum}.
3349 @item transparent_union
3350 This attribute, attached to a @code{union} type definition, indicates
3351 that any function parameter having that union type causes calls to that
3352 function to be treated in a special way.
3354 First, the argument corresponding to a transparent union type can be of
3355 any type in the union; no cast is required. Also, if the union contains
3356 a pointer type, the corresponding argument can be a null pointer
3357 constant or a void pointer expression; and if the union contains a void
3358 pointer type, the corresponding argument can be any pointer expression.
3359 If the union member type is a pointer, qualifiers like @code{const} on
3360 the referenced type must be respected, just as with normal pointer
3363 Second, the argument is passed to the function using the calling
3364 conventions of first member of the transparent union, not the calling
3365 conventions of the union itself. All members of the union must have the
3366 same machine representation; this is necessary for this argument passing
3369 Transparent unions are designed for library functions that have multiple
3370 interfaces for compatibility reasons. For example, suppose the
3371 @code{wait} function must accept either a value of type @code{int *} to
3372 comply with Posix, or a value of type @code{union wait *} to comply with
3373 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3374 @code{wait} would accept both kinds of arguments, but it would also
3375 accept any other pointer type and this would make argument type checking
3376 less useful. Instead, @code{<sys/wait.h>} might define the interface
3384 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3386 pid_t wait (wait_status_ptr_t);
3389 This interface allows either @code{int *} or @code{union wait *}
3390 arguments to be passed, using the @code{int *} calling convention.
3391 The program can call @code{wait} with arguments of either type:
3394 int w1 () @{ int w; return wait (&w); @}
3395 int w2 () @{ union wait w; return wait (&w); @}
3398 With this interface, @code{wait}'s implementation might look like this:
3401 pid_t wait (wait_status_ptr_t p)
3403 return waitpid (-1, p.__ip, 0);
3408 When attached to a type (including a @code{union} or a @code{struct}),
3409 this attribute means that variables of that type are meant to appear
3410 possibly unused. GCC will not produce a warning for any variables of
3411 that type, even if the variable appears to do nothing. This is often
3412 the case with lock or thread classes, which are usually defined and then
3413 not referenced, but contain constructors and destructors that have
3414 nontrivial bookkeeping functions.
3417 The @code{deprecated} attribute results in a warning if the type
3418 is used anywhere in the source file. This is useful when identifying
3419 types that are expected to be removed in a future version of a program.
3420 If possible, the warning also includes the location of the declaration
3421 of the deprecated type, to enable users to easily find further
3422 information about why the type is deprecated, or what they should do
3423 instead. Note that the warnings only occur for uses and then only
3424 if the type is being applied to an identifier that itself is not being
3425 declared as deprecated.
3428 typedef int T1 __attribute__ ((deprecated));
3432 typedef T1 T3 __attribute__ ((deprecated));
3433 T3 z __attribute__ ((deprecated));
3436 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3437 warning is issued for line 4 because T2 is not explicitly
3438 deprecated. Line 5 has no warning because T3 is explicitly
3439 deprecated. Similarly for line 6.
3441 The @code{deprecated} attribute can also be used for functions and
3442 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3445 Accesses to objects with types with this attribute are not subjected to
3446 type-based alias analysis, but are instead assumed to be able to alias
3447 any other type of objects, just like the @code{char} type. See
3448 @option{-fstrict-aliasing} for more information on aliasing issues.
3453 typedef short __attribute__((__may_alias__)) short_a;
3459 short_a *b = (short_a *) &a;
3463 if (a == 0x12345678)
3470 If you replaced @code{short_a} with @code{short} in the variable
3471 declaration, the above program would abort when compiled with
3472 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3473 above in recent GCC versions.
3476 To specify multiple attributes, separate them by commas within the
3477 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3481 @section An Inline Function is As Fast As a Macro
3482 @cindex inline functions
3483 @cindex integrating function code
3485 @cindex macros, inline alternative
3487 By declaring a function @code{inline}, you can direct GCC to
3488 integrate that function's code into the code for its callers. This
3489 makes execution faster by eliminating the function-call overhead; in
3490 addition, if any of the actual argument values are constant, their known
3491 values may permit simplifications at compile time so that not all of the
3492 inline function's code needs to be included. The effect on code size is
3493 less predictable; object code may be larger or smaller with function
3494 inlining, depending on the particular case. Inlining of functions is an
3495 optimization and it really ``works'' only in optimizing compilation. If
3496 you don't use @option{-O}, no function is really inline.
3498 Inline functions are included in the ISO C99 standard, but there are
3499 currently substantial differences between what GCC implements and what
3500 the ISO C99 standard requires.
3502 To declare a function inline, use the @code{inline} keyword in its
3503 declaration, like this:
3513 (If you are writing a header file to be included in ISO C programs, write
3514 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3515 You can also make all ``simple enough'' functions inline with the option
3516 @option{-finline-functions}.
3519 Note that certain usages in a function definition can make it unsuitable
3520 for inline substitution. Among these usages are: use of varargs, use of
3521 alloca, use of variable sized data types (@pxref{Variable Length}),
3522 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3523 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3524 will warn when a function marked @code{inline} could not be substituted,
3525 and will give the reason for the failure.
3527 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3528 does not affect the linkage of the function.
3530 @cindex automatic @code{inline} for C++ member fns
3531 @cindex @code{inline} automatic for C++ member fns
3532 @cindex member fns, automatically @code{inline}
3533 @cindex C++ member fns, automatically @code{inline}
3534 @opindex fno-default-inline
3535 GCC automatically inlines member functions defined within the class
3536 body of C++ programs even if they are not explicitly declared
3537 @code{inline}. (You can override this with @option{-fno-default-inline};
3538 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3540 @cindex inline functions, omission of
3541 @opindex fkeep-inline-functions
3542 When a function is both inline and @code{static}, if all calls to the
3543 function are integrated into the caller, and the function's address is
3544 never used, then the function's own assembler code is never referenced.
3545 In this case, GCC does not actually output assembler code for the
3546 function, unless you specify the option @option{-fkeep-inline-functions}.
3547 Some calls cannot be integrated for various reasons (in particular,
3548 calls that precede the function's definition cannot be integrated, and
3549 neither can recursive calls within the definition). If there is a
3550 nonintegrated call, then the function is compiled to assembler code as
3551 usual. The function must also be compiled as usual if the program
3552 refers to its address, because that can't be inlined.
3554 @cindex non-static inline function
3555 When an inline function is not @code{static}, then the compiler must assume
3556 that there may be calls from other source files; since a global symbol can
3557 be defined only once in any program, the function must not be defined in
3558 the other source files, so the calls therein cannot be integrated.
3559 Therefore, a non-@code{static} inline function is always compiled on its
3560 own in the usual fashion.
3562 If you specify both @code{inline} and @code{extern} in the function
3563 definition, then the definition is used only for inlining. In no case
3564 is the function compiled on its own, not even if you refer to its
3565 address explicitly. Such an address becomes an external reference, as
3566 if you had only declared the function, and had not defined it.
3568 This combination of @code{inline} and @code{extern} has almost the
3569 effect of a macro. The way to use it is to put a function definition in
3570 a header file with these keywords, and put another copy of the
3571 definition (lacking @code{inline} and @code{extern}) in a library file.
3572 The definition in the header file will cause most calls to the function
3573 to be inlined. If any uses of the function remain, they will refer to
3574 the single copy in the library.
3576 For future compatibility with when GCC implements ISO C99 semantics for
3577 inline functions, it is best to use @code{static inline} only. (The
3578 existing semantics will remain available when @option{-std=gnu89} is
3579 specified, but eventually the default will be @option{-std=gnu99} and
3580 that will implement the C99 semantics, though it does not do so yet.)
3582 GCC does not inline any functions when not optimizing unless you specify
3583 the @samp{always_inline} attribute for the function, like this:
3587 inline void foo (const char) __attribute__((always_inline));
3591 @section Assembler Instructions with C Expression Operands
3592 @cindex extended @code{asm}
3593 @cindex @code{asm} expressions
3594 @cindex assembler instructions
3597 In an assembler instruction using @code{asm}, you can specify the
3598 operands of the instruction using C expressions. This means you need not
3599 guess which registers or memory locations will contain the data you want
3602 You must specify an assembler instruction template much like what
3603 appears in a machine description, plus an operand constraint string for
3606 For example, here is how to use the 68881's @code{fsinx} instruction:
3609 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3613 Here @code{angle} is the C expression for the input operand while
3614 @code{result} is that of the output operand. Each has @samp{"f"} as its
3615 operand constraint, saying that a floating point register is required.
3616 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3617 output operands' constraints must use @samp{=}. The constraints use the
3618 same language used in the machine description (@pxref{Constraints}).
3620 Each operand is described by an operand-constraint string followed by
3621 the C expression in parentheses. A colon separates the assembler
3622 template from the first output operand and another separates the last
3623 output operand from the first input, if any. Commas separate the
3624 operands within each group. The total number of operands is currently
3625 limited to 30; this limitation may be lifted in some future version of
3628 If there are no output operands but there are input operands, you must
3629 place two consecutive colons surrounding the place where the output
3632 As of GCC version 3.1, it is also possible to specify input and output
3633 operands using symbolic names which can be referenced within the
3634 assembler code. These names are specified inside square brackets
3635 preceding the constraint string, and can be referenced inside the
3636 assembler code using @code{%[@var{name}]} instead of a percentage sign
3637 followed by the operand number. Using named operands the above example
3641 asm ("fsinx %[angle],%[output]"
3642 : [output] "=f" (result)
3643 : [angle] "f" (angle));
3647 Note that the symbolic operand names have no relation whatsoever to
3648 other C identifiers. You may use any name you like, even those of
3649 existing C symbols, but must ensure that no two operands within the same
3650 assembler construct use the same symbolic name.
3652 Output operand expressions must be lvalues; the compiler can check this.
3653 The input operands need not be lvalues. The compiler cannot check
3654 whether the operands have data types that are reasonable for the
3655 instruction being executed. It does not parse the assembler instruction
3656 template and does not know what it means or even whether it is valid
3657 assembler input. The extended @code{asm} feature is most often used for
3658 machine instructions the compiler itself does not know exist. If
3659 the output expression cannot be directly addressed (for example, it is a
3660 bit-field), your constraint must allow a register. In that case, GCC
3661 will use the register as the output of the @code{asm}, and then store
3662 that register into the output.
3664 The ordinary output operands must be write-only; GCC will assume that
3665 the values in these operands before the instruction are dead and need
3666 not be generated. Extended asm supports input-output or read-write
3667 operands. Use the constraint character @samp{+} to indicate such an
3668 operand and list it with the output operands.
3670 When the constraints for the read-write operand (or the operand in which
3671 only some of the bits are to be changed) allows a register, you may, as
3672 an alternative, logically split its function into two separate operands,
3673 one input operand and one write-only output operand. The connection
3674 between them is expressed by constraints which say they need to be in
3675 the same location when the instruction executes. You can use the same C
3676 expression for both operands, or different expressions. For example,
3677 here we write the (fictitious) @samp{combine} instruction with
3678 @code{bar} as its read-only source operand and @code{foo} as its
3679 read-write destination:
3682 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3686 The constraint @samp{"0"} for operand 1 says that it must occupy the
3687 same location as operand 0. A number in constraint is allowed only in
3688 an input operand and it must refer to an output operand.
3690 Only a number in the constraint can guarantee that one operand will be in
3691 the same place as another. The mere fact that @code{foo} is the value
3692 of both operands is not enough to guarantee that they will be in the
3693 same place in the generated assembler code. The following would not
3697 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3700 Various optimizations or reloading could cause operands 0 and 1 to be in
3701 different registers; GCC knows no reason not to do so. For example, the
3702 compiler might find a copy of the value of @code{foo} in one register and
3703 use it for operand 1, but generate the output operand 0 in a different
3704 register (copying it afterward to @code{foo}'s own address). Of course,
3705 since the register for operand 1 is not even mentioned in the assembler
3706 code, the result will not work, but GCC can't tell that.
3708 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3709 the operand number for a matching constraint. For example:
3712 asm ("cmoveq %1,%2,%[result]"
3713 : [result] "=r"(result)
3714 : "r" (test), "r"(new), "[result]"(old));
3717 Some instructions clobber specific hard registers. To describe this,
3718 write a third colon after the input operands, followed by the names of
3719 the clobbered hard registers (given as strings). Here is a realistic
3720 example for the VAX:
3723 asm volatile ("movc3 %0,%1,%2"
3725 : "g" (from), "g" (to), "g" (count)
3726 : "r0", "r1", "r2", "r3", "r4", "r5");
3729 You may not write a clobber description in a way that overlaps with an
3730 input or output operand. For example, you may not have an operand
3731 describing a register class with one member if you mention that register
3732 in the clobber list. Variables declared to live in specific registers
3733 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
3734 have no part mentioned in the clobber description.
3735 There is no way for you to specify that an input
3736 operand is modified without also specifying it as an output
3737 operand. Note that if all the output operands you specify are for this
3738 purpose (and hence unused), you will then also need to specify
3739 @code{volatile} for the @code{asm} construct, as described below, to
3740 prevent GCC from deleting the @code{asm} statement as unused.
3742 If you refer to a particular hardware register from the assembler code,
3743 you will probably have to list the register after the third colon to
3744 tell the compiler the register's value is modified. In some assemblers,
3745 the register names begin with @samp{%}; to produce one @samp{%} in the
3746 assembler code, you must write @samp{%%} in the input.
3748 If your assembler instruction can alter the condition code register, add
3749 @samp{cc} to the list of clobbered registers. GCC on some machines
3750 represents the condition codes as a specific hardware register;
3751 @samp{cc} serves to name this register. On other machines, the
3752 condition code is handled differently, and specifying @samp{cc} has no
3753 effect. But it is valid no matter what the machine.
3755 If your assembler instruction modifies memory in an unpredictable
3756 fashion, add @samp{memory} to the list of clobbered registers. This
3757 will cause GCC to not keep memory values cached in registers across
3758 the assembler instruction. You will also want to add the
3759 @code{volatile} keyword if the memory affected is not listed in the
3760 inputs or outputs of the @code{asm}, as the @samp{memory} clobber does
3761 not count as a side-effect of the @code{asm}.
3763 You can put multiple assembler instructions together in a single
3764 @code{asm} template, separated by the characters normally used in assembly
3765 code for the system. A combination that works in most places is a newline
3766 to break the line, plus a tab character to move to the instruction field
3767 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3768 assembler allows semicolons as a line-breaking character. Note that some
3769 assembler dialects use semicolons to start a comment.
3770 The input operands are guaranteed not to use any of the clobbered
3771 registers, and neither will the output operands' addresses, so you can
3772 read and write the clobbered registers as many times as you like. Here
3773 is an example of multiple instructions in a template; it assumes the
3774 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3777 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3779 : "g" (from), "g" (to)
3783 Unless an output operand has the @samp{&} constraint modifier, GCC
3784 may allocate it in the same register as an unrelated input operand, on
3785 the assumption the inputs are consumed before the outputs are produced.
3786 This assumption may be false if the assembler code actually consists of
3787 more than one instruction. In such a case, use @samp{&} for each output
3788 operand that may not overlap an input. @xref{Modifiers}.
3790 If you want to test the condition code produced by an assembler
3791 instruction, you must include a branch and a label in the @code{asm}
3792 construct, as follows:
3795 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3801 This assumes your assembler supports local labels, as the GNU assembler
3802 and most Unix assemblers do.
3804 Speaking of labels, jumps from one @code{asm} to another are not
3805 supported. The compiler's optimizers do not know about these jumps, and
3806 therefore they cannot take account of them when deciding how to
3809 @cindex macros containing @code{asm}
3810 Usually the most convenient way to use these @code{asm} instructions is to
3811 encapsulate them in macros that look like functions. For example,
3815 (@{ double __value, __arg = (x); \
3816 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3821 Here the variable @code{__arg} is used to make sure that the instruction
3822 operates on a proper @code{double} value, and to accept only those
3823 arguments @code{x} which can convert automatically to a @code{double}.
3825 Another way to make sure the instruction operates on the correct data
3826 type is to use a cast in the @code{asm}. This is different from using a
3827 variable @code{__arg} in that it converts more different types. For
3828 example, if the desired type were @code{int}, casting the argument to
3829 @code{int} would accept a pointer with no complaint, while assigning the
3830 argument to an @code{int} variable named @code{__arg} would warn about
3831 using a pointer unless the caller explicitly casts it.
3833 If an @code{asm} has output operands, GCC assumes for optimization
3834 purposes the instruction has no side effects except to change the output
3835 operands. This does not mean instructions with a side effect cannot be
3836 used, but you must be careful, because the compiler may eliminate them
3837 if the output operands aren't used, or move them out of loops, or
3838 replace two with one if they constitute a common subexpression. Also,
3839 if your instruction does have a side effect on a variable that otherwise
3840 appears not to change, the old value of the variable may be reused later
3841 if it happens to be found in a register.
3843 You can prevent an @code{asm} instruction from being deleted, moved
3844 significantly, or combined, by writing the keyword @code{volatile} after
3845 the @code{asm}. For example:
3848 #define get_and_set_priority(new) \
3850 asm volatile ("get_and_set_priority %0, %1" \
3851 : "=g" (__old) : "g" (new)); \
3856 If you write an @code{asm} instruction with no outputs, GCC will know
3857 the instruction has side-effects and will not delete the instruction or
3858 move it outside of loops.
3860 The @code{volatile} keyword indicates that the instruction has
3861 important side-effects. GCC will not delete a volatile @code{asm} if
3862 it is reachable. (The instruction can still be deleted if GCC can
3863 prove that control-flow will never reach the location of the
3864 instruction.) In addition, GCC will not reschedule instructions
3865 across a volatile @code{asm} instruction. For example:
3868 *(volatile int *)addr = foo;
3869 asm volatile ("eieio" : : );
3873 Assume @code{addr} contains the address of a memory mapped device
3874 register. The PowerPC @code{eieio} instruction (Enforce In-order
3875 Execution of I/O) tells the CPU to make sure that the store to that
3876 device register happens before it issues any other I/O@.
3878 Note that even a volatile @code{asm} instruction can be moved in ways
3879 that appear insignificant to the compiler, such as across jump
3880 instructions. You can't expect a sequence of volatile @code{asm}
3881 instructions to remain perfectly consecutive. If you want consecutive
3882 output, use a single @code{asm}. Also, GCC will perform some
3883 optimizations across a volatile @code{asm} instruction; GCC does not
3884 ``forget everything'' when it encounters a volatile @code{asm}
3885 instruction the way some other compilers do.
3887 An @code{asm} instruction without any operands or clobbers (an ``old
3888 style'' @code{asm}) will be treated identically to a volatile
3889 @code{asm} instruction.
3891 It is a natural idea to look for a way to give access to the condition
3892 code left by the assembler instruction. However, when we attempted to
3893 implement this, we found no way to make it work reliably. The problem
3894 is that output operands might need reloading, which would result in
3895 additional following ``store'' instructions. On most machines, these
3896 instructions would alter the condition code before there was time to
3897 test it. This problem doesn't arise for ordinary ``test'' and
3898 ``compare'' instructions because they don't have any output operands.
3900 For reasons similar to those described above, it is not possible to give
3901 an assembler instruction access to the condition code left by previous
3904 If you are writing a header file that should be includable in ISO C
3905 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3908 @subsection i386 floating point asm operands
3910 There are several rules on the usage of stack-like regs in
3911 asm_operands insns. These rules apply only to the operands that are
3916 Given a set of input regs that die in an asm_operands, it is
3917 necessary to know which are implicitly popped by the asm, and
3918 which must be explicitly popped by gcc.
3920 An input reg that is implicitly popped by the asm must be
3921 explicitly clobbered, unless it is constrained to match an
3925 For any input reg that is implicitly popped by an asm, it is
3926 necessary to know how to adjust the stack to compensate for the pop.
3927 If any non-popped input is closer to the top of the reg-stack than
3928 the implicitly popped reg, it would not be possible to know what the
3929 stack looked like---it's not clear how the rest of the stack ``slides
3932 All implicitly popped input regs must be closer to the top of
3933 the reg-stack than any input that is not implicitly popped.
3935 It is possible that if an input dies in an insn, reload might
3936 use the input reg for an output reload. Consider this example:
3939 asm ("foo" : "=t" (a) : "f" (b));
3942 This asm says that input B is not popped by the asm, and that
3943 the asm pushes a result onto the reg-stack, i.e., the stack is one
3944 deeper after the asm than it was before. But, it is possible that
3945 reload will think that it can use the same reg for both the input and
3946 the output, if input B dies in this insn.
3948 If any input operand uses the @code{f} constraint, all output reg
3949 constraints must use the @code{&} earlyclobber.
3951 The asm above would be written as
3954 asm ("foo" : "=&t" (a) : "f" (b));
3958 Some operands need to be in particular places on the stack. All
3959 output operands fall in this category---there is no other way to
3960 know which regs the outputs appear in unless the user indicates
3961 this in the constraints.
3963 Output operands must specifically indicate which reg an output
3964 appears in after an asm. @code{=f} is not allowed: the operand
3965 constraints must select a class with a single reg.
3968 Output operands may not be ``inserted'' between existing stack regs.
3969 Since no 387 opcode uses a read/write operand, all output operands
3970 are dead before the asm_operands, and are pushed by the asm_operands.
3971 It makes no sense to push anywhere but the top of the reg-stack.
3973 Output operands must start at the top of the reg-stack: output
3974 operands may not ``skip'' a reg.
3977 Some asm statements may need extra stack space for internal
3978 calculations. This can be guaranteed by clobbering stack registers
3979 unrelated to the inputs and outputs.
3983 Here are a couple of reasonable asms to want to write. This asm
3984 takes one input, which is internally popped, and produces two outputs.
3987 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
3990 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
3991 and replaces them with one output. The user must code the @code{st(1)}
3992 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
3995 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
4001 @section Controlling Names Used in Assembler Code
4002 @cindex assembler names for identifiers
4003 @cindex names used in assembler code
4004 @cindex identifiers, names in assembler code
4006 You can specify the name to be used in the assembler code for a C
4007 function or variable by writing the @code{asm} (or @code{__asm__})
4008 keyword after the declarator as follows:
4011 int foo asm ("myfoo") = 2;
4015 This specifies that the name to be used for the variable @code{foo} in
4016 the assembler code should be @samp{myfoo} rather than the usual
4019 On systems where an underscore is normally prepended to the name of a C
4020 function or variable, this feature allows you to define names for the
4021 linker that do not start with an underscore.
4023 It does not make sense to use this feature with a non-static local
4024 variable since such variables do not have assembler names. If you are
4025 trying to put the variable in a particular register, see @ref{Explicit
4026 Reg Vars}. GCC presently accepts such code with a warning, but will
4027 probably be changed to issue an error, rather than a warning, in the
4030 You cannot use @code{asm} in this way in a function @emph{definition}; but
4031 you can get the same effect by writing a declaration for the function
4032 before its definition and putting @code{asm} there, like this:
4035 extern func () asm ("FUNC");
4042 It is up to you to make sure that the assembler names you choose do not
4043 conflict with any other assembler symbols. Also, you must not use a
4044 register name; that would produce completely invalid assembler code. GCC
4045 does not as yet have the ability to store static variables in registers.
4046 Perhaps that will be added.
4048 @node Explicit Reg Vars
4049 @section Variables in Specified Registers
4050 @cindex explicit register variables
4051 @cindex variables in specified registers
4052 @cindex specified registers
4053 @cindex registers, global allocation
4055 GNU C allows you to put a few global variables into specified hardware
4056 registers. You can also specify the register in which an ordinary
4057 register variable should be allocated.
4061 Global register variables reserve registers throughout the program.
4062 This may be useful in programs such as programming language
4063 interpreters which have a couple of global variables that are accessed
4067 Local register variables in specific registers do not reserve the
4068 registers. The compiler's data flow analysis is capable of determining
4069 where the specified registers contain live values, and where they are
4070 available for other uses. Stores into local register variables may be deleted
4071 when they appear to be dead according to dataflow analysis. References
4072 to local register variables may be deleted or moved or simplified.
4074 These local variables are sometimes convenient for use with the extended
4075 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
4076 output of the assembler instruction directly into a particular register.
4077 (This will work provided the register you specify fits the constraints
4078 specified for that operand in the @code{asm}.)
4086 @node Global Reg Vars
4087 @subsection Defining Global Register Variables
4088 @cindex global register variables
4089 @cindex registers, global variables in
4091 You can define a global register variable in GNU C like this:
4094 register int *foo asm ("a5");
4098 Here @code{a5} is the name of the register which should be used. Choose a
4099 register which is normally saved and restored by function calls on your
4100 machine, so that library routines will not clobber it.
4102 Naturally the register name is cpu-dependent, so you would need to
4103 conditionalize your program according to cpu type. The register
4104 @code{a5} would be a good choice on a 68000 for a variable of pointer
4105 type. On machines with register windows, be sure to choose a ``global''
4106 register that is not affected magically by the function call mechanism.
4108 In addition, operating systems on one type of cpu may differ in how they
4109 name the registers; then you would need additional conditionals. For
4110 example, some 68000 operating systems call this register @code{%a5}.
4112 Eventually there may be a way of asking the compiler to choose a register
4113 automatically, but first we need to figure out how it should choose and
4114 how to enable you to guide the choice. No solution is evident.
4116 Defining a global register variable in a certain register reserves that
4117 register entirely for this use, at least within the current compilation.
4118 The register will not be allocated for any other purpose in the functions
4119 in the current compilation. The register will not be saved and restored by
4120 these functions. Stores into this register are never deleted even if they
4121 would appear to be dead, but references may be deleted or moved or
4124 It is not safe to access the global register variables from signal
4125 handlers, or from more than one thread of control, because the system
4126 library routines may temporarily use the register for other things (unless
4127 you recompile them specially for the task at hand).
4129 @cindex @code{qsort}, and global register variables
4130 It is not safe for one function that uses a global register variable to
4131 call another such function @code{foo} by way of a third function
4132 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4133 different source file in which the variable wasn't declared). This is
4134 because @code{lose} might save the register and put some other value there.
4135 For example, you can't expect a global register variable to be available in
4136 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4137 might have put something else in that register. (If you are prepared to
4138 recompile @code{qsort} with the same global register variable, you can
4139 solve this problem.)
4141 If you want to recompile @code{qsort} or other source files which do not
4142 actually use your global register variable, so that they will not use that
4143 register for any other purpose, then it suffices to specify the compiler
4144 option @option{-ffixed-@var{reg}}. You need not actually add a global
4145 register declaration to their source code.
4147 A function which can alter the value of a global register variable cannot
4148 safely be called from a function compiled without this variable, because it
4149 could clobber the value the caller expects to find there on return.
4150 Therefore, the function which is the entry point into the part of the
4151 program that uses the global register variable must explicitly save and
4152 restore the value which belongs to its caller.
4154 @cindex register variable after @code{longjmp}
4155 @cindex global register after @code{longjmp}
4156 @cindex value after @code{longjmp}
4159 On most machines, @code{longjmp} will restore to each global register
4160 variable the value it had at the time of the @code{setjmp}. On some
4161 machines, however, @code{longjmp} will not change the value of global
4162 register variables. To be portable, the function that called @code{setjmp}
4163 should make other arrangements to save the values of the global register
4164 variables, and to restore them in a @code{longjmp}. This way, the same
4165 thing will happen regardless of what @code{longjmp} does.
4167 All global register variable declarations must precede all function
4168 definitions. If such a declaration could appear after function
4169 definitions, the declaration would be too late to prevent the register from
4170 being used for other purposes in the preceding functions.
4172 Global register variables may not have initial values, because an
4173 executable file has no means to supply initial contents for a register.
4175 On the SPARC, there are reports that g3 @dots{} g7 are suitable
4176 registers, but certain library functions, such as @code{getwd}, as well
4177 as the subroutines for division and remainder, modify g3 and g4. g1 and
4178 g2 are local temporaries.
4180 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4181 Of course, it will not do to use more than a few of those.
4183 @node Local Reg Vars
4184 @subsection Specifying Registers for Local Variables
4185 @cindex local variables, specifying registers
4186 @cindex specifying registers for local variables
4187 @cindex registers for local variables
4189 You can define a local register variable with a specified register
4193 register int *foo asm ("a5");
4197 Here @code{a5} is the name of the register which should be used. Note
4198 that this is the same syntax used for defining global register
4199 variables, but for a local variable it would appear within a function.
4201 Naturally the register name is cpu-dependent, but this is not a
4202 problem, since specific registers are most often useful with explicit
4203 assembler instructions (@pxref{Extended Asm}). Both of these things
4204 generally require that you conditionalize your program according to
4207 In addition, operating systems on one type of cpu may differ in how they
4208 name the registers; then you would need additional conditionals. For
4209 example, some 68000 operating systems call this register @code{%a5}.
4211 Defining such a register variable does not reserve the register; it
4212 remains available for other uses in places where flow control determines
4213 the variable's value is not live. However, these registers are made
4214 unavailable for use in the reload pass; excessive use of this feature
4215 leaves the compiler too few available registers to compile certain
4218 This option does not guarantee that GCC will generate code that has
4219 this variable in the register you specify at all times. You may not
4220 code an explicit reference to this register in an @code{asm} statement
4221 and assume it will always refer to this variable.
4223 Stores into local register variables may be deleted when they appear to be dead
4224 according to dataflow analysis. References to local register variables may
4225 be deleted or moved or simplified.
4227 @node Alternate Keywords
4228 @section Alternate Keywords
4229 @cindex alternate keywords
4230 @cindex keywords, alternate
4232 @option{-ansi} and the various @option{-std} options disable certain
4233 keywords. This causes trouble when you want to use GNU C extensions, or
4234 a general-purpose header file that should be usable by all programs,
4235 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4236 @code{inline} are not available in programs compiled with
4237 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4238 program compiled with @option{-std=c99}). The ISO C99 keyword
4239 @code{restrict} is only available when @option{-std=gnu99} (which will
4240 eventually be the default) or @option{-std=c99} (or the equivalent
4241 @option{-std=iso9899:1999}) is used.
4243 The way to solve these problems is to put @samp{__} at the beginning and
4244 end of each problematical keyword. For example, use @code{__asm__}
4245 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4247 Other C compilers won't accept these alternative keywords; if you want to
4248 compile with another compiler, you can define the alternate keywords as
4249 macros to replace them with the customary keywords. It looks like this:
4257 @findex __extension__
4259 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4261 prevent such warnings within one expression by writing
4262 @code{__extension__} before the expression. @code{__extension__} has no
4263 effect aside from this.
4265 @node Incomplete Enums
4266 @section Incomplete @code{enum} Types
4268 You can define an @code{enum} tag without specifying its possible values.
4269 This results in an incomplete type, much like what you get if you write
4270 @code{struct foo} without describing the elements. A later declaration
4271 which does specify the possible values completes the type.
4273 You can't allocate variables or storage using the type while it is
4274 incomplete. However, you can work with pointers to that type.
4276 This extension may not be very useful, but it makes the handling of
4277 @code{enum} more consistent with the way @code{struct} and @code{union}
4280 This extension is not supported by GNU C++.
4282 @node Function Names
4283 @section Function Names as Strings
4284 @cindex @code{__FUNCTION__} identifier
4285 @cindex @code{__PRETTY_FUNCTION__} identifier
4286 @cindex @code{__func__} identifier
4288 GCC predefines two magic identifiers to hold the name of the current
4289 function. The identifier @code{__FUNCTION__} holds the name of the function
4290 as it appears in the source. The identifier @code{__PRETTY_FUNCTION__}
4291 holds the name of the function pretty printed in a language specific
4294 These names are always the same in a C function, but in a C++ function
4295 they may be different. For example, this program:
4299 extern int printf (char *, ...);
4306 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4307 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4325 __PRETTY_FUNCTION__ = int a::sub (int)
4328 The compiler automagically replaces the identifiers with a string
4329 literal containing the appropriate name. Thus, they are neither
4330 preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor
4331 variables. This means that they catenate with other string literals, and
4332 that they can be used to initialize char arrays. For example
4335 char here[] = "Function " __FUNCTION__ " in " __FILE__;
4338 On the other hand, @samp{#ifdef __FUNCTION__} does not have any special
4339 meaning inside a function, since the preprocessor does not do anything
4340 special with the identifier @code{__FUNCTION__}.
4342 Note that these semantics are deprecated, and that GCC 3.2 will handle
4343 @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} the same way as
4344 @code{__func__}. @code{__func__} is defined by the ISO standard C99:
4347 The identifier @code{__func__} is implicitly declared by the translator
4348 as if, immediately following the opening brace of each function
4349 definition, the declaration
4352 static const char __func__[] = "function-name";
4355 appeared, where function-name is the name of the lexically-enclosing
4356 function. This name is the unadorned name of the function.
4359 By this definition, @code{__func__} is a variable, not a string literal.
4360 In particular, @code{__func__} does not catenate with other string
4363 In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are
4364 variables, declared in the same way as @code{__func__}.
4366 @node Return Address
4367 @section Getting the Return or Frame Address of a Function
4369 These functions may be used to get information about the callers of a
4372 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4373 This function returns the return address of the current function, or of
4374 one of its callers. The @var{level} argument is number of frames to
4375 scan up the call stack. A value of @code{0} yields the return address
4376 of the current function, a value of @code{1} yields the return address
4377 of the caller of the current function, and so forth. When inlining
4378 the expected behavior is that the function will return the address of
4379 the function that will be returned to. To work around this behavior use
4380 the @code{noinline} function attribute.
4382 The @var{level} argument must be a constant integer.
4384 On some machines it may be impossible to determine the return address of
4385 any function other than the current one; in such cases, or when the top
4386 of the stack has been reached, this function will return @code{0} or a
4387 random value. In addition, @code{__builtin_frame_address} may be used
4388 to determine if the top of the stack has been reached.
4390 This function should only be used with a nonzero argument for debugging
4394 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4395 This function is similar to @code{__builtin_return_address}, but it
4396 returns the address of the function frame rather than the return address
4397 of the function. Calling @code{__builtin_frame_address} with a value of
4398 @code{0} yields the frame address of the current function, a value of
4399 @code{1} yields the frame address of the caller of the current function,
4402 The frame is the area on the stack which holds local variables and saved
4403 registers. The frame address is normally the address of the first word
4404 pushed on to the stack by the function. However, the exact definition
4405 depends upon the processor and the calling convention. If the processor
4406 has a dedicated frame pointer register, and the function has a frame,
4407 then @code{__builtin_frame_address} will return the value of the frame
4410 On some machines it may be impossible to determine the frame address of
4411 any function other than the current one; in such cases, or when the top
4412 of the stack has been reached, this function will return @code{0} if
4413 the first frame pointer is properly initialized by the startup code.
4415 This function should only be used with a nonzero argument for debugging
4419 @node Vector Extensions
4420 @section Using vector instructions through built-in functions
4422 On some targets, the instruction set contains SIMD vector instructions that
4423 operate on multiple values contained in one large register at the same time.
4424 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4427 The first step in using these extensions is to provide the necessary data
4428 types. This should be done using an appropriate @code{typedef}:
4431 typedef int v4si __attribute__ ((mode(V4SI)));
4434 The base type @code{int} is effectively ignored by the compiler, the
4435 actual properties of the new type @code{v4si} are defined by the
4436 @code{__attribute__}. It defines the machine mode to be used; for vector
4437 types these have the form @code{V@var{n}@var{B}}; @var{n} should be the
4438 number of elements in the vector, and @var{B} should be the base mode of the
4439 individual elements. The following can be used as base modes:
4443 An integer that is as wide as the smallest addressable unit, usually 8 bits.
4445 An integer, twice as wide as a QI mode integer, usually 16 bits.
4447 An integer, four times as wide as a QI mode integer, usually 32 bits.
4449 An integer, eight times as wide as a QI mode integer, usually 64 bits.
4451 A floating point value, as wide as a SI mode integer, usually 32 bits.
4453 A floating point value, as wide as a DI mode integer, usually 64 bits.
4456 Specifying a combination that is not valid for the current architecture
4457 will cause gcc to synthesize the instructions using a narrower mode.
4458 For example, if you specify a variable of type @code{V4SI} and your
4459 architecture does not allow for this specific SIMD type, gcc will
4460 produce code that uses 4 @code{SIs}.
4462 The types defined in this manner can be used with a subset of normal C
4463 operations. Currently, gcc will allow using the following operators on
4464 these types: @code{+, -, *, /, unary minus}@.
4466 The operations behave like C++ @code{valarrays}. Addition is defined as
4467 the addition of the corresponding elements of the operands. For
4468 example, in the code below, each of the 4 elements in @var{a} will be
4469 added to the corresponding 4 elements in @var{b} and the resulting
4470 vector will be stored in @var{c}.
4473 typedef int v4si __attribute__ ((mode(V4SI)));
4480 Subtraction, multiplication, and division operate in a similar manner.
4481 Likewise, the result of using the unary minus operator on a vector type
4482 is a vector whose elements are the negative value of the corresponding
4483 elements in the operand.
4485 You can declare variables and use them in function calls and returns, as
4486 well as in assignments and some casts. You can specify a vector type as
4487 a return type for a function. Vector types can also be used as function
4488 arguments. It is possible to cast from one vector type to another,
4489 provided they are of the same size (in fact, you can also cast vectors
4490 to and from other datatypes of the same size).
4492 You cannot operate between vectors of different lengths or different
4493 signness without a cast.
4495 A port that supports hardware vector operations, usually provides a set
4496 of built-in functions that can be used to operate on vectors. For
4497 example, a function to add two vectors and multiply the result by a
4498 third could look like this:
4501 v4si f (v4si a, v4si b, v4si c)
4503 v4si tmp = __builtin_addv4si (a, b);
4504 return __builtin_mulv4si (tmp, c);
4509 @node Other Builtins
4510 @section Other built-in functions provided by GCC
4511 @cindex built-in functions
4512 @findex __builtin_isgreater
4513 @findex __builtin_isgreaterequal
4514 @findex __builtin_isless
4515 @findex __builtin_islessequal
4516 @findex __builtin_islessgreater
4517 @findex __builtin_isunordered
4546 @findex fprintf_unlocked
4548 @findex fputs_unlocked
4560 @findex printf_unlocked
4597 GCC provides a large number of built-in functions other than the ones
4598 mentioned above. Some of these are for internal use in the processing
4599 of exceptions or variable-length argument lists and will not be
4600 documented here because they may change from time to time; we do not
4601 recommend general use of these functions.
4603 The remaining functions are provided for optimization purposes.
4605 @opindex fno-builtin
4606 GCC includes built-in versions of many of the functions in the standard
4607 C library. The versions prefixed with @code{__builtin_} will always be
4608 treated as having the same meaning as the C library function even if you
4609 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
4610 Many of these functions are only optimized in certain cases; if they are
4611 not optimized in a particular case, a call to the library function will
4616 The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit}
4617 are recognized and presumed not to return, but otherwise are not built
4618 in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi},
4619 @option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in
4620 strict C89 mode (@option{-ansi} or @option{-std=c89}). All these functions
4621 have corresponding versions prefixed with @code{__builtin_}, which may be
4622 used even in strict C89 mode.
4624 Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp},
4625 @code{bzero}, @code{index}, @code{rindex}, @code{ffs}, @code{fputs_unlocked},
4626 @code{printf_unlocked} and @code{fprintf_unlocked} may be handled as
4627 built-in functions. All these functions have corresponding versions
4628 prefixed with @code{__builtin_}, which may be used even in strict C89
4631 The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl}, @code{creal},
4632 @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf}, @code{cimagl},
4633 @code{llabs}, @code{imaxabs}, @code{round}, @code{trunc}, @code{nearbyint},
4634 @code{roundf}, @code{truncf}, @code{nearbyintf}, @code{roundl}, @code{truncl} and
4635 @code{nearbyintl} are handled as built-in functions except in strict ISO C90 mode.
4636 There are also built-in versions of the ISO C99 functions @code{cosf},
4637 @code{cosl}, @code{expf}, @code{expl}, @code{fabsf}, @code{fabsl}, @code{logf},
4638 @code{logl}, @code{sinf}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
4639 @code{ceilf}, @code{ceill}, @code{floorf} and @code{floorl} that are recognized
4640 in any mode since ISO C90 reserves these names for the purpose to which ISO C99
4641 puts them. All these functions have corresponding versions prefixed with
4644 The ISO C90 functions @code{abs}, @code{cos}, @code{exp}, @code{fabs},
4645 @code{fprintf}, @code{fputs}, @code{labs}, @code{log}, @code{floor},
4646 @code{ceil} @code{memcmp}, @code{memcpy}, @code{memset}, @code{printf},
4647 @code{sin}, @code{sqrt}, @code{strcat}, @code{strchr}, @code{strcmp},
4648 @code{strcpy}, @code{strcspn}, @code{strlen}, @code{strncat}, @code{strncmp},
4649 @code{strncpy}, @code{strpbrk}, @code{strrchr}, @code{strspn}, and
4650 @code{strstr} are all recognized as built-in functions unless
4651 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}} is
4652 specified for an individual function). All of these functions have
4653 corresponding versions prefixed with @code{__builtin_}.
4655 GCC provides built-in versions of the ISO C99 floating point comparison
4656 macros that avoid raising exceptions for unordered operands. They have
4657 the same names as the standard macros ( @code{isgreater},
4658 @code{isgreaterequal}, @code{isless}, @code{islessequal},
4659 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
4660 prefixed. We intend for a library implementor to be able to simply
4661 @code{#define} each standard macro to its built-in equivalent.
4663 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
4665 You can use the built-in function @code{__builtin_types_compatible_p} to
4666 determine whether two types are the same.
4668 This built-in function returns 1 if the unqualified versions of the
4669 types @var{type1} and @var{type2} (which are types, not expressions) are
4670 compatible, 0 otherwise. The result of this built-in function can be
4671 used in integer constant expressions.
4673 This built-in function ignores top level qualifiers (e.g., @code{const},
4674 @code{volatile}). For example, @code{int} is equivalent to @code{const
4677 The type @code{int[]} and @code{int[5]} are compatible. On the other
4678 hand, @code{int} and @code{char *} are not compatible, even if the size
4679 of their types, on the particular architecture are the same. Also, the
4680 amount of pointer indirection is taken into account when determining
4681 similarity. Consequently, @code{short *} is not similar to
4682 @code{short **}. Furthermore, two types that are typedefed are
4683 considered compatible if their underlying types are compatible.
4685 An @code{enum} type is considered to be compatible with another
4686 @code{enum} type. For example, @code{enum @{foo, bar@}} is similar to
4687 @code{enum @{hot, dog@}}.
4689 You would typically use this function in code whose execution varies
4690 depending on the arguments' types. For example:
4696 if (__builtin_types_compatible_p (typeof (x), long double)) \
4697 tmp = foo_long_double (tmp); \
4698 else if (__builtin_types_compatible_p (typeof (x), double)) \
4699 tmp = foo_double (tmp); \
4700 else if (__builtin_types_compatible_p (typeof (x), float)) \
4701 tmp = foo_float (tmp); \
4708 @emph{Note:} This construct is only available for C.
4712 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
4714 You can use the built-in function @code{__builtin_choose_expr} to
4715 evaluate code depending on the value of a constant expression. This
4716 built-in function returns @var{exp1} if @var{const_exp}, which is a
4717 constant expression that must be able to be determined at compile time,
4718 is nonzero. Otherwise it returns 0.
4720 This built-in function is analogous to the @samp{? :} operator in C,
4721 except that the expression returned has its type unaltered by promotion
4722 rules. Also, the built-in function does not evaluate the expression
4723 that was not chosen. For example, if @var{const_exp} evaluates to true,
4724 @var{exp2} is not evaluated even if it has side-effects.
4726 This built-in function can return an lvalue if the chosen argument is an
4729 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
4730 type. Similarly, if @var{exp2} is returned, its return type is the same
4737 __builtin_choose_expr ( \
4738 __builtin_types_compatible_p (typeof (x), double), \
4740 __builtin_choose_expr ( \
4741 __builtin_types_compatible_p (typeof (x), float), \
4743 /* @r{The void expression results in a compile-time error} \
4744 @r{when assigning the result to something.} */ \
4748 @emph{Note:} This construct is only available for C. Furthermore, the
4749 unused expression (@var{exp1} or @var{exp2} depending on the value of
4750 @var{const_exp}) may still generate syntax errors. This may change in
4755 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
4756 You can use the built-in function @code{__builtin_constant_p} to
4757 determine if a value is known to be constant at compile-time and hence
4758 that GCC can perform constant-folding on expressions involving that
4759 value. The argument of the function is the value to test. The function
4760 returns the integer 1 if the argument is known to be a compile-time
4761 constant and 0 if it is not known to be a compile-time constant. A
4762 return of 0 does not indicate that the value is @emph{not} a constant,
4763 but merely that GCC cannot prove it is a constant with the specified
4764 value of the @option{-O} option.
4766 You would typically use this function in an embedded application where
4767 memory was a critical resource. If you have some complex calculation,
4768 you may want it to be folded if it involves constants, but need to call
4769 a function if it does not. For example:
4772 #define Scale_Value(X) \
4773 (__builtin_constant_p (X) \
4774 ? ((X) * SCALE + OFFSET) : Scale (X))
4777 You may use this built-in function in either a macro or an inline
4778 function. However, if you use it in an inlined function and pass an
4779 argument of the function as the argument to the built-in, GCC will
4780 never return 1 when you call the inline function with a string constant
4781 or compound literal (@pxref{Compound Literals}) and will not return 1
4782 when you pass a constant numeric value to the inline function unless you
4783 specify the @option{-O} option.
4785 You may also use @code{__builtin_constant_p} in initializers for static
4786 data. For instance, you can write
4789 static const int table[] = @{
4790 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
4796 This is an acceptable initializer even if @var{EXPRESSION} is not a
4797 constant expression. GCC must be more conservative about evaluating the
4798 built-in in this case, because it has no opportunity to perform
4801 Previous versions of GCC did not accept this built-in in data
4802 initializers. The earliest version where it is completely safe is
4806 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
4807 @opindex fprofile-arcs
4808 You may use @code{__builtin_expect} to provide the compiler with
4809 branch prediction information. In general, you should prefer to
4810 use actual profile feedback for this (@option{-fprofile-arcs}), as
4811 programmers are notoriously bad at predicting how their programs
4812 actually perform. However, there are applications in which this
4813 data is hard to collect.
4815 The return value is the value of @var{exp}, which should be an
4816 integral expression. The value of @var{c} must be a compile-time
4817 constant. The semantics of the built-in are that it is expected
4818 that @var{exp} == @var{c}. For example:
4821 if (__builtin_expect (x, 0))
4826 would indicate that we do not expect to call @code{foo}, since
4827 we expect @code{x} to be zero. Since you are limited to integral
4828 expressions for @var{exp}, you should use constructions such as
4831 if (__builtin_expect (ptr != NULL, 1))
4836 when testing pointer or floating-point values.
4839 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
4840 This function is used to minimize cache-miss latency by moving data into
4841 a cache before it is accessed.
4842 You can insert calls to @code{__builtin_prefetch} into code for which
4843 you know addresses of data in memory that is likely to be accessed soon.
4844 If the target supports them, data prefetch instructions will be generated.
4845 If the prefetch is done early enough before the access then the data will
4846 be in the cache by the time it is accessed.
4848 The value of @var{addr} is the address of the memory to prefetch.
4849 There are two optional arguments, @var{rw} and @var{locality}.
4850 The value of @var{rw} is a compile-time constant one or zero; one
4851 means that the prefetch is preparing for a write to the memory address
4852 and zero, the default, means that the prefetch is preparing for a read.
4853 The value @var{locality} must be a compile-time constant integer between
4854 zero and three. A value of zero means that the data has no temporal
4855 locality, so it need not be left in the cache after the access. A value
4856 of three means that the data has a high degree of temporal locality and
4857 should be left in all levels of cache possible. Values of one and two
4858 mean, respectively, a low or moderate degree of temporal locality. The
4862 for (i = 0; i < n; i++)
4865 __builtin_prefetch (&a[i+j], 1, 1);
4866 __builtin_prefetch (&b[i+j], 0, 1);
4871 Data prefetch does not generate faults if @var{addr} is invalid, but
4872 the address expression itself must be valid. For example, a prefetch
4873 of @code{p->next} will not fault if @code{p->next} is not a valid
4874 address, but evaluation will fault if @code{p} is not a valid address.
4876 If the target does not support data prefetch, the address expression
4877 is evaluated if it includes side effects but no other code is generated
4878 and GCC does not issue a warning.
4881 @deftypefn {Built-in Function} double __builtin_huge_val (void)
4882 Returns a positive infinity, if supported by the floating-point format,
4883 else @code{DBL_MAX}. This function is suitable for implementing the
4884 ISO C macro @code{HUGE_VAL}.
4887 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
4888 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
4891 @deftypefn {Built-in Function} long double __builtin_huge_vall (void)
4892 Similar to @code{__builtin_huge_val}, except the return
4893 type is @code{long double}.
4896 @deftypefn {Built-in Function} double __builtin_inf (void)
4897 Similar to @code{__builtin_huge_val}, except a warning is generated
4898 if the target floating-point format does not support infinities.
4899 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
4902 @deftypefn {Built-in Function} float __builtin_inff (void)
4903 Similar to @code{__builtin_inf}, except the return type is @code{float}.
4906 @deftypefn {Built-in Function} long double __builtin_infl (void)
4907 Similar to @code{__builtin_inf}, except the return
4908 type is @code{long double}.
4911 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
4912 This is an implementation of the ISO C99 function @code{nan}.
4914 Since ISO C99 defines this function in terms of @code{strtod}, which we
4915 do not implement, a description of the parsing is in order. The string
4916 is parsed as by @code{strtol}; that is, the base is recognized by
4917 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
4918 in the significand such that the least significant bit of the number
4919 is at the least significant bit of the significand. The number is
4920 truncated to fit the significand field provided. The significand is
4921 forced to be a quiet NaN.
4923 This function, if given a string literal, is evaluated early enough
4924 that it is considered a compile-time constant.
4927 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
4928 Similar to @code{__builtin_nan}, except the return type is @code{float}.
4931 @deftypefn {Built-in Function} long double __builtin_nanl (const char *str)
4932 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
4935 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
4936 Similar to @code{__builtin_nan}, except the significand is forced
4937 to be a signaling NaN. The @code{nans} function is proposed by
4938 @uref{http://std.dkuug.dk/JTC1/SC22/WG14/www/docs/n965.htm,,WG14 N965}.
4941 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
4942 Similar to @code{__builtin_nans}, except the return type is @code{float}.
4945 @deftypefn {Built-in Function} long double __builtin_nansl (const char *str)
4946 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
4949 @node Target Builtins
4950 @section Built-in Functions Specific to Particular Target Machines
4952 On some target machines, GCC supports many built-in functions specific
4953 to those machines. Generally these generate calls to specific machine
4954 instructions, but allow the compiler to schedule those calls.
4957 * Alpha Built-in Functions::
4958 * X86 Built-in Functions::
4959 * PowerPC AltiVec Built-in Functions::
4962 @node Alpha Built-in Functions
4963 @subsection Alpha Built-in Functions
4965 These built-in functions are available for the Alpha family of
4966 processors, depending on the command-line switches used.
4968 The following built-in functions are always available. They
4969 all generate the machine instruction that is part of the name.
4972 long __builtin_alpha_implver (void)
4973 long __builtin_alpha_rpcc (void)
4974 long __builtin_alpha_amask (long)
4975 long __builtin_alpha_cmpbge (long, long)
4976 long __builtin_alpha_extbl (long, long)
4977 long __builtin_alpha_extwl (long, long)
4978 long __builtin_alpha_extll (long, long)
4979 long __builtin_alpha_extql (long, long)
4980 long __builtin_alpha_extwh (long, long)
4981 long __builtin_alpha_extlh (long, long)
4982 long __builtin_alpha_extqh (long, long)
4983 long __builtin_alpha_insbl (long, long)
4984 long __builtin_alpha_inswl (long, long)
4985 long __builtin_alpha_insll (long, long)
4986 long __builtin_alpha_insql (long, long)
4987 long __builtin_alpha_inswh (long, long)
4988 long __builtin_alpha_inslh (long, long)
4989 long __builtin_alpha_insqh (long, long)
4990 long __builtin_alpha_mskbl (long, long)
4991 long __builtin_alpha_mskwl (long, long)
4992 long __builtin_alpha_mskll (long, long)
4993 long __builtin_alpha_mskql (long, long)
4994 long __builtin_alpha_mskwh (long, long)
4995 long __builtin_alpha_msklh (long, long)
4996 long __builtin_alpha_mskqh (long, long)
4997 long __builtin_alpha_umulh (long, long)
4998 long __builtin_alpha_zap (long, long)
4999 long __builtin_alpha_zapnot (long, long)
5002 The following built-in functions are always with @option{-mmax}
5003 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
5004 later. They all generate the machine instruction that is part
5008 long __builtin_alpha_pklb (long)
5009 long __builtin_alpha_pkwb (long)
5010 long __builtin_alpha_unpkbl (long)
5011 long __builtin_alpha_unpkbw (long)
5012 long __builtin_alpha_minub8 (long, long)
5013 long __builtin_alpha_minsb8 (long, long)
5014 long __builtin_alpha_minuw4 (long, long)
5015 long __builtin_alpha_minsw4 (long, long)
5016 long __builtin_alpha_maxub8 (long, long)
5017 long __builtin_alpha_maxsb8 (long, long)
5018 long __builtin_alpha_maxuw4 (long, long)
5019 long __builtin_alpha_maxsw4 (long, long)
5020 long __builtin_alpha_perr (long, long)
5023 The following built-in functions are always with @option{-mcix}
5024 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
5025 later. They all generate the machine instruction that is part
5029 long __builtin_alpha_cttz (long)
5030 long __builtin_alpha_ctlz (long)
5031 long __builtin_alpha_ctpop (long)
5034 The following builtins are available on systems that use the OSF/1
5035 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
5036 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
5037 @code{rdval} and @code{wrval}.
5040 void *__builtin_thread_pointer (void)
5041 void __builtin_set_thread_pointer (void *)
5044 @node X86 Built-in Functions
5045 @subsection X86 Built-in Functions
5047 These built-in functions are available for the i386 and x86-64 family
5048 of computers, depending on the command-line switches used.
5050 The following machine modes are available for use with MMX built-in functions
5051 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
5052 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
5053 vector of eight 8-bit integers. Some of the built-in functions operate on
5054 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
5056 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
5057 of two 32-bit floating point values.
5059 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
5060 floating point values. Some instructions use a vector of four 32-bit
5061 integers, these use @code{V4SI}. Finally, some instructions operate on an
5062 entire vector register, interpreting it as a 128-bit integer, these use mode
5065 The following built-in functions are made available by @option{-mmmx}.
5066 All of them generate the machine instruction that is part of the name.
5069 v8qi __builtin_ia32_paddb (v8qi, v8qi)
5070 v4hi __builtin_ia32_paddw (v4hi, v4hi)
5071 v2si __builtin_ia32_paddd (v2si, v2si)
5072 v8qi __builtin_ia32_psubb (v8qi, v8qi)
5073 v4hi __builtin_ia32_psubw (v4hi, v4hi)
5074 v2si __builtin_ia32_psubd (v2si, v2si)
5075 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
5076 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
5077 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
5078 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
5079 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
5080 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
5081 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
5082 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
5083 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
5084 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
5085 di __builtin_ia32_pand (di, di)
5086 di __builtin_ia32_pandn (di,di)
5087 di __builtin_ia32_por (di, di)
5088 di __builtin_ia32_pxor (di, di)
5089 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
5090 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
5091 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
5092 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
5093 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
5094 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
5095 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
5096 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
5097 v2si __builtin_ia32_punpckhdq (v2si, v2si)
5098 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
5099 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
5100 v2si __builtin_ia32_punpckldq (v2si, v2si)
5101 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
5102 v4hi __builtin_ia32_packssdw (v2si, v2si)
5103 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
5106 The following built-in functions are made available either with
5107 @option{-msse}, or with a combination of @option{-m3dnow} and
5108 @option{-march=athlon}. All of them generate the machine
5109 instruction that is part of the name.
5112 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
5113 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
5114 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
5115 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
5116 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
5117 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
5118 v8qi __builtin_ia32_pminub (v8qi, v8qi)
5119 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
5120 int __builtin_ia32_pextrw (v4hi, int)
5121 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
5122 int __builtin_ia32_pmovmskb (v8qi)
5123 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
5124 void __builtin_ia32_movntq (di *, di)
5125 void __builtin_ia32_sfence (void)
5128 The following built-in functions are available when @option{-msse} is used.
5129 All of them generate the machine instruction that is part of the name.
5132 int __builtin_ia32_comieq (v4sf, v4sf)
5133 int __builtin_ia32_comineq (v4sf, v4sf)
5134 int __builtin_ia32_comilt (v4sf, v4sf)
5135 int __builtin_ia32_comile (v4sf, v4sf)
5136 int __builtin_ia32_comigt (v4sf, v4sf)
5137 int __builtin_ia32_comige (v4sf, v4sf)
5138 int __builtin_ia32_ucomieq (v4sf, v4sf)
5139 int __builtin_ia32_ucomineq (v4sf, v4sf)
5140 int __builtin_ia32_ucomilt (v4sf, v4sf)
5141 int __builtin_ia32_ucomile (v4sf, v4sf)
5142 int __builtin_ia32_ucomigt (v4sf, v4sf)
5143 int __builtin_ia32_ucomige (v4sf, v4sf)
5144 v4sf __builtin_ia32_addps (v4sf, v4sf)
5145 v4sf __builtin_ia32_subps (v4sf, v4sf)
5146 v4sf __builtin_ia32_mulps (v4sf, v4sf)
5147 v4sf __builtin_ia32_divps (v4sf, v4sf)
5148 v4sf __builtin_ia32_addss (v4sf, v4sf)
5149 v4sf __builtin_ia32_subss (v4sf, v4sf)
5150 v4sf __builtin_ia32_mulss (v4sf, v4sf)
5151 v4sf __builtin_ia32_divss (v4sf, v4sf)
5152 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
5153 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
5154 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
5155 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
5156 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
5157 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
5158 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
5159 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
5160 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
5161 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
5162 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
5163 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
5164 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
5165 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
5166 v4si __builtin_ia32_cmpless (v4sf, v4sf)
5167 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
5168 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
5169 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
5170 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
5171 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
5172 v4sf __builtin_ia32_maxps (v4sf, v4sf)
5173 v4sf __builtin_ia32_maxss (v4sf, v4sf)
5174 v4sf __builtin_ia32_minps (v4sf, v4sf)
5175 v4sf __builtin_ia32_minss (v4sf, v4sf)
5176 v4sf __builtin_ia32_andps (v4sf, v4sf)
5177 v4sf __builtin_ia32_andnps (v4sf, v4sf)
5178 v4sf __builtin_ia32_orps (v4sf, v4sf)
5179 v4sf __builtin_ia32_xorps (v4sf, v4sf)
5180 v4sf __builtin_ia32_movss (v4sf, v4sf)
5181 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
5182 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
5183 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
5184 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
5185 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
5186 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
5187 v2si __builtin_ia32_cvtps2pi (v4sf)
5188 int __builtin_ia32_cvtss2si (v4sf)
5189 v2si __builtin_ia32_cvttps2pi (v4sf)
5190 int __builtin_ia32_cvttss2si (v4sf)
5191 v4sf __builtin_ia32_rcpps (v4sf)
5192 v4sf __builtin_ia32_rsqrtps (v4sf)
5193 v4sf __builtin_ia32_sqrtps (v4sf)
5194 v4sf __builtin_ia32_rcpss (v4sf)
5195 v4sf __builtin_ia32_rsqrtss (v4sf)
5196 v4sf __builtin_ia32_sqrtss (v4sf)
5197 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
5198 void __builtin_ia32_movntps (float *, v4sf)
5199 int __builtin_ia32_movmskps (v4sf)
5202 The following built-in functions are available when @option{-msse} is used.
5205 @item v4sf __builtin_ia32_loadaps (float *)
5206 Generates the @code{movaps} machine instruction as a load from memory.
5207 @item void __builtin_ia32_storeaps (float *, v4sf)
5208 Generates the @code{movaps} machine instruction as a store to memory.
5209 @item v4sf __builtin_ia32_loadups (float *)
5210 Generates the @code{movups} machine instruction as a load from memory.
5211 @item void __builtin_ia32_storeups (float *, v4sf)
5212 Generates the @code{movups} machine instruction as a store to memory.
5213 @item v4sf __builtin_ia32_loadsss (float *)
5214 Generates the @code{movss} machine instruction as a load from memory.
5215 @item void __builtin_ia32_storess (float *, v4sf)
5216 Generates the @code{movss} machine instruction as a store to memory.
5217 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
5218 Generates the @code{movhps} machine instruction as a load from memory.
5219 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
5220 Generates the @code{movlps} machine instruction as a load from memory
5221 @item void __builtin_ia32_storehps (v4sf, v2si *)
5222 Generates the @code{movhps} machine instruction as a store to memory.
5223 @item void __builtin_ia32_storelps (v4sf, v2si *)
5224 Generates the @code{movlps} machine instruction as a store to memory.
5227 The following built-in functions are available when @option{-m3dnow} is used.
5228 All of them generate the machine instruction that is part of the name.
5231 void __builtin_ia32_femms (void)
5232 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
5233 v2si __builtin_ia32_pf2id (v2sf)
5234 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
5235 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
5236 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
5237 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
5238 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
5239 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
5240 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
5241 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
5242 v2sf __builtin_ia32_pfrcp (v2sf)
5243 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
5244 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
5245 v2sf __builtin_ia32_pfrsqrt (v2sf)
5246 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
5247 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
5248 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
5249 v2sf __builtin_ia32_pi2fd (v2si)
5250 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
5253 The following built-in functions are available when both @option{-m3dnow}
5254 and @option{-march=athlon} are used. All of them generate the machine
5255 instruction that is part of the name.
5258 v2si __builtin_ia32_pf2iw (v2sf)
5259 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
5260 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
5261 v2sf __builtin_ia32_pi2fw (v2si)
5262 v2sf __builtin_ia32_pswapdsf (v2sf)
5263 v2si __builtin_ia32_pswapdsi (v2si)
5266 @node PowerPC AltiVec Built-in Functions
5267 @subsection PowerPC AltiVec Built-in Functions
5269 These built-in functions are available for the PowerPC family
5270 of computers, depending on the command-line switches used.
5272 The following machine modes are available for use with AltiVec built-in
5273 functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four
5274 32-bit integers, @code{V4SF} for a vector of four 32-bit floating point
5275 numbers, @code{V8HI} for a vector of eight 16-bit integers, and
5276 @code{V16QI} for a vector of sixteen 8-bit integers.
5278 The following functions are made available by including
5279 @code{<altivec.h>} and using @option{-maltivec} and
5280 @option{-mabi=altivec}. The functions implement the functionality
5281 described in Motorola's AltiVec Programming Interface Manual.
5283 There are a few differences from Motorola's documentation and GCC's
5284 implementation. Vector constants are done with curly braces (not
5285 parentheses). Vector initializers require no casts if the vector
5286 constant is of the same type as the variable it is initializing. The
5287 @code{vector bool} type is deprecated and will be discontinued in
5288 further revisions. Use @code{vector signed} instead. If @code{signed}
5289 or @code{unsigned} is omitted, the vector type will default to
5290 @code{signed}. Lastly, all overloaded functions are implemented with macros
5291 for the C implementation. So code the following example will not work:
5294 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
5297 Since vec_add is a macro, the vector constant in the above example will
5298 be treated as four different arguments. Wrap the entire argument in
5299 parentheses for this to work. The C++ implementation does not use
5302 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
5303 Internally, GCC uses built-in functions to achieve the functionality in
5304 the aforementioned header file, but they are not supported and are
5305 subject to change without notice.
5308 vector signed char vec_abs (vector signed char, vector signed char);
5309 vector signed short vec_abs (vector signed short, vector signed short);
5310 vector signed int vec_abs (vector signed int, vector signed int);
5311 vector signed float vec_abs (vector signed float, vector signed float);
5313 vector signed char vec_abss (vector signed char, vector signed char);
5314 vector signed short vec_abss (vector signed short, vector signed short);
5316 vector signed char vec_add (vector signed char, vector signed char);
5317 vector unsigned char vec_add (vector signed char, vector unsigned char);
5319 vector unsigned char vec_add (vector unsigned char, vector signed char);
5321 vector unsigned char vec_add (vector unsigned char,
5322 vector unsigned char);
5323 vector signed short vec_add (vector signed short, vector signed short);
5324 vector unsigned short vec_add (vector signed short,
5325 vector unsigned short);
5326 vector unsigned short vec_add (vector unsigned short,
5327 vector signed short);
5328 vector unsigned short vec_add (vector unsigned short,
5329 vector unsigned short);
5330 vector signed int vec_add (vector signed int, vector signed int);
5331 vector unsigned int vec_add (vector signed int, vector unsigned int);
5332 vector unsigned int vec_add (vector unsigned int, vector signed int);
5333 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
5334 vector float vec_add (vector float, vector float);
5336 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
5338 vector unsigned char vec_adds (vector signed char,
5339 vector unsigned char);
5340 vector unsigned char vec_adds (vector unsigned char,
5341 vector signed char);
5342 vector unsigned char vec_adds (vector unsigned char,
5343 vector unsigned char);
5344 vector signed char vec_adds (vector signed char, vector signed char);
5345 vector unsigned short vec_adds (vector signed short,
5346 vector unsigned short);
5347 vector unsigned short vec_adds (vector unsigned short,
5348 vector signed short);
5349 vector unsigned short vec_adds (vector unsigned short,
5350 vector unsigned short);
5351 vector signed short vec_adds (vector signed short, vector signed short);
5353 vector unsigned int vec_adds (vector signed int, vector unsigned int);
5354 vector unsigned int vec_adds (vector unsigned int, vector signed int);
5355 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
5357 vector signed int vec_adds (vector signed int, vector signed int);
5359 vector float vec_and (vector float, vector float);
5360 vector float vec_and (vector float, vector signed int);
5361 vector float vec_and (vector signed int, vector float);
5362 vector signed int vec_and (vector signed int, vector signed int);
5363 vector unsigned int vec_and (vector signed int, vector unsigned int);
5364 vector unsigned int vec_and (vector unsigned int, vector signed int);
5365 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
5366 vector signed short vec_and (vector signed short, vector signed short);
5367 vector unsigned short vec_and (vector signed short,
5368 vector unsigned short);
5369 vector unsigned short vec_and (vector unsigned short,
5370 vector signed short);
5371 vector unsigned short vec_and (vector unsigned short,
5372 vector unsigned short);
5373 vector signed char vec_and (vector signed char, vector signed char);
5374 vector unsigned char vec_and (vector signed char, vector unsigned char);
5376 vector unsigned char vec_and (vector unsigned char, vector signed char);
5378 vector unsigned char vec_and (vector unsigned char,
5379 vector unsigned char);
5381 vector float vec_andc (vector float, vector float);
5382 vector float vec_andc (vector float, vector signed int);
5383 vector float vec_andc (vector signed int, vector float);
5384 vector signed int vec_andc (vector signed int, vector signed int);
5385 vector unsigned int vec_andc (vector signed int, vector unsigned int);
5386 vector unsigned int vec_andc (vector unsigned int, vector signed int);
5387 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
5389 vector signed short vec_andc (vector signed short, vector signed short);
5391 vector unsigned short vec_andc (vector signed short,
5392 vector unsigned short);
5393 vector unsigned short vec_andc (vector unsigned short,
5394 vector signed short);
5395 vector unsigned short vec_andc (vector unsigned short,
5396 vector unsigned short);
5397 vector signed char vec_andc (vector signed char, vector signed char);
5398 vector unsigned char vec_andc (vector signed char,
5399 vector unsigned char);
5400 vector unsigned char vec_andc (vector unsigned char,
5401 vector signed char);
5402 vector unsigned char vec_andc (vector unsigned char,
5403 vector unsigned char);
5405 vector unsigned char vec_avg (vector unsigned char,
5406 vector unsigned char);
5407 vector signed char vec_avg (vector signed char, vector signed char);
5408 vector unsigned short vec_avg (vector unsigned short,
5409 vector unsigned short);
5410 vector signed short vec_avg (vector signed short, vector signed short);
5411 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
5412 vector signed int vec_avg (vector signed int, vector signed int);
5414 vector float vec_ceil (vector float);
5416 vector signed int vec_cmpb (vector float, vector float);
5418 vector signed char vec_cmpeq (vector signed char, vector signed char);
5419 vector signed char vec_cmpeq (vector unsigned char,
5420 vector unsigned char);
5421 vector signed short vec_cmpeq (vector signed short,
5422 vector signed short);
5423 vector signed short vec_cmpeq (vector unsigned short,
5424 vector unsigned short);
5425 vector signed int vec_cmpeq (vector signed int, vector signed int);
5426 vector signed int vec_cmpeq (vector unsigned int, vector unsigned int);
5427 vector signed int vec_cmpeq (vector float, vector float);
5429 vector signed int vec_cmpge (vector float, vector float);
5431 vector signed char vec_cmpgt (vector unsigned char,
5432 vector unsigned char);
5433 vector signed char vec_cmpgt (vector signed char, vector signed char);
5434 vector signed short vec_cmpgt (vector unsigned short,
5435 vector unsigned short);
5436 vector signed short vec_cmpgt (vector signed short,
5437 vector signed short);
5438 vector signed int vec_cmpgt (vector unsigned int, vector unsigned int);
5439 vector signed int vec_cmpgt (vector signed int, vector signed int);
5440 vector signed int vec_cmpgt (vector float, vector float);
5442 vector signed int vec_cmple (vector float, vector float);
5444 vector signed char vec_cmplt (vector unsigned char,
5445 vector unsigned char);
5446 vector signed char vec_cmplt (vector signed char, vector signed char);
5447 vector signed short vec_cmplt (vector unsigned short,
5448 vector unsigned short);
5449 vector signed short vec_cmplt (vector signed short,
5450 vector signed short);
5451 vector signed int vec_cmplt (vector unsigned int, vector unsigned int);
5452 vector signed int vec_cmplt (vector signed int, vector signed int);
5453 vector signed int vec_cmplt (vector float, vector float);
5455 vector float vec_ctf (vector unsigned int, const char);
5456 vector float vec_ctf (vector signed int, const char);
5458 vector signed int vec_cts (vector float, const char);
5460 vector unsigned int vec_ctu (vector float, const char);
5462 void vec_dss (const char);
5464 void vec_dssall (void);
5466 void vec_dst (void *, int, const char);
5468 void vec_dstst (void *, int, const char);
5470 void vec_dststt (void *, int, const char);
5472 void vec_dstt (void *, int, const char);
5474 vector float vec_expte (vector float, vector float);
5476 vector float vec_floor (vector float, vector float);
5478 vector float vec_ld (int, vector float *);
5479 vector float vec_ld (int, float *):
5480 vector signed int vec_ld (int, int *);
5481 vector signed int vec_ld (int, vector signed int *);
5482 vector unsigned int vec_ld (int, vector unsigned int *);
5483 vector unsigned int vec_ld (int, unsigned int *);
5484 vector signed short vec_ld (int, short *, vector signed short *);
5485 vector unsigned short vec_ld (int, unsigned short *,
5486 vector unsigned short *);
5487 vector signed char vec_ld (int, signed char *);
5488 vector signed char vec_ld (int, vector signed char *);
5489 vector unsigned char vec_ld (int, unsigned char *);
5490 vector unsigned char vec_ld (int, vector unsigned char *);
5492 vector signed char vec_lde (int, signed char *);
5493 vector unsigned char vec_lde (int, unsigned char *);
5494 vector signed short vec_lde (int, short *);
5495 vector unsigned short vec_lde (int, unsigned short *);
5496 vector float vec_lde (int, float *);
5497 vector signed int vec_lde (int, int *);
5498 vector unsigned int vec_lde (int, unsigned int *);
5500 void float vec_ldl (int, float *);
5501 void float vec_ldl (int, vector float *);
5502 void signed int vec_ldl (int, vector signed int *);
5503 void signed int vec_ldl (int, int *);
5504 void unsigned int vec_ldl (int, unsigned int *);
5505 void unsigned int vec_ldl (int, vector unsigned int *);
5506 void signed short vec_ldl (int, vector signed short *);
5507 void signed short vec_ldl (int, short *);
5508 void unsigned short vec_ldl (int, vector unsigned short *);
5509 void unsigned short vec_ldl (int, unsigned short *);
5510 void signed char vec_ldl (int, vector signed char *);
5511 void signed char vec_ldl (int, signed char *);
5512 void unsigned char vec_ldl (int, vector unsigned char *);
5513 void unsigned char vec_ldl (int, unsigned char *);
5515 vector float vec_loge (vector float);
5517 vector unsigned char vec_lvsl (int, void *, int *);
5519 vector unsigned char vec_lvsr (int, void *, int *);
5521 vector float vec_madd (vector float, vector float, vector float);
5523 vector signed short vec_madds (vector signed short, vector signed short,
5524 vector signed short);
5526 vector unsigned char vec_max (vector signed char, vector unsigned char);
5528 vector unsigned char vec_max (vector unsigned char, vector signed char);
5530 vector unsigned char vec_max (vector unsigned char,
5531 vector unsigned char);
5532 vector signed char vec_max (vector signed char, vector signed char);
5533 vector unsigned short vec_max (vector signed short,
5534 vector unsigned short);
5535 vector unsigned short vec_max (vector unsigned short,
5536 vector signed short);
5537 vector unsigned short vec_max (vector unsigned short,
5538 vector unsigned short);
5539 vector signed short vec_max (vector signed short, vector signed short);
5540 vector unsigned int vec_max (vector signed int, vector unsigned int);
5541 vector unsigned int vec_max (vector unsigned int, vector signed int);
5542 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
5543 vector signed int vec_max (vector signed int, vector signed int);
5544 vector float vec_max (vector float, vector float);
5546 vector signed char vec_mergeh (vector signed char, vector signed char);
5547 vector unsigned char vec_mergeh (vector unsigned char,
5548 vector unsigned char);
5549 vector signed short vec_mergeh (vector signed short,
5550 vector signed short);
5551 vector unsigned short vec_mergeh (vector unsigned short,
5552 vector unsigned short);
5553 vector float vec_mergeh (vector float, vector float);
5554 vector signed int vec_mergeh (vector signed int, vector signed int);
5555 vector unsigned int vec_mergeh (vector unsigned int,
5556 vector unsigned int);
5558 vector signed char vec_mergel (vector signed char, vector signed char);
5559 vector unsigned char vec_mergel (vector unsigned char,
5560 vector unsigned char);
5561 vector signed short vec_mergel (vector signed short,
5562 vector signed short);
5563 vector unsigned short vec_mergel (vector unsigned short,
5564 vector unsigned short);
5565 vector float vec_mergel (vector float, vector float);
5566 vector signed int vec_mergel (vector signed int, vector signed int);
5567 vector unsigned int vec_mergel (vector unsigned int,
5568 vector unsigned int);
5570 vector unsigned short vec_mfvscr (void);
5572 vector unsigned char vec_min (vector signed char, vector unsigned char);
5574 vector unsigned char vec_min (vector unsigned char, vector signed char);
5576 vector unsigned char vec_min (vector unsigned char,
5577 vector unsigned char);
5578 vector signed char vec_min (vector signed char, vector signed char);
5579 vector unsigned short vec_min (vector signed short,
5580 vector unsigned short);
5581 vector unsigned short vec_min (vector unsigned short,
5582 vector signed short);
5583 vector unsigned short vec_min (vector unsigned short,
5584 vector unsigned short);
5585 vector signed short vec_min (vector signed short, vector signed short);
5586 vector unsigned int vec_min (vector signed int, vector unsigned int);
5587 vector unsigned int vec_min (vector unsigned int, vector signed int);
5588 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
5589 vector signed int vec_min (vector signed int, vector signed int);
5590 vector float vec_min (vector float, vector float);
5592 vector signed short vec_mladd (vector signed short, vector signed short,
5593 vector signed short);
5594 vector signed short vec_mladd (vector signed short,
5595 vector unsigned short,
5596 vector unsigned short);
5597 vector signed short vec_mladd (vector unsigned short,
5598 vector signed short,
5599 vector signed short);
5600 vector unsigned short vec_mladd (vector unsigned short,
5601 vector unsigned short,
5602 vector unsigned short);
5604 vector signed short vec_mradds (vector signed short,
5605 vector signed short,
5606 vector signed short);
5608 vector unsigned int vec_msum (vector unsigned char,
5609 vector unsigned char,
5610 vector unsigned int);
5611 vector signed int vec_msum (vector signed char, vector unsigned char,
5613 vector unsigned int vec_msum (vector unsigned short,
5614 vector unsigned short,
5615 vector unsigned int);
5616 vector signed int vec_msum (vector signed short, vector signed short,
5619 vector unsigned int vec_msums (vector unsigned short,
5620 vector unsigned short,
5621 vector unsigned int);
5622 vector signed int vec_msums (vector signed short, vector signed short,
5625 void vec_mtvscr (vector signed int);
5626 void vec_mtvscr (vector unsigned int);
5627 void vec_mtvscr (vector signed short);
5628 void vec_mtvscr (vector unsigned short);
5629 void vec_mtvscr (vector signed char);
5630 void vec_mtvscr (vector unsigned char);
5632 vector unsigned short vec_mule (vector unsigned char,
5633 vector unsigned char);
5634 vector signed short vec_mule (vector signed char, vector signed char);
5635 vector unsigned int vec_mule (vector unsigned short,
5636 vector unsigned short);
5637 vector signed int vec_mule (vector signed short, vector signed short);
5639 vector unsigned short vec_mulo (vector unsigned char,
5640 vector unsigned char);
5641 vector signed short vec_mulo (vector signed char, vector signed char);
5642 vector unsigned int vec_mulo (vector unsigned short,
5643 vector unsigned short);
5644 vector signed int vec_mulo (vector signed short, vector signed short);
5646 vector float vec_nmsub (vector float, vector float, vector float);
5648 vector float vec_nor (vector float, vector float);
5649 vector signed int vec_nor (vector signed int, vector signed int);
5650 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
5651 vector signed short vec_nor (vector signed short, vector signed short);
5652 vector unsigned short vec_nor (vector unsigned short,
5653 vector unsigned short);
5654 vector signed char vec_nor (vector signed char, vector signed char);
5655 vector unsigned char vec_nor (vector unsigned char,
5656 vector unsigned char);
5658 vector float vec_or (vector float, vector float);
5659 vector float vec_or (vector float, vector signed int);
5660 vector float vec_or (vector signed int, vector float);
5661 vector signed int vec_or (vector signed int, vector signed int);
5662 vector unsigned int vec_or (vector signed int, vector unsigned int);
5663 vector unsigned int vec_or (vector unsigned int, vector signed int);
5664 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
5665 vector signed short vec_or (vector signed short, vector signed short);
5666 vector unsigned short vec_or (vector signed short,
5667 vector unsigned short);
5668 vector unsigned short vec_or (vector unsigned short,
5669 vector signed short);
5670 vector unsigned short vec_or (vector unsigned short,
5671 vector unsigned short);
5672 vector signed char vec_or (vector signed char, vector signed char);
5673 vector unsigned char vec_or (vector signed char, vector unsigned char);
5674 vector unsigned char vec_or (vector unsigned char, vector signed char);
5675 vector unsigned char vec_or (vector unsigned char,
5676 vector unsigned char);
5678 vector signed char vec_pack (vector signed short, vector signed short);
5679 vector unsigned char vec_pack (vector unsigned short,
5680 vector unsigned short);
5681 vector signed short vec_pack (vector signed int, vector signed int);
5682 vector unsigned short vec_pack (vector unsigned int,
5683 vector unsigned int);
5685 vector signed short vec_packpx (vector unsigned int,
5686 vector unsigned int);
5688 vector unsigned char vec_packs (vector unsigned short,
5689 vector unsigned short);
5690 vector signed char vec_packs (vector signed short, vector signed short);
5692 vector unsigned short vec_packs (vector unsigned int,
5693 vector unsigned int);
5694 vector signed short vec_packs (vector signed int, vector signed int);
5696 vector unsigned char vec_packsu (vector unsigned short,
5697 vector unsigned short);
5698 vector unsigned char vec_packsu (vector signed short,
5699 vector signed short);
5700 vector unsigned short vec_packsu (vector unsigned int,
5701 vector unsigned int);
5702 vector unsigned short vec_packsu (vector signed int, vector signed int);
5704 vector float vec_perm (vector float, vector float,
5705 vector unsigned char);
5706 vector signed int vec_perm (vector signed int, vector signed int,
5707 vector unsigned char);
5708 vector unsigned int vec_perm (vector unsigned int, vector unsigned int,
5709 vector unsigned char);
5710 vector signed short vec_perm (vector signed short, vector signed short,
5711 vector unsigned char);
5712 vector unsigned short vec_perm (vector unsigned short,
5713 vector unsigned short,
5714 vector unsigned char);
5715 vector signed char vec_perm (vector signed char, vector signed char,
5716 vector unsigned char);
5717 vector unsigned char vec_perm (vector unsigned char,
5718 vector unsigned char,
5719 vector unsigned char);
5721 vector float vec_re (vector float);
5723 vector signed char vec_rl (vector signed char, vector unsigned char);
5724 vector unsigned char vec_rl (vector unsigned char,
5725 vector unsigned char);
5726 vector signed short vec_rl (vector signed short, vector unsigned short);
5728 vector unsigned short vec_rl (vector unsigned short,
5729 vector unsigned short);
5730 vector signed int vec_rl (vector signed int, vector unsigned int);
5731 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
5733 vector float vec_round (vector float);
5735 vector float vec_rsqrte (vector float);
5737 vector float vec_sel (vector float, vector float, vector signed int);
5738 vector float vec_sel (vector float, vector float, vector unsigned int);
5739 vector signed int vec_sel (vector signed int, vector signed int,
5741 vector signed int vec_sel (vector signed int, vector signed int,
5742 vector unsigned int);
5743 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
5745 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
5746 vector unsigned int);
5747 vector signed short vec_sel (vector signed short, vector signed short,
5748 vector signed short);
5749 vector signed short vec_sel (vector signed short, vector signed short,
5750 vector unsigned short);
5751 vector unsigned short vec_sel (vector unsigned short,
5752 vector unsigned short,
5753 vector signed short);
5754 vector unsigned short vec_sel (vector unsigned short,
5755 vector unsigned short,
5756 vector unsigned short);
5757 vector signed char vec_sel (vector signed char, vector signed char,
5758 vector signed char);
5759 vector signed char vec_sel (vector signed char, vector signed char,
5760 vector unsigned char);
5761 vector unsigned char vec_sel (vector unsigned char,
5762 vector unsigned char,
5763 vector signed char);
5764 vector unsigned char vec_sel (vector unsigned char,
5765 vector unsigned char,
5766 vector unsigned char);
5768 vector signed char vec_sl (vector signed char, vector unsigned char);
5769 vector unsigned char vec_sl (vector unsigned char,
5770 vector unsigned char);
5771 vector signed short vec_sl (vector signed short, vector unsigned short);
5773 vector unsigned short vec_sl (vector unsigned short,
5774 vector unsigned short);
5775 vector signed int vec_sl (vector signed int, vector unsigned int);
5776 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
5778 vector float vec_sld (vector float, vector float, const char);
5779 vector signed int vec_sld (vector signed int, vector signed int,
5781 vector unsigned int vec_sld (vector unsigned int, vector unsigned int,
5783 vector signed short vec_sld (vector signed short, vector signed short,
5785 vector unsigned short vec_sld (vector unsigned short,
5786 vector unsigned short, const char);
5787 vector signed char vec_sld (vector signed char, vector signed char,
5789 vector unsigned char vec_sld (vector unsigned char,
5790 vector unsigned char,
5793 vector signed int vec_sll (vector signed int, vector unsigned int);
5794 vector signed int vec_sll (vector signed int, vector unsigned short);
5795 vector signed int vec_sll (vector signed int, vector unsigned char);
5796 vector unsigned int vec_sll (vector unsigned int, vector unsigned int);
5797 vector unsigned int vec_sll (vector unsigned int,
5798 vector unsigned short);
5799 vector unsigned int vec_sll (vector unsigned int, vector unsigned char);
5801 vector signed short vec_sll (vector signed short, vector unsigned int);
5802 vector signed short vec_sll (vector signed short,
5803 vector unsigned short);
5804 vector signed short vec_sll (vector signed short, vector unsigned char);
5806 vector unsigned short vec_sll (vector unsigned short,
5807 vector unsigned int);
5808 vector unsigned short vec_sll (vector unsigned short,
5809 vector unsigned short);
5810 vector unsigned short vec_sll (vector unsigned short,
5811 vector unsigned char);
5812 vector signed char vec_sll (vector signed char, vector unsigned int);
5813 vector signed char vec_sll (vector signed char, vector unsigned short);
5814 vector signed char vec_sll (vector signed char, vector unsigned char);
5815 vector unsigned char vec_sll (vector unsigned char,
5816 vector unsigned int);
5817 vector unsigned char vec_sll (vector unsigned char,
5818 vector unsigned short);
5819 vector unsigned char vec_sll (vector unsigned char,
5820 vector unsigned char);
5822 vector float vec_slo (vector float, vector signed char);
5823 vector float vec_slo (vector float, vector unsigned char);
5824 vector signed int vec_slo (vector signed int, vector signed char);
5825 vector signed int vec_slo (vector signed int, vector unsigned char);
5826 vector unsigned int vec_slo (vector unsigned int, vector signed char);
5827 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
5829 vector signed short vec_slo (vector signed short, vector signed char);
5830 vector signed short vec_slo (vector signed short, vector unsigned char);
5832 vector unsigned short vec_slo (vector unsigned short,
5833 vector signed char);
5834 vector unsigned short vec_slo (vector unsigned short,
5835 vector unsigned char);
5836 vector signed char vec_slo (vector signed char, vector signed char);
5837 vector signed char vec_slo (vector signed char, vector unsigned char);
5838 vector unsigned char vec_slo (vector unsigned char, vector signed char);
5840 vector unsigned char vec_slo (vector unsigned char,
5841 vector unsigned char);
5843 vector signed char vec_splat (vector signed char, const char);
5844 vector unsigned char vec_splat (vector unsigned char, const char);
5845 vector signed short vec_splat (vector signed short, const char);
5846 vector unsigned short vec_splat (vector unsigned short, const char);
5847 vector float vec_splat (vector float, const char);
5848 vector signed int vec_splat (vector signed int, const char);
5849 vector unsigned int vec_splat (vector unsigned int, const char);
5851 vector signed char vec_splat_s8 (const char);
5853 vector signed short vec_splat_s16 (const char);
5855 vector signed int vec_splat_s32 (const char);
5857 vector unsigned char vec_splat_u8 (const char);
5859 vector unsigned short vec_splat_u16 (const char);
5861 vector unsigned int vec_splat_u32 (const char);
5863 vector signed char vec_sr (vector signed char, vector unsigned char);
5864 vector unsigned char vec_sr (vector unsigned char,
5865 vector unsigned char);
5866 vector signed short vec_sr (vector signed short, vector unsigned short);
5868 vector unsigned short vec_sr (vector unsigned short,
5869 vector unsigned short);
5870 vector signed int vec_sr (vector signed int, vector unsigned int);
5871 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
5873 vector signed char vec_sra (vector signed char, vector unsigned char);
5874 vector unsigned char vec_sra (vector unsigned char,
5875 vector unsigned char);
5876 vector signed short vec_sra (vector signed short,
5877 vector unsigned short);
5878 vector unsigned short vec_sra (vector unsigned short,
5879 vector unsigned short);
5880 vector signed int vec_sra (vector signed int, vector unsigned int);
5881 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
5883 vector signed int vec_srl (vector signed int, vector unsigned int);
5884 vector signed int vec_srl (vector signed int, vector unsigned short);
5885 vector signed int vec_srl (vector signed int, vector unsigned char);
5886 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
5887 vector unsigned int vec_srl (vector unsigned int,
5888 vector unsigned short);
5889 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
5891 vector signed short vec_srl (vector signed short, vector unsigned int);
5892 vector signed short vec_srl (vector signed short,
5893 vector unsigned short);
5894 vector signed short vec_srl (vector signed short, vector unsigned char);
5896 vector unsigned short vec_srl (vector unsigned short,
5897 vector unsigned int);
5898 vector unsigned short vec_srl (vector unsigned short,
5899 vector unsigned short);
5900 vector unsigned short vec_srl (vector unsigned short,
5901 vector unsigned char);
5902 vector signed char vec_srl (vector signed char, vector unsigned int);
5903 vector signed char vec_srl (vector signed char, vector unsigned short);
5904 vector signed char vec_srl (vector signed char, vector unsigned char);
5905 vector unsigned char vec_srl (vector unsigned char,
5906 vector unsigned int);
5907 vector unsigned char vec_srl (vector unsigned char,
5908 vector unsigned short);
5909 vector unsigned char vec_srl (vector unsigned char,
5910 vector unsigned char);
5912 vector float vec_sro (vector float, vector signed char);
5913 vector float vec_sro (vector float, vector unsigned char);
5914 vector signed int vec_sro (vector signed int, vector signed char);
5915 vector signed int vec_sro (vector signed int, vector unsigned char);
5916 vector unsigned int vec_sro (vector unsigned int, vector signed char);
5917 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
5919 vector signed short vec_sro (vector signed short, vector signed char);
5920 vector signed short vec_sro (vector signed short, vector unsigned char);
5922 vector unsigned short vec_sro (vector unsigned short,
5923 vector signed char);
5924 vector unsigned short vec_sro (vector unsigned short,
5925 vector unsigned char);
5926 vector signed char vec_sro (vector signed char, vector signed char);
5927 vector signed char vec_sro (vector signed char, vector unsigned char);
5928 vector unsigned char vec_sro (vector unsigned char, vector signed char);
5930 vector unsigned char vec_sro (vector unsigned char,
5931 vector unsigned char);
5933 void vec_st (vector float, int, float *);
5934 void vec_st (vector float, int, vector float *);
5935 void vec_st (vector signed int, int, int *);
5936 void vec_st (vector signed int, int, unsigned int *);
5937 void vec_st (vector unsigned int, int, unsigned int *);
5938 void vec_st (vector unsigned int, int, vector unsigned int *);
5939 void vec_st (vector signed short, int, short *);
5940 void vec_st (vector signed short, int, vector unsigned short *);
5941 void vec_st (vector signed short, int, vector signed short *);
5942 void vec_st (vector unsigned short, int, unsigned short *);
5943 void vec_st (vector unsigned short, int, vector unsigned short *);
5944 void vec_st (vector signed char, int, signed char *);
5945 void vec_st (vector signed char, int, unsigned char *);
5946 void vec_st (vector signed char, int, vector signed char *);
5947 void vec_st (vector unsigned char, int, unsigned char *);
5948 void vec_st (vector unsigned char, int, vector unsigned char *);
5950 void vec_ste (vector signed char, int, unsigned char *);
5951 void vec_ste (vector signed char, int, signed char *);
5952 void vec_ste (vector unsigned char, int, unsigned char *);
5953 void vec_ste (vector signed short, int, short *);
5954 void vec_ste (vector signed short, int, unsigned short *);
5955 void vec_ste (vector unsigned short, int, void *);
5956 void vec_ste (vector signed int, int, unsigned int *);
5957 void vec_ste (vector signed int, int, int *);
5958 void vec_ste (vector unsigned int, int, unsigned int *);
5959 void vec_ste (vector float, int, float *);
5961 void vec_stl (vector float, int, vector float *);
5962 void vec_stl (vector float, int, float *);
5963 void vec_stl (vector signed int, int, vector signed int *);
5964 void vec_stl (vector signed int, int, int *);
5965 void vec_stl (vector signed int, int, unsigned int *);
5966 void vec_stl (vector unsigned int, int, vector unsigned int *);
5967 void vec_stl (vector unsigned int, int, unsigned int *);
5968 void vec_stl (vector signed short, int, short *);
5969 void vec_stl (vector signed short, int, unsigned short *);
5970 void vec_stl (vector signed short, int, vector signed short *);
5971 void vec_stl (vector unsigned short, int, unsigned short *);
5972 void vec_stl (vector unsigned short, int, vector signed short *);
5973 void vec_stl (vector signed char, int, signed char *);
5974 void vec_stl (vector signed char, int, unsigned char *);
5975 void vec_stl (vector signed char, int, vector signed char *);
5976 void vec_stl (vector unsigned char, int, unsigned char *);
5977 void vec_stl (vector unsigned char, int, vector unsigned char *);
5979 vector signed char vec_sub (vector signed char, vector signed char);
5980 vector unsigned char vec_sub (vector signed char, vector unsigned char);
5982 vector unsigned char vec_sub (vector unsigned char, vector signed char);
5984 vector unsigned char vec_sub (vector unsigned char,
5985 vector unsigned char);
5986 vector signed short vec_sub (vector signed short, vector signed short);
5987 vector unsigned short vec_sub (vector signed short,
5988 vector unsigned short);
5989 vector unsigned short vec_sub (vector unsigned short,
5990 vector signed short);
5991 vector unsigned short vec_sub (vector unsigned short,
5992 vector unsigned short);
5993 vector signed int vec_sub (vector signed int, vector signed int);
5994 vector unsigned int vec_sub (vector signed int, vector unsigned int);
5995 vector unsigned int vec_sub (vector unsigned int, vector signed int);
5996 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
5997 vector float vec_sub (vector float, vector float);
5999 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
6001 vector unsigned char vec_subs (vector signed char,
6002 vector unsigned char);
6003 vector unsigned char vec_subs (vector unsigned char,
6004 vector signed char);
6005 vector unsigned char vec_subs (vector unsigned char,
6006 vector unsigned char);
6007 vector signed char vec_subs (vector signed char, vector signed char);
6008 vector unsigned short vec_subs (vector signed short,
6009 vector unsigned short);
6010 vector unsigned short vec_subs (vector unsigned short,
6011 vector signed short);
6012 vector unsigned short vec_subs (vector unsigned short,
6013 vector unsigned short);
6014 vector signed short vec_subs (vector signed short, vector signed short);
6016 vector unsigned int vec_subs (vector signed int, vector unsigned int);
6017 vector unsigned int vec_subs (vector unsigned int, vector signed int);
6018 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
6020 vector signed int vec_subs (vector signed int, vector signed int);
6022 vector unsigned int vec_sum4s (vector unsigned char,
6023 vector unsigned int);
6024 vector signed int vec_sum4s (vector signed char, vector signed int);
6025 vector signed int vec_sum4s (vector signed short, vector signed int);
6027 vector signed int vec_sum2s (vector signed int, vector signed int);
6029 vector signed int vec_sums (vector signed int, vector signed int);
6031 vector float vec_trunc (vector float);
6033 vector signed short vec_unpackh (vector signed char);
6034 vector unsigned int vec_unpackh (vector signed short);
6035 vector signed int vec_unpackh (vector signed short);
6037 vector signed short vec_unpackl (vector signed char);
6038 vector unsigned int vec_unpackl (vector signed short);
6039 vector signed int vec_unpackl (vector signed short);
6041 vector float vec_xor (vector float, vector float);
6042 vector float vec_xor (vector float, vector signed int);
6043 vector float vec_xor (vector signed int, vector float);
6044 vector signed int vec_xor (vector signed int, vector signed int);
6045 vector unsigned int vec_xor (vector signed int, vector unsigned int);
6046 vector unsigned int vec_xor (vector unsigned int, vector signed int);
6047 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
6048 vector signed short vec_xor (vector signed short, vector signed short);
6049 vector unsigned short vec_xor (vector signed short,
6050 vector unsigned short);
6051 vector unsigned short vec_xor (vector unsigned short,
6052 vector signed short);
6053 vector unsigned short vec_xor (vector unsigned short,
6054 vector unsigned short);
6055 vector signed char vec_xor (vector signed char, vector signed char);
6056 vector unsigned char vec_xor (vector signed char, vector unsigned char);
6058 vector unsigned char vec_xor (vector unsigned char, vector signed char);
6060 vector unsigned char vec_xor (vector unsigned char,
6061 vector unsigned char);
6063 vector signed int vec_all_eq (vector signed char, vector unsigned char);
6065 vector signed int vec_all_eq (vector signed char, vector signed char);
6066 vector signed int vec_all_eq (vector unsigned char, vector signed char);
6068 vector signed int vec_all_eq (vector unsigned char,
6069 vector unsigned char);
6070 vector signed int vec_all_eq (vector signed short,
6071 vector unsigned short);
6072 vector signed int vec_all_eq (vector signed short, vector signed short);
6074 vector signed int vec_all_eq (vector unsigned short,
6075 vector signed short);
6076 vector signed int vec_all_eq (vector unsigned short,
6077 vector unsigned short);
6078 vector signed int vec_all_eq (vector signed int, vector unsigned int);
6079 vector signed int vec_all_eq (vector signed int, vector signed int);
6080 vector signed int vec_all_eq (vector unsigned int, vector signed int);
6081 vector signed int vec_all_eq (vector unsigned int, vector unsigned int);
6083 vector signed int vec_all_eq (vector float, vector float);
6085 vector signed int vec_all_ge (vector signed char, vector unsigned char);
6087 vector signed int vec_all_ge (vector unsigned char, vector signed char);
6089 vector signed int vec_all_ge (vector unsigned char,
6090 vector unsigned char);
6091 vector signed int vec_all_ge (vector signed char, vector signed char);
6092 vector signed int vec_all_ge (vector signed short,
6093 vector unsigned short);
6094 vector signed int vec_all_ge (vector unsigned short,
6095 vector signed short);
6096 vector signed int vec_all_ge (vector unsigned short,
6097 vector unsigned short);
6098 vector signed int vec_all_ge (vector signed short, vector signed short);
6100 vector signed int vec_all_ge (vector signed int, vector unsigned int);
6101 vector signed int vec_all_ge (vector unsigned int, vector signed int);
6102 vector signed int vec_all_ge (vector unsigned int, vector unsigned int);
6104 vector signed int vec_all_ge (vector signed int, vector signed int);
6105 vector signed int vec_all_ge (vector float, vector float);
6107 vector signed int vec_all_gt (vector signed char, vector unsigned char);
6109 vector signed int vec_all_gt (vector unsigned char, vector signed char);
6111 vector signed int vec_all_gt (vector unsigned char,
6112 vector unsigned char);
6113 vector signed int vec_all_gt (vector signed char, vector signed char);
6114 vector signed int vec_all_gt (vector signed short,
6115 vector unsigned short);
6116 vector signed int vec_all_gt (vector unsigned short,
6117 vector signed short);
6118 vector signed int vec_all_gt (vector unsigned short,
6119 vector unsigned short);
6120 vector signed int vec_all_gt (vector signed short, vector signed short);
6122 vector signed int vec_all_gt (vector signed int, vector unsigned int);
6123 vector signed int vec_all_gt (vector unsigned int, vector signed int);
6124 vector signed int vec_all_gt (vector unsigned int, vector unsigned int);
6126 vector signed int vec_all_gt (vector signed int, vector signed int);
6127 vector signed int vec_all_gt (vector float, vector float);
6129 vector signed int vec_all_in (vector float, vector float);
6131 vector signed int vec_all_le (vector signed char, vector unsigned char);
6133 vector signed int vec_all_le (vector unsigned char, vector signed char);
6135 vector signed int vec_all_le (vector unsigned char,
6136 vector unsigned char);
6137 vector signed int vec_all_le (vector signed char, vector signed char);
6138 vector signed int vec_all_le (vector signed short,
6139 vector unsigned short);
6140 vector signed int vec_all_le (vector unsigned short,
6141 vector signed short);
6142 vector signed int vec_all_le (vector unsigned short,
6143 vector unsigned short);
6144 vector signed int vec_all_le (vector signed short, vector signed short);
6146 vector signed int vec_all_le (vector signed int, vector unsigned int);
6147 vector signed int vec_all_le (vector unsigned int, vector signed int);
6148 vector signed int vec_all_le (vector unsigned int, vector unsigned int);
6150 vector signed int vec_all_le (vector signed int, vector signed int);
6151 vector signed int vec_all_le (vector float, vector float);
6153 vector signed int vec_all_lt (vector signed char, vector unsigned char);
6155 vector signed int vec_all_lt (vector unsigned char, vector signed char);
6157 vector signed int vec_all_lt (vector unsigned char,
6158 vector unsigned char);
6159 vector signed int vec_all_lt (vector signed char, vector signed char);
6160 vector signed int vec_all_lt (vector signed short,
6161 vector unsigned short);
6162 vector signed int vec_all_lt (vector unsigned short,
6163 vector signed short);
6164 vector signed int vec_all_lt (vector unsigned short,
6165 vector unsigned short);
6166 vector signed int vec_all_lt (vector signed short, vector signed short);
6168 vector signed int vec_all_lt (vector signed int, vector unsigned int);
6169 vector signed int vec_all_lt (vector unsigned int, vector signed int);
6170 vector signed int vec_all_lt (vector unsigned int, vector unsigned int);
6172 vector signed int vec_all_lt (vector signed int, vector signed int);
6173 vector signed int vec_all_lt (vector float, vector float);
6175 vector signed int vec_all_nan (vector float);
6177 vector signed int vec_all_ne (vector signed char, vector unsigned char);
6179 vector signed int vec_all_ne (vector signed char, vector signed char);
6180 vector signed int vec_all_ne (vector unsigned char, vector signed char);
6182 vector signed int vec_all_ne (vector unsigned char,
6183 vector unsigned char);
6184 vector signed int vec_all_ne (vector signed short,
6185 vector unsigned short);
6186 vector signed int vec_all_ne (vector signed short, vector signed short);
6188 vector signed int vec_all_ne (vector unsigned short,
6189 vector signed short);
6190 vector signed int vec_all_ne (vector unsigned short,
6191 vector unsigned short);
6192 vector signed int vec_all_ne (vector signed int, vector unsigned int);
6193 vector signed int vec_all_ne (vector signed int, vector signed int);
6194 vector signed int vec_all_ne (vector unsigned int, vector signed int);
6195 vector signed int vec_all_ne (vector unsigned int, vector unsigned int);
6197 vector signed int vec_all_ne (vector float, vector float);
6199 vector signed int vec_all_nge (vector float, vector float);
6201 vector signed int vec_all_ngt (vector float, vector float);
6203 vector signed int vec_all_nle (vector float, vector float);
6205 vector signed int vec_all_nlt (vector float, vector float);
6207 vector signed int vec_all_numeric (vector float);
6209 vector signed int vec_any_eq (vector signed char, vector unsigned char);
6211 vector signed int vec_any_eq (vector signed char, vector signed char);
6212 vector signed int vec_any_eq (vector unsigned char, vector signed char);
6214 vector signed int vec_any_eq (vector unsigned char,
6215 vector unsigned char);
6216 vector signed int vec_any_eq (vector signed short,
6217 vector unsigned short);
6218 vector signed int vec_any_eq (vector signed short, vector signed short);
6220 vector signed int vec_any_eq (vector unsigned short,
6221 vector signed short);
6222 vector signed int vec_any_eq (vector unsigned short,
6223 vector unsigned short);
6224 vector signed int vec_any_eq (vector signed int, vector unsigned int);
6225 vector signed int vec_any_eq (vector signed int, vector signed int);
6226 vector signed int vec_any_eq (vector unsigned int, vector signed int);
6227 vector signed int vec_any_eq (vector unsigned int, vector unsigned int);
6229 vector signed int vec_any_eq (vector float, vector float);
6231 vector signed int vec_any_ge (vector signed char, vector unsigned char);
6233 vector signed int vec_any_ge (vector unsigned char, vector signed char);
6235 vector signed int vec_any_ge (vector unsigned char,
6236 vector unsigned char);
6237 vector signed int vec_any_ge (vector signed char, vector signed char);
6238 vector signed int vec_any_ge (vector signed short,
6239 vector unsigned short);
6240 vector signed int vec_any_ge (vector unsigned short,
6241 vector signed short);
6242 vector signed int vec_any_ge (vector unsigned short,
6243 vector unsigned short);
6244 vector signed int vec_any_ge (vector signed short, vector signed short);
6246 vector signed int vec_any_ge (vector signed int, vector unsigned int);
6247 vector signed int vec_any_ge (vector unsigned int, vector signed int);
6248 vector signed int vec_any_ge (vector unsigned int, vector unsigned int);
6250 vector signed int vec_any_ge (vector signed int, vector signed int);
6251 vector signed int vec_any_ge (vector float, vector float);
6253 vector signed int vec_any_gt (vector signed char, vector unsigned char);
6255 vector signed int vec_any_gt (vector unsigned char, vector signed char);
6257 vector signed int vec_any_gt (vector unsigned char,
6258 vector unsigned char);
6259 vector signed int vec_any_gt (vector signed char, vector signed char);
6260 vector signed int vec_any_gt (vector signed short,
6261 vector unsigned short);
6262 vector signed int vec_any_gt (vector unsigned short,
6263 vector signed short);
6264 vector signed int vec_any_gt (vector unsigned short,
6265 vector unsigned short);
6266 vector signed int vec_any_gt (vector signed short, vector signed short);
6268 vector signed int vec_any_gt (vector signed int, vector unsigned int);
6269 vector signed int vec_any_gt (vector unsigned int, vector signed int);
6270 vector signed int vec_any_gt (vector unsigned int, vector unsigned int);
6272 vector signed int vec_any_gt (vector signed int, vector signed int);
6273 vector signed int vec_any_gt (vector float, vector float);
6275 vector signed int vec_any_le (vector signed char, vector unsigned char);
6277 vector signed int vec_any_le (vector unsigned char, vector signed char);
6279 vector signed int vec_any_le (vector unsigned char,
6280 vector unsigned char);
6281 vector signed int vec_any_le (vector signed char, vector signed char);
6282 vector signed int vec_any_le (vector signed short,
6283 vector unsigned short);
6284 vector signed int vec_any_le (vector unsigned short,
6285 vector signed short);
6286 vector signed int vec_any_le (vector unsigned short,
6287 vector unsigned short);
6288 vector signed int vec_any_le (vector signed short, vector signed short);
6290 vector signed int vec_any_le (vector signed int, vector unsigned int);
6291 vector signed int vec_any_le (vector unsigned int, vector signed int);
6292 vector signed int vec_any_le (vector unsigned int, vector unsigned int);
6294 vector signed int vec_any_le (vector signed int, vector signed int);
6295 vector signed int vec_any_le (vector float, vector float);
6297 vector signed int vec_any_lt (vector signed char, vector unsigned char);
6299 vector signed int vec_any_lt (vector unsigned char, vector signed char);
6301 vector signed int vec_any_lt (vector unsigned char,
6302 vector unsigned char);
6303 vector signed int vec_any_lt (vector signed char, vector signed char);
6304 vector signed int vec_any_lt (vector signed short,
6305 vector unsigned short);
6306 vector signed int vec_any_lt (vector unsigned short,
6307 vector signed short);
6308 vector signed int vec_any_lt (vector unsigned short,
6309 vector unsigned short);
6310 vector signed int vec_any_lt (vector signed short, vector signed short);
6312 vector signed int vec_any_lt (vector signed int, vector unsigned int);
6313 vector signed int vec_any_lt (vector unsigned int, vector signed int);
6314 vector signed int vec_any_lt (vector unsigned int, vector unsigned int);
6316 vector signed int vec_any_lt (vector signed int, vector signed int);
6317 vector signed int vec_any_lt (vector float, vector float);
6319 vector signed int vec_any_nan (vector float);
6321 vector signed int vec_any_ne (vector signed char, vector unsigned char);
6323 vector signed int vec_any_ne (vector signed char, vector signed char);
6324 vector signed int vec_any_ne (vector unsigned char, vector signed char);
6326 vector signed int vec_any_ne (vector unsigned char,
6327 vector unsigned char);
6328 vector signed int vec_any_ne (vector signed short,
6329 vector unsigned short);
6330 vector signed int vec_any_ne (vector signed short, vector signed short);
6332 vector signed int vec_any_ne (vector unsigned short,
6333 vector signed short);
6334 vector signed int vec_any_ne (vector unsigned short,
6335 vector unsigned short);
6336 vector signed int vec_any_ne (vector signed int, vector unsigned int);
6337 vector signed int vec_any_ne (vector signed int, vector signed int);
6338 vector signed int vec_any_ne (vector unsigned int, vector signed int);
6339 vector signed int vec_any_ne (vector unsigned int, vector unsigned int);
6341 vector signed int vec_any_ne (vector float, vector float);
6343 vector signed int vec_any_nge (vector float, vector float);
6345 vector signed int vec_any_ngt (vector float, vector float);
6347 vector signed int vec_any_nle (vector float, vector float);
6349 vector signed int vec_any_nlt (vector float, vector float);
6351 vector signed int vec_any_numeric (vector float);
6353 vector signed int vec_any_out (vector float, vector float);
6357 @section Pragmas Accepted by GCC
6361 GCC supports several types of pragmas, primarily in order to compile
6362 code originally written for other compilers. Note that in general
6363 we do not recommend the use of pragmas; @xref{Function Attributes},
6364 for further explanation.
6368 * RS/6000 and PowerPC Pragmas::
6375 @subsection ARM Pragmas
6377 The ARM target defines pragmas for controlling the default addition of
6378 @code{long_call} and @code{short_call} attributes to functions.
6379 @xref{Function Attributes}, for information about the effects of these
6384 @cindex pragma, long_calls
6385 Set all subsequent functions to have the @code{long_call} attribute.
6388 @cindex pragma, no_long_calls
6389 Set all subsequent functions to have the @code{short_call} attribute.
6391 @item long_calls_off
6392 @cindex pragma, long_calls_off
6393 Do not affect the @code{long_call} or @code{short_call} attributes of
6394 subsequent functions.
6397 @node RS/6000 and PowerPC Pragmas
6398 @subsection RS/6000 and PowerPC Pragmas
6400 The RS/6000 and PowerPC targets define one pragma for controlling
6401 whether or not the @code{longcall} attribute is added to function
6402 declarations by default. This pragma overrides the @option{-mlongcall}
6403 option, but not the @code{longcall} and @code{shortcall} attributes.
6404 @xref{RS/6000 and PowerPC Options}, for more information about when long
6405 calls are and are not necessary.
6409 @cindex pragma, longcall
6410 Apply the @code{longcall} attribute to all subsequent function
6414 Do not apply the @code{longcall} attribute to subsequent function
6418 @c Describe c4x pragmas here.
6419 @c Describe h8300 pragmas here.
6420 @c Describe i370 pragmas here.
6421 @c Describe i960 pragmas here.
6422 @c Describe sh pragmas here.
6423 @c Describe v850 pragmas here.
6425 @node Darwin Pragmas
6426 @subsection Darwin Pragmas
6428 The following pragmas are available for all architectures running the
6429 Darwin operating system. These are useful for compatibility with other
6433 @item mark @var{tokens}@dots{}
6434 @cindex pragma, mark
6435 This pragma is accepted, but has no effect.
6437 @item options align=@var{alignment}
6438 @cindex pragma, options align
6439 This pragma sets the alignment of fields in structures. The values of
6440 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
6441 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
6442 properly; to restore the previous setting, use @code{reset} for the
6445 @item segment @var{tokens}@dots{}
6446 @cindex pragma, segment
6447 This pragma is accepted, but has no effect.
6449 @item unused (@var{var} [, @var{var}]@dots{})
6450 @cindex pragma, unused
6451 This pragma declares variables to be possibly unused. GCC will not
6452 produce warnings for the listed variables. The effect is similar to
6453 that of the @code{unused} attribute, except that this pragma may appear
6454 anywhere within the variables' scopes.
6457 @node Solaris Pragmas
6458 @subsection Solaris Pragmas
6460 For compatibility with the SunPRO compiler, the following pragma
6464 @item redefine_extname @var{oldname} @var{newname}
6465 @cindex pragma, redefine_extname
6467 This pragma gives the C function @var{oldname} the assembler label
6468 @var{newname}. The pragma must appear before the function declaration.
6469 This pragma is equivalent to the asm labels extension (@pxref{Asm
6470 Labels}). The preprocessor defines @code{__PRAGMA_REDEFINE_EXTNAME}
6471 if the pragma is available.
6475 @subsection Tru64 Pragmas
6477 For compatibility with the Compaq C compiler, the following pragma
6481 @item extern_prefix @var{string}
6482 @cindex pragma, extern_prefix
6484 This pragma renames all subsequent function and variable declarations
6485 such that @var{string} is prepended to the name. This effect may be
6486 terminated by using another @code{extern_prefix} pragma with the
6489 This pragma is similar in intent to to the asm labels extension
6490 (@pxref{Asm Labels}) in that the system programmer wants to change
6491 the assembly-level ABI without changing the source-level API. The
6492 preprocessor defines @code{__EXTERN_PREFIX} if the pragma is available.
6495 @node Unnamed Fields
6496 @section Unnamed struct/union fields within structs/unions.
6500 For compatibility with other compilers, GCC allows you to define
6501 a structure or union that contains, as fields, structures and unions
6502 without names. For example:
6515 In this example, the user would be able to access members of the unnamed
6516 union with code like @samp{foo.b}. Note that only unnamed structs and
6517 unions are allowed, you may not have, for example, an unnamed
6520 You must never create such structures that cause ambiguous field definitions.
6521 For example, this structure:
6532 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
6533 Such constructs are not supported and must be avoided. In the future,
6534 such constructs may be detected and treated as compilation errors.
6537 @section Thread-Local Storage
6538 @cindex Thread-Local Storage
6539 @cindex @acronym{TLS}
6542 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
6543 are allocated such that there is one instance of the variable per extant
6544 thread. The run-time model GCC uses to implement this originates
6545 in the IA-64 processor-specific ABI, but has since been migrated
6546 to other processors as well. It requires significant support from
6547 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
6548 system libraries (@file{libc.so} and @file{libpthread.so}), so it
6549 is not available everywhere.
6551 At the user level, the extension is visible with a new storage
6552 class keyword: @code{__thread}. For example:
6556 extern __thread struct state s;
6557 static __thread char *p;
6560 The @code{__thread} specifier may be used alone, with the @code{extern}
6561 or @code{static} specifiers, but with no other storage class specifier.
6562 When used with @code{extern} or @code{static}, @code{__thread} must appear
6563 immediately after the other storage class specifier.
6565 The @code{__thread} specifier may be applied to any global, file-scoped
6566 static, function-scoped static, or static data member of a class. It may
6567 not be applied to block-scoped automatic or non-static data member.
6569 When the address-of operator is applied to a thread-local variable, it is
6570 evaluated at run-time and returns the address of the current thread's
6571 instance of that variable. An address so obtained may be used by any
6572 thread. When a thread terminates, any pointers to thread-local variables
6573 in that thread become invalid.
6575 No static initialization may refer to the address of a thread-local variable.
6577 In C++, if an initializer is present for a thread-local variable, it must
6578 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
6581 See @uref{http://people.redhat.com/drepper/tls.pdf,
6582 ELF Handling For Thread-Local Storage} for a detailed explanation of
6583 the four thread-local storage addressing models, and how the run-time
6584 is expected to function.
6587 * C99 Thread-Local Edits::
6588 * C++98 Thread-Local Edits::
6591 @node C99 Thread-Local Edits
6592 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
6594 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
6595 that document the exact semantics of the language extension.
6599 @cite{5.1.2 Execution environments}
6601 Add new text after paragraph 1
6604 Within either execution environment, a @dfn{thread} is a flow of
6605 control within a program. It is implementation defined whether
6606 or not there may be more than one thread associated with a program.
6607 It is implementation defined how threads beyond the first are
6608 created, the name and type of the function called at thread
6609 startup, and how threads may be terminated. However, objects
6610 with thread storage duration shall be initialized before thread
6615 @cite{6.2.4 Storage durations of objects}
6617 Add new text before paragraph 3
6620 An object whose identifier is declared with the storage-class
6621 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
6622 Its lifetime is the entire execution of the thread, and its
6623 stored value is initialized only once, prior to thread startup.
6627 @cite{6.4.1 Keywords}
6629 Add @code{__thread}.
6632 @cite{6.7.1 Storage-class specifiers}
6634 Add @code{__thread} to the list of storage class specifiers in
6637 Change paragraph 2 to
6640 With the exception of @code{__thread}, at most one storage-class
6641 specifier may be given [@dots{}]. The @code{__thread} specifier may
6642 be used alone, or immediately following @code{extern} or
6646 Add new text after paragraph 6
6649 The declaration of an identifier for a variable that has
6650 block scope that specifies @code{__thread} shall also
6651 specify either @code{extern} or @code{static}.
6653 The @code{__thread} specifier shall be used only with
6658 @node C++98 Thread-Local Edits
6659 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
6661 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
6662 that document the exact semantics of the language extension.
6665 @b{[intro.execution]}
6667 New text after paragraph 4
6670 A @dfn{thread} is a flow of control within the abstract machine.
6671 It is implementation defined whether or not there may be more than
6675 New text after paragraph 7
6678 It is unspecified whether additional action must be taken to
6679 ensure when and whether side effects are visible to other threads.
6685 Add @code{__thread}.
6688 @b{[basic.start.main]}
6690 Add after paragraph 5
6693 The thread that begins execution at the @code{main} function is called
6694 the @dfn{main thread}. It is implementation defined how functions
6695 beginning threads other than the main thread are designated or typed.
6696 A function so designated, as well as the @code{main} function, is called
6697 a @dfn{thread startup function}. It is implementation defined what
6698 happens if a thread startup function returns. It is implementation
6699 defined what happens to other threads when any thread calls @code{exit}.
6703 @b{[basic.start.init]}
6705 Add after paragraph 4
6708 The storage for an object of thread storage duration shall be
6709 statically initialized before the first statement of the thread startup
6710 function. An object of thread storage duration shall not require
6711 dynamic initialization.
6715 @b{[basic.start.term]}
6717 Add after paragraph 3
6720 The type of an object with thread storage duration shall not have a
6721 non-trivial destructor, nor shall it be an array type whose elements
6722 (directly or indirectly) have non-trivial destructors.
6728 Add ``thread storage duration'' to the list in paragraph 1.
6733 Thread, static, and automatic storage durations are associated with
6734 objects introduced by declarations [@dots{}].
6737 Add @code{__thread} to the list of specifiers in paragraph 3.
6740 @b{[basic.stc.thread]}
6742 New section before @b{[basic.stc.static]}
6745 The keyword @code{__thread} applied to an non-local object gives the
6746 object thread storage duration.
6748 A local variable or class data member declared both @code{static}
6749 and @code{__thread} gives the variable or member thread storage
6754 @b{[basic.stc.static]}
6759 All objects which have neither thread storage duration, dynamic
6760 storage duration nor are local [@dots{}].
6766 Add @code{__thread} to the list in paragraph 1.
6771 With the exception of @code{__thread}, at most one
6772 @var{storage-class-specifier} shall appear in a given
6773 @var{decl-specifier-seq}. The @code{__thread} specifier may
6774 be used alone, or immediately following the @code{extern} or
6775 @code{static} specifiers. [@dots{}]
6778 Add after paragraph 5
6781 The @code{__thread} specifier can be applied only to the names of objects
6782 and to anonymous unions.
6788 Add after paragraph 6
6791 Non-@code{static} members shall not be @code{__thread}.
6795 @node C++ Extensions
6796 @chapter Extensions to the C++ Language
6797 @cindex extensions, C++ language
6798 @cindex C++ language extensions
6800 The GNU compiler provides these extensions to the C++ language (and you
6801 can also use most of the C language extensions in your C++ programs). If you
6802 want to write code that checks whether these features are available, you can
6803 test for the GNU compiler the same way as for C programs: check for a
6804 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
6805 test specifically for GNU C++ (@pxref{Standard Predefined,,Standard
6806 Predefined Macros,cpp.info,The C Preprocessor}).
6809 * Min and Max:: C++ Minimum and maximum operators.
6810 * Volatiles:: What constitutes an access to a volatile object.
6811 * Restricted Pointers:: C99 restricted pointers and references.
6812 * Vague Linkage:: Where G++ puts inlines, vtables and such.
6813 * C++ Interface:: You can use a single C++ header file for both
6814 declarations and definitions.
6815 * Template Instantiation:: Methods for ensuring that exactly one copy of
6816 each needed template instantiation is emitted.
6817 * Bound member functions:: You can extract a function pointer to the
6818 method denoted by a @samp{->*} or @samp{.*} expression.
6819 * C++ Attributes:: Variable, function, and type attributes for C++ only.
6820 * Java Exceptions:: Tweaking exception handling to work with Java.
6821 * Deprecated Features:: Things might disappear from g++.
6822 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
6826 @section Minimum and Maximum Operators in C++
6828 It is very convenient to have operators which return the ``minimum'' or the
6829 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
6832 @item @var{a} <? @var{b}
6834 @cindex minimum operator
6835 is the @dfn{minimum}, returning the smaller of the numeric values
6836 @var{a} and @var{b};
6838 @item @var{a} >? @var{b}
6840 @cindex maximum operator
6841 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
6845 These operations are not primitive in ordinary C++, since you can
6846 use a macro to return the minimum of two things in C++, as in the
6850 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
6854 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
6855 the minimum value of variables @var{i} and @var{j}.
6857 However, side effects in @code{X} or @code{Y} may cause unintended
6858 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
6859 the smaller counter twice. The GNU C @code{typeof} extension allows you
6860 to write safe macros that avoid this kind of problem (@pxref{Typeof}).
6861 However, writing @code{MIN} and @code{MAX} as macros also forces you to
6862 use function-call notation for a fundamental arithmetic operation.
6863 Using GNU C++ extensions, you can write @w{@samp{int min = i <? j;}}
6866 Since @code{<?} and @code{>?} are built into the compiler, they properly
6867 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
6871 @section When is a Volatile Object Accessed?
6872 @cindex accessing volatiles
6873 @cindex volatile read
6874 @cindex volatile write
6875 @cindex volatile access
6877 Both the C and C++ standard have the concept of volatile objects. These
6878 are normally accessed by pointers and used for accessing hardware. The
6879 standards encourage compilers to refrain from optimizations
6880 concerning accesses to volatile objects that it might perform on
6881 non-volatile objects. The C standard leaves it implementation defined
6882 as to what constitutes a volatile access. The C++ standard omits to
6883 specify this, except to say that C++ should behave in a similar manner
6884 to C with respect to volatiles, where possible. The minimum either
6885 standard specifies is that at a sequence point all previous accesses to
6886 volatile objects have stabilized and no subsequent accesses have
6887 occurred. Thus an implementation is free to reorder and combine
6888 volatile accesses which occur between sequence points, but cannot do so
6889 for accesses across a sequence point. The use of volatiles does not
6890 allow you to violate the restriction on updating objects multiple times
6891 within a sequence point.
6893 In most expressions, it is intuitively obvious what is a read and what is
6894 a write. For instance
6897 volatile int *dst = @var{somevalue};
6898 volatile int *src = @var{someothervalue};
6903 will cause a read of the volatile object pointed to by @var{src} and stores the
6904 value into the volatile object pointed to by @var{dst}. There is no
6905 guarantee that these reads and writes are atomic, especially for objects
6906 larger than @code{int}.
6908 Less obvious expressions are where something which looks like an access
6909 is used in a void context. An example would be,
6912 volatile int *src = @var{somevalue};
6916 With C, such expressions are rvalues, and as rvalues cause a read of
6917 the object, GCC interprets this as a read of the volatile being pointed
6918 to. The C++ standard specifies that such expressions do not undergo
6919 lvalue to rvalue conversion, and that the type of the dereferenced
6920 object may be incomplete. The C++ standard does not specify explicitly
6921 that it is this lvalue to rvalue conversion which is responsible for
6922 causing an access. However, there is reason to believe that it is,
6923 because otherwise certain simple expressions become undefined. However,
6924 because it would surprise most programmers, G++ treats dereferencing a
6925 pointer to volatile object of complete type in a void context as a read
6926 of the object. When the object has incomplete type, G++ issues a
6931 struct T @{int m;@};
6932 volatile S *ptr1 = @var{somevalue};
6933 volatile T *ptr2 = @var{somevalue};
6938 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
6939 causes a read of the object pointed to. If you wish to force an error on
6940 the first case, you must force a conversion to rvalue with, for instance
6941 a static cast, @code{static_cast<S>(*ptr1)}.
6943 When using a reference to volatile, G++ does not treat equivalent
6944 expressions as accesses to volatiles, but instead issues a warning that
6945 no volatile is accessed. The rationale for this is that otherwise it
6946 becomes difficult to determine where volatile access occur, and not
6947 possible to ignore the return value from functions returning volatile
6948 references. Again, if you wish to force a read, cast the reference to
6951 @node Restricted Pointers
6952 @section Restricting Pointer Aliasing
6953 @cindex restricted pointers
6954 @cindex restricted references
6955 @cindex restricted this pointer
6957 As with gcc, g++ understands the C99 feature of restricted pointers,
6958 specified with the @code{__restrict__}, or @code{__restrict} type
6959 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
6960 language flag, @code{restrict} is not a keyword in C++.
6962 In addition to allowing restricted pointers, you can specify restricted
6963 references, which indicate that the reference is not aliased in the local
6967 void fn (int *__restrict__ rptr, int &__restrict__ rref)
6974 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
6975 @var{rref} refers to a (different) unaliased integer.
6977 You may also specify whether a member function's @var{this} pointer is
6978 unaliased by using @code{__restrict__} as a member function qualifier.
6981 void T::fn () __restrict__
6988 Within the body of @code{T::fn}, @var{this} will have the effective
6989 definition @code{T *__restrict__ const this}. Notice that the
6990 interpretation of a @code{__restrict__} member function qualifier is
6991 different to that of @code{const} or @code{volatile} qualifier, in that it
6992 is applied to the pointer rather than the object. This is consistent with
6993 other compilers which implement restricted pointers.
6995 As with all outermost parameter qualifiers, @code{__restrict__} is
6996 ignored in function definition matching. This means you only need to
6997 specify @code{__restrict__} in a function definition, rather than
6998 in a function prototype as well.
7001 @section Vague Linkage
7002 @cindex vague linkage
7004 There are several constructs in C++ which require space in the object
7005 file but are not clearly tied to a single translation unit. We say that
7006 these constructs have ``vague linkage''. Typically such constructs are
7007 emitted wherever they are needed, though sometimes we can be more
7011 @item Inline Functions
7012 Inline functions are typically defined in a header file which can be
7013 included in many different compilations. Hopefully they can usually be
7014 inlined, but sometimes an out-of-line copy is necessary, if the address
7015 of the function is taken or if inlining fails. In general, we emit an
7016 out-of-line copy in all translation units where one is needed. As an
7017 exception, we only emit inline virtual functions with the vtable, since
7018 it will always require a copy.
7020 Local static variables and string constants used in an inline function
7021 are also considered to have vague linkage, since they must be shared
7022 between all inlined and out-of-line instances of the function.
7026 C++ virtual functions are implemented in most compilers using a lookup
7027 table, known as a vtable. The vtable contains pointers to the virtual
7028 functions provided by a class, and each object of the class contains a
7029 pointer to its vtable (or vtables, in some multiple-inheritance
7030 situations). If the class declares any non-inline, non-pure virtual
7031 functions, the first one is chosen as the ``key method'' for the class,
7032 and the vtable is only emitted in the translation unit where the key
7035 @emph{Note:} If the chosen key method is later defined as inline, the
7036 vtable will still be emitted in every translation unit which defines it.
7037 Make sure that any inline virtuals are declared inline in the class
7038 body, even if they are not defined there.
7040 @item type_info objects
7043 C++ requires information about types to be written out in order to
7044 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
7045 For polymorphic classes (classes with virtual functions), the type_info
7046 object is written out along with the vtable so that @samp{dynamic_cast}
7047 can determine the dynamic type of a class object at runtime. For all
7048 other types, we write out the type_info object when it is used: when
7049 applying @samp{typeid} to an expression, throwing an object, or
7050 referring to a type in a catch clause or exception specification.
7052 @item Template Instantiations
7053 Most everything in this section also applies to template instantiations,
7054 but there are other options as well.
7055 @xref{Template Instantiation,,Where's the Template?}.
7059 When used with GNU ld version 2.8 or later on an ELF system such as
7060 Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of
7061 these constructs will be discarded at link time. This is known as
7064 On targets that don't support COMDAT, but do support weak symbols, GCC
7065 will use them. This way one copy will override all the others, but
7066 the unused copies will still take up space in the executable.
7068 For targets which do not support either COMDAT or weak symbols,
7069 most entities with vague linkage will be emitted as local symbols to
7070 avoid duplicate definition errors from the linker. This will not happen
7071 for local statics in inlines, however, as having multiple copies will
7072 almost certainly break things.
7074 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
7075 another way to control placement of these constructs.
7078 @section Declarations and Definitions in One Header
7080 @cindex interface and implementation headers, C++
7081 @cindex C++ interface and implementation headers
7082 C++ object definitions can be quite complex. In principle, your source
7083 code will need two kinds of things for each object that you use across
7084 more than one source file. First, you need an @dfn{interface}
7085 specification, describing its structure with type declarations and
7086 function prototypes. Second, you need the @dfn{implementation} itself.
7087 It can be tedious to maintain a separate interface description in a
7088 header file, in parallel to the actual implementation. It is also
7089 dangerous, since separate interface and implementation definitions may
7090 not remain parallel.
7092 @cindex pragmas, interface and implementation
7093 With GNU C++, you can use a single header file for both purposes.
7096 @emph{Warning:} The mechanism to specify this is in transition. For the
7097 nonce, you must use one of two @code{#pragma} commands; in a future
7098 release of GNU C++, an alternative mechanism will make these
7099 @code{#pragma} commands unnecessary.
7102 The header file contains the full definitions, but is marked with
7103 @samp{#pragma interface} in the source code. This allows the compiler
7104 to use the header file only as an interface specification when ordinary
7105 source files incorporate it with @code{#include}. In the single source
7106 file where the full implementation belongs, you can use either a naming
7107 convention or @samp{#pragma implementation} to indicate this alternate
7108 use of the header file.
7111 @item #pragma interface
7112 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
7113 @kindex #pragma interface
7114 Use this directive in @emph{header files} that define object classes, to save
7115 space in most of the object files that use those classes. Normally,
7116 local copies of certain information (backup copies of inline member
7117 functions, debugging information, and the internal tables that implement
7118 virtual functions) must be kept in each object file that includes class
7119 definitions. You can use this pragma to avoid such duplication. When a
7120 header file containing @samp{#pragma interface} is included in a
7121 compilation, this auxiliary information will not be generated (unless
7122 the main input source file itself uses @samp{#pragma implementation}).
7123 Instead, the object files will contain references to be resolved at link
7126 The second form of this directive is useful for the case where you have
7127 multiple headers with the same name in different directories. If you
7128 use this form, you must specify the same string to @samp{#pragma
7131 @item #pragma implementation
7132 @itemx #pragma implementation "@var{objects}.h"
7133 @kindex #pragma implementation
7134 Use this pragma in a @emph{main input file}, when you want full output from
7135 included header files to be generated (and made globally visible). The
7136 included header file, in turn, should use @samp{#pragma interface}.
7137 Backup copies of inline member functions, debugging information, and the
7138 internal tables used to implement virtual functions are all generated in
7139 implementation files.
7141 @cindex implied @code{#pragma implementation}
7142 @cindex @code{#pragma implementation}, implied
7143 @cindex naming convention, implementation headers
7144 If you use @samp{#pragma implementation} with no argument, it applies to
7145 an include file with the same basename@footnote{A file's @dfn{basename}
7146 was the name stripped of all leading path information and of trailing
7147 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
7148 file. For example, in @file{allclass.cc}, giving just
7149 @samp{#pragma implementation}
7150 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
7152 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
7153 an implementation file whenever you would include it from
7154 @file{allclass.cc} even if you never specified @samp{#pragma
7155 implementation}. This was deemed to be more trouble than it was worth,
7156 however, and disabled.
7158 If you use an explicit @samp{#pragma implementation}, it must appear in
7159 your source file @emph{before} you include the affected header files.
7161 Use the string argument if you want a single implementation file to
7162 include code from multiple header files. (You must also use
7163 @samp{#include} to include the header file; @samp{#pragma
7164 implementation} only specifies how to use the file---it doesn't actually
7167 There is no way to split up the contents of a single header file into
7168 multiple implementation files.
7171 @cindex inlining and C++ pragmas
7172 @cindex C++ pragmas, effect on inlining
7173 @cindex pragmas in C++, effect on inlining
7174 @samp{#pragma implementation} and @samp{#pragma interface} also have an
7175 effect on function inlining.
7177 If you define a class in a header file marked with @samp{#pragma
7178 interface}, the effect on a function defined in that class is similar to
7179 an explicit @code{extern} declaration---the compiler emits no code at
7180 all to define an independent version of the function. Its definition
7181 is used only for inlining with its callers.
7183 @opindex fno-implement-inlines
7184 Conversely, when you include the same header file in a main source file
7185 that declares it as @samp{#pragma implementation}, the compiler emits
7186 code for the function itself; this defines a version of the function
7187 that can be found via pointers (or by callers compiled without
7188 inlining). If all calls to the function can be inlined, you can avoid
7189 emitting the function by compiling with @option{-fno-implement-inlines}.
7190 If any calls were not inlined, you will get linker errors.
7192 @node Template Instantiation
7193 @section Where's the Template?
7195 @cindex template instantiation
7197 C++ templates are the first language feature to require more
7198 intelligence from the environment than one usually finds on a UNIX
7199 system. Somehow the compiler and linker have to make sure that each
7200 template instance occurs exactly once in the executable if it is needed,
7201 and not at all otherwise. There are two basic approaches to this
7202 problem, which I will refer to as the Borland model and the Cfront model.
7206 Borland C++ solved the template instantiation problem by adding the code
7207 equivalent of common blocks to their linker; the compiler emits template
7208 instances in each translation unit that uses them, and the linker
7209 collapses them together. The advantage of this model is that the linker
7210 only has to consider the object files themselves; there is no external
7211 complexity to worry about. This disadvantage is that compilation time
7212 is increased because the template code is being compiled repeatedly.
7213 Code written for this model tends to include definitions of all
7214 templates in the header file, since they must be seen to be
7218 The AT&T C++ translator, Cfront, solved the template instantiation
7219 problem by creating the notion of a template repository, an
7220 automatically maintained place where template instances are stored. A
7221 more modern version of the repository works as follows: As individual
7222 object files are built, the compiler places any template definitions and
7223 instantiations encountered in the repository. At link time, the link
7224 wrapper adds in the objects in the repository and compiles any needed
7225 instances that were not previously emitted. The advantages of this
7226 model are more optimal compilation speed and the ability to use the
7227 system linker; to implement the Borland model a compiler vendor also
7228 needs to replace the linker. The disadvantages are vastly increased
7229 complexity, and thus potential for error; for some code this can be
7230 just as transparent, but in practice it can been very difficult to build
7231 multiple programs in one directory and one program in multiple
7232 directories. Code written for this model tends to separate definitions
7233 of non-inline member templates into a separate file, which should be
7234 compiled separately.
7237 When used with GNU ld version 2.8 or later on an ELF system such as
7238 Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the
7239 Borland model. On other systems, g++ implements neither automatic
7242 A future version of g++ will support a hybrid model whereby the compiler
7243 will emit any instantiations for which the template definition is
7244 included in the compile, and store template definitions and
7245 instantiation context information into the object file for the rest.
7246 The link wrapper will extract that information as necessary and invoke
7247 the compiler to produce the remaining instantiations. The linker will
7248 then combine duplicate instantiations.
7250 In the mean time, you have the following options for dealing with
7251 template instantiations:
7256 Compile your template-using code with @option{-frepo}. The compiler will
7257 generate files with the extension @samp{.rpo} listing all of the
7258 template instantiations used in the corresponding object files which
7259 could be instantiated there; the link wrapper, @samp{collect2}, will
7260 then update the @samp{.rpo} files to tell the compiler where to place
7261 those instantiations and rebuild any affected object files. The
7262 link-time overhead is negligible after the first pass, as the compiler
7263 will continue to place the instantiations in the same files.
7265 This is your best option for application code written for the Borland
7266 model, as it will just work. Code written for the Cfront model will
7267 need to be modified so that the template definitions are available at
7268 one or more points of instantiation; usually this is as simple as adding
7269 @code{#include <tmethods.cc>} to the end of each template header.
7271 For library code, if you want the library to provide all of the template
7272 instantiations it needs, just try to link all of its object files
7273 together; the link will fail, but cause the instantiations to be
7274 generated as a side effect. Be warned, however, that this may cause
7275 conflicts if multiple libraries try to provide the same instantiations.
7276 For greater control, use explicit instantiation as described in the next
7280 @opindex fno-implicit-templates
7281 Compile your code with @option{-fno-implicit-templates} to disable the
7282 implicit generation of template instances, and explicitly instantiate
7283 all the ones you use. This approach requires more knowledge of exactly
7284 which instances you need than do the others, but it's less
7285 mysterious and allows greater control. You can scatter the explicit
7286 instantiations throughout your program, perhaps putting them in the
7287 translation units where the instances are used or the translation units
7288 that define the templates themselves; you can put all of the explicit
7289 instantiations you need into one big file; or you can create small files
7296 template class Foo<int>;
7297 template ostream& operator <<
7298 (ostream&, const Foo<int>&);
7301 for each of the instances you need, and create a template instantiation
7304 If you are using Cfront-model code, you can probably get away with not
7305 using @option{-fno-implicit-templates} when compiling files that don't
7306 @samp{#include} the member template definitions.
7308 If you use one big file to do the instantiations, you may want to
7309 compile it without @option{-fno-implicit-templates} so you get all of the
7310 instances required by your explicit instantiations (but not by any
7311 other files) without having to specify them as well.
7313 g++ has extended the template instantiation syntax outlined in the
7314 Working Paper to allow forward declaration of explicit instantiations
7315 (with @code{extern}), instantiation of the compiler support data for a
7316 template class (i.e.@: the vtable) without instantiating any of its
7317 members (with @code{inline}), and instantiation of only the static data
7318 members of a template class, without the support data or member
7319 functions (with (@code{static}):
7322 extern template int max (int, int);
7323 inline template class Foo<int>;
7324 static template class Foo<int>;
7328 Do nothing. Pretend g++ does implement automatic instantiation
7329 management. Code written for the Borland model will work fine, but
7330 each translation unit will contain instances of each of the templates it
7331 uses. In a large program, this can lead to an unacceptable amount of code
7334 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
7335 more discussion of these pragmas.
7338 @node Bound member functions
7339 @section Extracting the function pointer from a bound pointer to member function
7342 @cindex pointer to member function
7343 @cindex bound pointer to member function
7345 In C++, pointer to member functions (PMFs) are implemented using a wide
7346 pointer of sorts to handle all the possible call mechanisms; the PMF
7347 needs to store information about how to adjust the @samp{this} pointer,
7348 and if the function pointed to is virtual, where to find the vtable, and
7349 where in the vtable to look for the member function. If you are using
7350 PMFs in an inner loop, you should really reconsider that decision. If
7351 that is not an option, you can extract the pointer to the function that
7352 would be called for a given object/PMF pair and call it directly inside
7353 the inner loop, to save a bit of time.
7355 Note that you will still be paying the penalty for the call through a
7356 function pointer; on most modern architectures, such a call defeats the
7357 branch prediction features of the CPU@. This is also true of normal
7358 virtual function calls.
7360 The syntax for this extension is
7364 extern int (A::*fp)();
7365 typedef int (*fptr)(A *);
7367 fptr p = (fptr)(a.*fp);
7370 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
7371 no object is needed to obtain the address of the function. They can be
7372 converted to function pointers directly:
7375 fptr p1 = (fptr)(&A::foo);
7378 @opindex Wno-pmf-conversions
7379 You must specify @option{-Wno-pmf-conversions} to use this extension.
7381 @node C++ Attributes
7382 @section C++-Specific Variable, Function, and Type Attributes
7384 Some attributes only make sense for C++ programs.
7387 @item init_priority (@var{priority})
7388 @cindex init_priority attribute
7391 In Standard C++, objects defined at namespace scope are guaranteed to be
7392 initialized in an order in strict accordance with that of their definitions
7393 @emph{in a given translation unit}. No guarantee is made for initializations
7394 across translation units. However, GNU C++ allows users to control the
7395 order of initialization of objects defined at namespace scope with the
7396 @code{init_priority} attribute by specifying a relative @var{priority},
7397 a constant integral expression currently bounded between 101 and 65535
7398 inclusive. Lower numbers indicate a higher priority.
7400 In the following example, @code{A} would normally be created before
7401 @code{B}, but the @code{init_priority} attribute has reversed that order:
7404 Some_Class A __attribute__ ((init_priority (2000)));
7405 Some_Class B __attribute__ ((init_priority (543)));
7409 Note that the particular values of @var{priority} do not matter; only their
7412 @item java_interface
7413 @cindex java_interface attribute
7415 This type attribute informs C++ that the class is a Java interface. It may
7416 only be applied to classes declared within an @code{extern "Java"} block.
7417 Calls to methods declared in this interface will be dispatched using GCJ's
7418 interface table mechanism, instead of regular virtual table dispatch.
7422 @node Java Exceptions
7423 @section Java Exceptions
7425 The Java language uses a slightly different exception handling model
7426 from C++. Normally, GNU C++ will automatically detect when you are
7427 writing C++ code that uses Java exceptions, and handle them
7428 appropriately. However, if C++ code only needs to execute destructors
7429 when Java exceptions are thrown through it, GCC will guess incorrectly.
7430 Sample problematic code is:
7433 struct S @{ ~S(); @};
7434 extern void bar(); // is written in Java, and may throw exceptions
7443 The usual effect of an incorrect guess is a link failure, complaining of
7444 a missing routine called @samp{__gxx_personality_v0}.
7446 You can inform the compiler that Java exceptions are to be used in a
7447 translation unit, irrespective of what it might think, by writing
7448 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
7449 @samp{#pragma} must appear before any functions that throw or catch
7450 exceptions, or run destructors when exceptions are thrown through them.
7452 You cannot mix Java and C++ exceptions in the same translation unit. It
7453 is believed to be safe to throw a C++ exception from one file through
7454 another file compiled for the Java exception model, or vice versa, but
7455 there may be bugs in this area.
7457 @node Deprecated Features
7458 @section Deprecated Features
7460 In the past, the GNU C++ compiler was extended to experiment with new
7461 features, at a time when the C++ language was still evolving. Now that
7462 the C++ standard is complete, some of those features are superseded by
7463 superior alternatives. Using the old features might cause a warning in
7464 some cases that the feature will be dropped in the future. In other
7465 cases, the feature might be gone already.
7467 While the list below is not exhaustive, it documents some of the options
7468 that are now deprecated:
7471 @item -fexternal-templates
7472 @itemx -falt-external-templates
7473 These are two of the many ways for g++ to implement template
7474 instantiation. @xref{Template Instantiation}. The C++ standard clearly
7475 defines how template definitions have to be organized across
7476 implementation units. g++ has an implicit instantiation mechanism that
7477 should work just fine for standard-conforming code.
7479 @item -fstrict-prototype
7480 @itemx -fno-strict-prototype
7481 Previously it was possible to use an empty prototype parameter list to
7482 indicate an unspecified number of parameters (like C), rather than no
7483 parameters, as C++ demands. This feature has been removed, except where
7484 it is required for backwards compatibility @xref{Backwards Compatibility}.
7487 The named return value extension has been deprecated, and is now
7490 The use of initializer lists with new expressions has been deprecated,
7491 and is now removed from g++.
7493 Floating and complex non-type template parameters have been deprecated,
7494 and are now removed from g++.
7496 The implicit typename extension has been deprecated and will be removed
7497 from g++ at some point. In some cases g++ determines that a dependent
7498 type such as @code{TPL<T>::X} is a type without needing a
7499 @code{typename} keyword, contrary to the standard.
7501 @node Backwards Compatibility
7502 @section Backwards Compatibility
7503 @cindex Backwards Compatibility
7504 @cindex ARM [Annotated C++ Reference Manual]
7506 Now that there is a definitive ISO standard C++, G++ has a specification
7507 to adhere to. The C++ language evolved over time, and features that
7508 used to be acceptable in previous drafts of the standard, such as the ARM
7509 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
7510 compilation of C++ written to such drafts, G++ contains some backwards
7511 compatibilities. @emph{All such backwards compatibility features are
7512 liable to disappear in future versions of G++.} They should be considered
7513 deprecated @xref{Deprecated Features}.
7517 If a variable is declared at for scope, it used to remain in scope until
7518 the end of the scope which contained the for statement (rather than just
7519 within the for scope). G++ retains this, but issues a warning, if such a
7520 variable is accessed outside the for scope.
7522 @item Implicit C language
7523 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
7524 scope to set the language. On such systems, all header files are
7525 implicitly scoped inside a C language scope. Also, an empty prototype
7526 @code{()} will be treated as an unspecified number of arguments, rather
7527 than no arguments, as C++ demands.