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 * Naming Types:: Giving a name to the type of some expression.
431 * Typeof:: @code{typeof}: referring to the type of an expression.
432 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
433 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
434 * Long Long:: Double-word integers---@code{long long int}.
435 * Complex:: Data types for complex numbers.
436 * Hex Floats:: Hexadecimal floating-point constants.
437 * Zero Length:: Zero-length arrays.
438 * Variable Length:: Arrays whose length is computed at run time.
439 * Variadic Macros:: Macros with a variable number of arguments.
440 * Escaped Newlines:: Slightly looser rules for escaped newlines.
441 * Multi-line Strings:: String literals with embedded newlines.
442 * Subscripting:: Any array can be subscripted, even if not an lvalue.
443 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
444 * Initializers:: Non-constant initializers.
445 * Compound Literals:: Compound literals give structures, unions
447 * Designated Inits:: Labeling elements of initializers.
448 * Cast to Union:: Casting to union type from any member of the union.
449 * Case Ranges:: `case 1 ... 9' and such.
450 * Mixed Declarations:: Mixing declarations and code.
451 * Function Attributes:: Declaring that functions have no side effects,
452 or that they can never return.
453 * Attribute Syntax:: Formal syntax for attributes.
454 * Function Prototypes:: Prototype declarations and old-style definitions.
455 * C++ Comments:: C++ comments are recognized.
456 * Dollar Signs:: Dollar sign is allowed in identifiers.
457 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
458 * Variable Attributes:: Specifying attributes of variables.
459 * Type Attributes:: Specifying attributes of types.
460 * Alignment:: Inquiring about the alignment of a type or variable.
461 * Inline:: Defining inline functions (as fast as macros).
462 * Extended Asm:: Assembler instructions with C expressions as operands.
463 (With them you can define ``built-in'' functions.)
464 * Constraints:: Constraints for asm operands
465 * Asm Labels:: Specifying the assembler name to use for a C symbol.
466 * Explicit Reg Vars:: Defining variables residing in specified registers.
467 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
468 * Incomplete Enums:: @code{enum foo;}, with details to follow.
469 * Function Names:: Printable strings which are the name of the current
471 * Return Address:: Getting the return or frame address of a function.
472 * Vector Extensions:: Using vector instructions through built-in functions.
473 * Other Builtins:: Other built-in functions.
474 * Target Builtins:: Built-in functions specific to particular targets.
475 * Pragmas:: Pragmas accepted by GCC.
476 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
477 * Thread-Local:: Per-thread variables.
480 @node Statement Exprs
481 @section Statements and Declarations in Expressions
482 @cindex statements inside expressions
483 @cindex declarations inside expressions
484 @cindex expressions containing statements
485 @cindex macros, statements in expressions
487 @c the above section title wrapped and causes an underfull hbox.. i
488 @c changed it from "within" to "in". --mew 4feb93
490 A compound statement enclosed in parentheses may appear as an expression
491 in GNU C@. This allows you to use loops, switches, and local variables
492 within an expression.
494 Recall that a compound statement is a sequence of statements surrounded
495 by braces; in this construct, parentheses go around the braces. For
499 (@{ int y = foo (); int z;
506 is a valid (though slightly more complex than necessary) expression
507 for the absolute value of @code{foo ()}.
509 The last thing in the compound statement should be an expression
510 followed by a semicolon; the value of this subexpression serves as the
511 value of the entire construct. (If you use some other kind of statement
512 last within the braces, the construct has type @code{void}, and thus
513 effectively no value.)
515 This feature is especially useful in making macro definitions ``safe'' (so
516 that they evaluate each operand exactly once). For example, the
517 ``maximum'' function is commonly defined as a macro in standard C as
521 #define max(a,b) ((a) > (b) ? (a) : (b))
525 @cindex side effects, macro argument
526 But this definition computes either @var{a} or @var{b} twice, with bad
527 results if the operand has side effects. In GNU C, if you know the
528 type of the operands (here let's assume @code{int}), you can define
529 the macro safely as follows:
532 #define maxint(a,b) \
533 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
536 Embedded statements are not allowed in constant expressions, such as
537 the value of an enumeration constant, the width of a bit-field, or
538 the initial value of a static variable.
540 If you don't know the type of the operand, you can still do this, but you
541 must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
544 Statement expressions are not supported fully in G++, and their fate
545 there is unclear. (It is possible that they will become fully supported
546 at some point, or that they will be deprecated, or that the bugs that
547 are present will continue to exist indefinitely.) Presently, statement
548 expressions do not work well as default arguments.
550 In addition, there are semantic issues with statement-expressions in
551 C++. If you try to use statement-expressions instead of inline
552 functions in C++, you may be surprised at the way object destruction is
553 handled. For example:
556 #define foo(a) (@{int b = (a); b + 3; @})
560 does not work the same way as:
563 inline int foo(int a) @{ int b = a; return b + 3; @}
567 In particular, if the expression passed into @code{foo} involves the
568 creation of temporaries, the destructors for those temporaries will be
569 run earlier in the case of the macro than in the case of the function.
571 These considerations mean that it is probably a bad idea to use
572 statement-expressions of this form in header files that are designed to
573 work with C++. (Note that some versions of the GNU C Library contained
574 header files using statement-expression that lead to precisely this
578 @section Locally Declared Labels
580 @cindex macros, local labels
582 Each statement expression is a scope in which @dfn{local labels} can be
583 declared. A local label is simply an identifier; you can jump to it
584 with an ordinary @code{goto} statement, but only from within the
585 statement expression it belongs to.
587 A local label declaration looks like this:
590 __label__ @var{label};
597 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
600 Local label declarations must come at the beginning of the statement
601 expression, right after the @samp{(@{}, before any ordinary
604 The label declaration defines the label @emph{name}, but does not define
605 the label itself. You must do this in the usual way, with
606 @code{@var{label}:}, within the statements of the statement expression.
608 The local label feature is useful because statement expressions are
609 often used in macros. If the macro contains nested loops, a @code{goto}
610 can be useful for breaking out of them. However, an ordinary label
611 whose scope is the whole function cannot be used: if the macro can be
612 expanded several times in one function, the label will be multiply
613 defined in that function. A local label avoids this problem. For
617 #define SEARCH(array, target) \
620 typeof (target) _SEARCH_target = (target); \
621 typeof (*(array)) *_SEARCH_array = (array); \
624 for (i = 0; i < max; i++) \
625 for (j = 0; j < max; j++) \
626 if (_SEARCH_array[i][j] == _SEARCH_target) \
627 @{ value = i; goto found; @} \
634 @node Labels as Values
635 @section Labels as Values
636 @cindex labels as values
637 @cindex computed gotos
638 @cindex goto with computed label
639 @cindex address of a label
641 You can get the address of a label defined in the current function
642 (or a containing function) with the unary operator @samp{&&}. The
643 value has type @code{void *}. This value is a constant and can be used
644 wherever a constant of that type is valid. For example:
652 To use these values, you need to be able to jump to one. This is done
653 with the computed goto statement@footnote{The analogous feature in
654 Fortran is called an assigned goto, but that name seems inappropriate in
655 C, where one can do more than simply store label addresses in label
656 variables.}, @code{goto *@var{exp};}. For example,
663 Any expression of type @code{void *} is allowed.
665 One way of using these constants is in initializing a static array that
666 will serve as a jump table:
669 static void *array[] = @{ &&foo, &&bar, &&hack @};
672 Then you can select a label with indexing, like this:
679 Note that this does not check whether the subscript is in bounds---array
680 indexing in C never does that.
682 Such an array of label values serves a purpose much like that of the
683 @code{switch} statement. The @code{switch} statement is cleaner, so
684 use that rather than an array unless the problem does not fit a
685 @code{switch} statement very well.
687 Another use of label values is in an interpreter for threaded code.
688 The labels within the interpreter function can be stored in the
689 threaded code for super-fast dispatching.
691 You may not use this mechanism to jump to code in a different function.
692 If you do that, totally unpredictable things will happen. The best way to
693 avoid this is to store the label address only in automatic variables and
694 never pass it as an argument.
696 An alternate way to write the above example is
699 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
701 goto *(&&foo + array[i]);
705 This is more friendly to code living in shared libraries, as it reduces
706 the number of dynamic relocations that are needed, and by consequence,
707 allows the data to be read-only.
709 @node Nested Functions
710 @section Nested Functions
711 @cindex nested functions
712 @cindex downward funargs
715 A @dfn{nested function} is a function defined inside another function.
716 (Nested functions are not supported for GNU C++.) The nested function's
717 name is local to the block where it is defined. For example, here we
718 define a nested function named @code{square}, and call it twice:
722 foo (double a, double b)
724 double square (double z) @{ return z * z; @}
726 return square (a) + square (b);
731 The nested function can access all the variables of the containing
732 function that are visible at the point of its definition. This is
733 called @dfn{lexical scoping}. For example, here we show a nested
734 function which uses an inherited variable named @code{offset}:
738 bar (int *array, int offset, int size)
740 int access (int *array, int index)
741 @{ return array[index + offset]; @}
744 for (i = 0; i < size; i++)
745 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
750 Nested function definitions are permitted within functions in the places
751 where variable definitions are allowed; that is, in any block, before
752 the first statement in the block.
754 It is possible to call the nested function from outside the scope of its
755 name by storing its address or passing the address to another function:
758 hack (int *array, int size)
760 void store (int index, int value)
761 @{ array[index] = value; @}
763 intermediate (store, size);
767 Here, the function @code{intermediate} receives the address of
768 @code{store} as an argument. If @code{intermediate} calls @code{store},
769 the arguments given to @code{store} are used to store into @code{array}.
770 But this technique works only so long as the containing function
771 (@code{hack}, in this example) does not exit.
773 If you try to call the nested function through its address after the
774 containing function has exited, all hell will break loose. If you try
775 to call it after a containing scope level has exited, and if it refers
776 to some of the variables that are no longer in scope, you may be lucky,
777 but it's not wise to take the risk. If, however, the nested function
778 does not refer to anything that has gone out of scope, you should be
781 GCC implements taking the address of a nested function using a technique
782 called @dfn{trampolines}. A paper describing them is available as
785 @uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}.
787 A nested function can jump to a label inherited from a containing
788 function, provided the label was explicitly declared in the containing
789 function (@pxref{Local Labels}). Such a jump returns instantly to the
790 containing function, exiting the nested function which did the
791 @code{goto} and any intermediate functions as well. Here is an example:
795 bar (int *array, int offset, int size)
798 int access (int *array, int index)
802 return array[index + offset];
806 for (i = 0; i < size; i++)
807 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
811 /* @r{Control comes here from @code{access}
812 if it detects an error.} */
819 A nested function always has internal linkage. Declaring one with
820 @code{extern} is erroneous. If you need to declare the nested function
821 before its definition, use @code{auto} (which is otherwise meaningless
822 for function declarations).
825 bar (int *array, int offset, int size)
828 auto int access (int *, int);
830 int access (int *array, int index)
834 return array[index + offset];
840 @node Constructing Calls
841 @section Constructing Function Calls
842 @cindex constructing calls
843 @cindex forwarding calls
845 Using the built-in functions described below, you can record
846 the arguments a function received, and call another function
847 with the same arguments, without knowing the number or types
850 You can also record the return value of that function call,
851 and later return that value, without knowing what data type
852 the function tried to return (as long as your caller expects
855 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
856 This built-in function returns a pointer to data
857 describing how to perform a call with the same arguments as were passed
858 to the current function.
860 The function saves the arg pointer register, structure value address,
861 and all registers that might be used to pass arguments to a function
862 into a block of memory allocated on the stack. Then it returns the
863 address of that block.
866 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
867 This built-in function invokes @var{function}
868 with a copy of the parameters described by @var{arguments}
871 The value of @var{arguments} should be the value returned by
872 @code{__builtin_apply_args}. The argument @var{size} specifies the size
873 of the stack argument data, in bytes.
875 This function returns a pointer to data describing
876 how to return whatever value was returned by @var{function}. The data
877 is saved in a block of memory allocated on the stack.
879 It is not always simple to compute the proper value for @var{size}. The
880 value is used by @code{__builtin_apply} to compute the amount of data
881 that should be pushed on the stack and copied from the incoming argument
885 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
886 This built-in function returns the value described by @var{result} from
887 the containing function. You should specify, for @var{result}, a value
888 returned by @code{__builtin_apply}.
892 @section Naming an Expression's Type
895 You can give a name to the type of an expression using a @code{typedef}
896 declaration with an initializer. Here is how to define @var{name} as a
897 type name for the type of @var{exp}:
900 typedef @var{name} = @var{exp};
903 This is useful in conjunction with the statements-within-expressions
904 feature. Here is how the two together can be used to define a safe
905 ``maximum'' macro that operates on any arithmetic type:
909 (@{typedef _ta = (a), _tb = (b); \
910 _ta _a = (a); _tb _b = (b); \
911 _a > _b ? _a : _b; @})
914 @cindex underscores in variables in macros
915 @cindex @samp{_} in variables in macros
916 @cindex local variables in macros
917 @cindex variables, local, in macros
918 @cindex macros, local variables in
920 The reason for using names that start with underscores for the local
921 variables is to avoid conflicts with variable names that occur within the
922 expressions that are substituted for @code{a} and @code{b}. Eventually we
923 hope to design a new form of declaration syntax that allows you to declare
924 variables whose scopes start only after their initializers; this will be a
925 more reliable way to prevent such conflicts.
928 @section Referring to a Type with @code{typeof}
931 @cindex macros, types of arguments
933 Another way to refer to the type of an expression is with @code{typeof}.
934 The syntax of using of this keyword looks like @code{sizeof}, but the
935 construct acts semantically like a type name defined with @code{typedef}.
937 There are two ways of writing the argument to @code{typeof}: with an
938 expression or with a type. Here is an example with an expression:
945 This assumes that @code{x} is an array of pointers to functions;
946 the type described is that of the values of the functions.
948 Here is an example with a typename as the argument:
955 Here the type described is that of pointers to @code{int}.
957 If you are writing a header file that must work when included in ISO C
958 programs, write @code{__typeof__} instead of @code{typeof}.
959 @xref{Alternate Keywords}.
961 A @code{typeof}-construct can be used anywhere a typedef name could be
962 used. For example, you can use it in a declaration, in a cast, or inside
963 of @code{sizeof} or @code{typeof}.
967 This declares @code{y} with the type of what @code{x} points to.
974 This declares @code{y} as an array of such values.
981 This declares @code{y} as an array of pointers to characters:
984 typeof (typeof (char *)[4]) y;
988 It is equivalent to the following traditional C declaration:
994 To see the meaning of the declaration using @code{typeof}, and why it
995 might be a useful way to write, let's rewrite it with these macros:
998 #define pointer(T) typeof(T *)
999 #define array(T, N) typeof(T [N])
1003 Now the declaration can be rewritten this way:
1006 array (pointer (char), 4) y;
1010 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1011 pointers to @code{char}.
1015 @section Generalized Lvalues
1016 @cindex compound expressions as lvalues
1017 @cindex expressions, compound, as lvalues
1018 @cindex conditional expressions as lvalues
1019 @cindex expressions, conditional, as lvalues
1020 @cindex casts as lvalues
1021 @cindex generalized lvalues
1022 @cindex lvalues, generalized
1023 @cindex extensions, @code{?:}
1024 @cindex @code{?:} extensions
1025 Compound expressions, conditional expressions and casts are allowed as
1026 lvalues provided their operands are lvalues. This means that you can take
1027 their addresses or store values into them.
1029 Standard C++ allows compound expressions and conditional expressions as
1030 lvalues, and permits casts to reference type, so use of this extension
1031 is deprecated for C++ code.
1033 For example, a compound expression can be assigned, provided the last
1034 expression in the sequence is an lvalue. These two expressions are
1042 Similarly, the address of the compound expression can be taken. These two
1043 expressions are equivalent:
1050 A conditional expression is a valid lvalue if its type is not void and the
1051 true and false branches are both valid lvalues. For example, these two
1052 expressions are equivalent:
1056 (a ? b = 5 : (c = 5))
1059 A cast is a valid lvalue if its operand is an lvalue. A simple
1060 assignment whose left-hand side is a cast works by converting the
1061 right-hand side first to the specified type, then to the type of the
1062 inner left-hand side expression. After this is stored, the value is
1063 converted back to the specified type to become the value of the
1064 assignment. Thus, if @code{a} has type @code{char *}, the following two
1065 expressions are equivalent:
1069 (int)(a = (char *)(int)5)
1072 An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1073 performs the arithmetic using the type resulting from the cast, and then
1074 continues as in the previous case. Therefore, these two expressions are
1079 (int)(a = (char *)(int) ((int)a + 5))
1082 You cannot take the address of an lvalue cast, because the use of its
1083 address would not work out coherently. Suppose that @code{&(int)f} were
1084 permitted, where @code{f} has type @code{float}. Then the following
1085 statement would try to store an integer bit-pattern where a floating
1086 point number belongs:
1092 This is quite different from what @code{(int)f = 1} would do---that
1093 would convert 1 to floating point and store it. Rather than cause this
1094 inconsistency, we think it is better to prohibit use of @samp{&} on a cast.
1096 If you really do want an @code{int *} pointer with the address of
1097 @code{f}, you can simply write @code{(int *)&f}.
1100 @section Conditionals with Omitted Operands
1101 @cindex conditional expressions, extensions
1102 @cindex omitted middle-operands
1103 @cindex middle-operands, omitted
1104 @cindex extensions, @code{?:}
1105 @cindex @code{?:} extensions
1107 The middle operand in a conditional expression may be omitted. Then
1108 if the first operand is nonzero, its value is the value of the conditional
1111 Therefore, the expression
1118 has the value of @code{x} if that is nonzero; otherwise, the value of
1121 This example is perfectly equivalent to
1127 @cindex side effect in ?:
1128 @cindex ?: side effect
1130 In this simple case, the ability to omit the middle operand is not
1131 especially useful. When it becomes useful is when the first operand does,
1132 or may (if it is a macro argument), contain a side effect. Then repeating
1133 the operand in the middle would perform the side effect twice. Omitting
1134 the middle operand uses the value already computed without the undesirable
1135 effects of recomputing it.
1138 @section Double-Word Integers
1139 @cindex @code{long long} data types
1140 @cindex double-word arithmetic
1141 @cindex multiprecision arithmetic
1142 @cindex @code{LL} integer suffix
1143 @cindex @code{ULL} integer suffix
1145 ISO C99 supports data types for integers that are at least 64 bits wide,
1146 and as an extension GCC supports them in C89 mode and in C++.
1147 Simply write @code{long long int} for a signed integer, or
1148 @code{unsigned long long int} for an unsigned integer. To make an
1149 integer constant of type @code{long long int}, add the suffix @samp{LL}
1150 to the integer. To make an integer constant of type @code{unsigned long
1151 long int}, add the suffix @samp{ULL} to the integer.
1153 You can use these types in arithmetic like any other integer types.
1154 Addition, subtraction, and bitwise boolean operations on these types
1155 are open-coded on all types of machines. Multiplication is open-coded
1156 if the machine supports fullword-to-doubleword a widening multiply
1157 instruction. Division and shifts are open-coded only on machines that
1158 provide special support. The operations that are not open-coded use
1159 special library routines that come with GCC@.
1161 There may be pitfalls when you use @code{long long} types for function
1162 arguments, unless you declare function prototypes. If a function
1163 expects type @code{int} for its argument, and you pass a value of type
1164 @code{long long int}, confusion will result because the caller and the
1165 subroutine will disagree about the number of bytes for the argument.
1166 Likewise, if the function expects @code{long long int} and you pass
1167 @code{int}. The best way to avoid such problems is to use prototypes.
1170 @section Complex Numbers
1171 @cindex complex numbers
1172 @cindex @code{_Complex} keyword
1173 @cindex @code{__complex__} keyword
1175 ISO C99 supports complex floating data types, and as an extension GCC
1176 supports them in C89 mode and in C++, and supports complex integer data
1177 types which are not part of ISO C99. You can declare complex types
1178 using the keyword @code{_Complex}. As an extension, the older GNU
1179 keyword @code{__complex__} is also supported.
1181 For example, @samp{_Complex double x;} declares @code{x} as a
1182 variable whose real part and imaginary part are both of type
1183 @code{double}. @samp{_Complex short int y;} declares @code{y} to
1184 have real and imaginary parts of type @code{short int}; this is not
1185 likely to be useful, but it shows that the set of complex types is
1188 To write a constant with a complex data type, use the suffix @samp{i} or
1189 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
1190 has type @code{_Complex float} and @code{3i} has type
1191 @code{_Complex int}. Such a constant always has a pure imaginary
1192 value, but you can form any complex value you like by adding one to a
1193 real constant. This is a GNU extension; if you have an ISO C99
1194 conforming C library (such as GNU libc), and want to construct complex
1195 constants of floating type, you should include @code{<complex.h>} and
1196 use the macros @code{I} or @code{_Complex_I} instead.
1198 @cindex @code{__real__} keyword
1199 @cindex @code{__imag__} keyword
1200 To extract the real part of a complex-valued expression @var{exp}, write
1201 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
1202 extract the imaginary part. This is a GNU extension; for values of
1203 floating type, you should use the ISO C99 functions @code{crealf},
1204 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
1205 @code{cimagl}, declared in @code{<complex.h>} and also provided as
1206 built-in functions by GCC@.
1208 @cindex complex conjugation
1209 The operator @samp{~} performs complex conjugation when used on a value
1210 with a complex type. This is a GNU extension; for values of
1211 floating type, you should use the ISO C99 functions @code{conjf},
1212 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
1213 provided as built-in functions by GCC@.
1215 GCC can allocate complex automatic variables in a noncontiguous
1216 fashion; it's even possible for the real part to be in a register while
1217 the imaginary part is on the stack (or vice-versa). None of the
1218 supported debugging info formats has a way to represent noncontiguous
1219 allocation like this, so GCC describes a noncontiguous complex
1220 variable as if it were two separate variables of noncomplex type.
1221 If the variable's actual name is @code{foo}, the two fictitious
1222 variables are named @code{foo$real} and @code{foo$imag}. You can
1223 examine and set these two fictitious variables with your debugger.
1225 A future version of GDB will know how to recognize such pairs and treat
1226 them as a single variable with a complex type.
1232 ISO C99 supports floating-point numbers written not only in the usual
1233 decimal notation, such as @code{1.55e1}, but also numbers such as
1234 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1235 supports this in C89 mode (except in some cases when strictly
1236 conforming) and in C++. In that format the
1237 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1238 mandatory. The exponent is a decimal number that indicates the power of
1239 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1246 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1247 is the same as @code{1.55e1}.
1249 Unlike for floating-point numbers in the decimal notation the exponent
1250 is always required in the hexadecimal notation. Otherwise the compiler
1251 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1252 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1253 extension for floating-point constants of type @code{float}.
1256 @section Arrays of Length Zero
1257 @cindex arrays of length zero
1258 @cindex zero-length arrays
1259 @cindex length-zero arrays
1260 @cindex flexible array members
1262 Zero-length arrays are allowed in GNU C@. They are very useful as the
1263 last element of a structure which is really a header for a variable-length
1272 struct line *thisline = (struct line *)
1273 malloc (sizeof (struct line) + this_length);
1274 thisline->length = this_length;
1277 In ISO C90, you would have to give @code{contents} a length of 1, which
1278 means either you waste space or complicate the argument to @code{malloc}.
1280 In ISO C99, you would use a @dfn{flexible array member}, which is
1281 slightly different in syntax and semantics:
1285 Flexible array members are written as @code{contents[]} without
1289 Flexible array members have incomplete type, and so the @code{sizeof}
1290 operator may not be applied. As a quirk of the original implementation
1291 of zero-length arrays, @code{sizeof} evaluates to zero.
1294 Flexible array members may only appear as the last member of a
1295 @code{struct} that is otherwise non-empty.
1298 GCC versions before 3.0 allowed zero-length arrays to be statically
1299 initialized, as if they were flexible arrays. In addition to those
1300 cases that were useful, it also allowed initializations in situations
1301 that would corrupt later data. Non-empty initialization of zero-length
1302 arrays is now treated like any case where there are more initializer
1303 elements than the array holds, in that a suitable warning about "excess
1304 elements in array" is given, and the excess elements (all of them, in
1305 this case) are ignored.
1307 Instead GCC allows static initialization of flexible array members.
1308 This is equivalent to defining a new structure containing the original
1309 structure followed by an array of sufficient size to contain the data.
1310 I.e.@: in the following, @code{f1} is constructed as if it were declared
1316 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1319 struct f1 f1; int data[3];
1320 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1324 The convenience of this extension is that @code{f1} has the desired
1325 type, eliminating the need to consistently refer to @code{f2.f1}.
1327 This has symmetry with normal static arrays, in that an array of
1328 unknown size is also written with @code{[]}.
1330 Of course, this extension only makes sense if the extra data comes at
1331 the end of a top-level object, as otherwise we would be overwriting
1332 data at subsequent offsets. To avoid undue complication and confusion
1333 with initialization of deeply nested arrays, we simply disallow any
1334 non-empty initialization except when the structure is the top-level
1335 object. For example:
1338 struct foo @{ int x; int y[]; @};
1339 struct bar @{ struct foo z; @};
1341 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1342 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1343 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1344 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1347 @node Variable Length
1348 @section Arrays of Variable Length
1349 @cindex variable-length arrays
1350 @cindex arrays of variable length
1353 Variable-length automatic arrays are allowed in ISO C99, and as an
1354 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1355 implementation of variable-length arrays does not yet conform in detail
1356 to the ISO C99 standard.) These arrays are
1357 declared like any other automatic arrays, but with a length that is not
1358 a constant expression. The storage is allocated at the point of
1359 declaration and deallocated when the brace-level is exited. For
1364 concat_fopen (char *s1, char *s2, char *mode)
1366 char str[strlen (s1) + strlen (s2) + 1];
1369 return fopen (str, mode);
1373 @cindex scope of a variable length array
1374 @cindex variable-length array scope
1375 @cindex deallocating variable length arrays
1376 Jumping or breaking out of the scope of the array name deallocates the
1377 storage. Jumping into the scope is not allowed; you get an error
1380 @cindex @code{alloca} vs variable-length arrays
1381 You can use the function @code{alloca} to get an effect much like
1382 variable-length arrays. The function @code{alloca} is available in
1383 many other C implementations (but not in all). On the other hand,
1384 variable-length arrays are more elegant.
1386 There are other differences between these two methods. Space allocated
1387 with @code{alloca} exists until the containing @emph{function} returns.
1388 The space for a variable-length array is deallocated as soon as the array
1389 name's scope ends. (If you use both variable-length arrays and
1390 @code{alloca} in the same function, deallocation of a variable-length array
1391 will also deallocate anything more recently allocated with @code{alloca}.)
1393 You can also use variable-length arrays as arguments to functions:
1397 tester (int len, char data[len][len])
1403 The length of an array is computed once when the storage is allocated
1404 and is remembered for the scope of the array in case you access it with
1407 If you want to pass the array first and the length afterward, you can
1408 use a forward declaration in the parameter list---another GNU extension.
1412 tester (int len; char data[len][len], int len)
1418 @cindex parameter forward declaration
1419 The @samp{int len} before the semicolon is a @dfn{parameter forward
1420 declaration}, and it serves the purpose of making the name @code{len}
1421 known when the declaration of @code{data} is parsed.
1423 You can write any number of such parameter forward declarations in the
1424 parameter list. They can be separated by commas or semicolons, but the
1425 last one must end with a semicolon, which is followed by the ``real''
1426 parameter declarations. Each forward declaration must match a ``real''
1427 declaration in parameter name and data type. ISO C99 does not support
1428 parameter forward declarations.
1430 @node Variadic Macros
1431 @section Macros with a Variable Number of Arguments.
1432 @cindex variable number of arguments
1433 @cindex macro with variable arguments
1434 @cindex rest argument (in macro)
1435 @cindex variadic macros
1437 In the ISO C standard of 1999, a macro can be declared to accept a
1438 variable number of arguments much as a function can. The syntax for
1439 defining the macro is similar to that of a function. Here is an
1443 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1446 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1447 such a macro, it represents the zero or more tokens until the closing
1448 parenthesis that ends the invocation, including any commas. This set of
1449 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1450 wherever it appears. See the CPP manual for more information.
1452 GCC has long supported variadic macros, and used a different syntax that
1453 allowed you to give a name to the variable arguments just like any other
1454 argument. Here is an example:
1457 #define debug(format, args...) fprintf (stderr, format, args)
1460 This is in all ways equivalent to the ISO C example above, but arguably
1461 more readable and descriptive.
1463 GNU CPP has two further variadic macro extensions, and permits them to
1464 be used with either of the above forms of macro definition.
1466 In standard C, you are not allowed to leave the variable argument out
1467 entirely; but you are allowed to pass an empty argument. For example,
1468 this invocation is invalid in ISO C, because there is no comma after
1475 GNU CPP permits you to completely omit the variable arguments in this
1476 way. In the above examples, the compiler would complain, though since
1477 the expansion of the macro still has the extra comma after the format
1480 To help solve this problem, CPP behaves specially for variable arguments
1481 used with the token paste operator, @samp{##}. If instead you write
1484 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1487 and if the variable arguments are omitted or empty, the @samp{##}
1488 operator causes the preprocessor to remove the comma before it. If you
1489 do provide some variable arguments in your macro invocation, GNU CPP
1490 does not complain about the paste operation and instead places the
1491 variable arguments after the comma. Just like any other pasted macro
1492 argument, these arguments are not macro expanded.
1494 @node Escaped Newlines
1495 @section Slightly Looser Rules for Escaped Newlines
1496 @cindex escaped newlines
1497 @cindex newlines (escaped)
1499 Recently, the preprocessor has relaxed its treatment of escaped
1500 newlines. Previously, the newline had to immediately follow a
1501 backslash. The current implementation allows whitespace in the form of
1502 spaces, horizontal and vertical tabs, and form feeds between the
1503 backslash and the subsequent newline. The preprocessor issues a
1504 warning, but treats it as a valid escaped newline and combines the two
1505 lines to form a single logical line. This works within comments and
1506 tokens, including multi-line strings, as well as between tokens.
1507 Comments are @emph{not} treated as whitespace for the purposes of this
1508 relaxation, since they have not yet been replaced with spaces.
1510 @node Multi-line Strings
1511 @section String Literals with Embedded Newlines
1512 @cindex multi-line string literals
1514 As an extension, GNU CPP permits string literals to cross multiple lines
1515 without escaping the embedded newlines. Each embedded newline is
1516 replaced with a single @samp{\n} character in the resulting string
1517 literal, regardless of what form the newline took originally.
1519 CPP currently allows such strings in directives as well (other than the
1520 @samp{#include} family). This is deprecated and will eventually be
1524 @section Non-Lvalue Arrays May Have Subscripts
1525 @cindex subscripting
1526 @cindex arrays, non-lvalue
1528 @cindex subscripting and function values
1529 In ISO C99, arrays that are not lvalues still decay to pointers, and
1530 may be subscripted, although they may not be modified or used after
1531 the next sequence point and the unary @samp{&} operator may not be
1532 applied to them. As an extension, GCC allows such arrays to be
1533 subscripted in C89 mode, though otherwise they do not decay to
1534 pointers outside C99 mode. For example,
1535 this is valid in GNU C though not valid in C89:
1539 struct foo @{int a[4];@};
1545 return f().a[index];
1551 @section Arithmetic on @code{void}- and Function-Pointers
1552 @cindex void pointers, arithmetic
1553 @cindex void, size of pointer to
1554 @cindex function pointers, arithmetic
1555 @cindex function, size of pointer to
1557 In GNU C, addition and subtraction operations are supported on pointers to
1558 @code{void} and on pointers to functions. This is done by treating the
1559 size of a @code{void} or of a function as 1.
1561 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1562 and on function types, and returns 1.
1564 @opindex Wpointer-arith
1565 The option @option{-Wpointer-arith} requests a warning if these extensions
1569 @section Non-Constant Initializers
1570 @cindex initializers, non-constant
1571 @cindex non-constant initializers
1573 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1574 automatic variable are not required to be constant expressions in GNU C@.
1575 Here is an example of an initializer with run-time varying elements:
1578 foo (float f, float g)
1580 float beat_freqs[2] = @{ f-g, f+g @};
1585 @node Compound Literals
1586 @section Compound Literals
1587 @cindex constructor expressions
1588 @cindex initializations in expressions
1589 @cindex structures, constructor expression
1590 @cindex expressions, constructor
1591 @cindex compound literals
1592 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1594 ISO C99 supports compound literals. A compound literal looks like
1595 a cast containing an initializer. Its value is an object of the
1596 type specified in the cast, containing the elements specified in
1597 the initializer; it is an lvalue. As an extension, GCC supports
1598 compound literals in C89 mode and in C++.
1600 Usually, the specified type is a structure. Assume that
1601 @code{struct foo} and @code{structure} are declared as shown:
1604 struct foo @{int a; char b[2];@} structure;
1608 Here is an example of constructing a @code{struct foo} with a compound literal:
1611 structure = ((struct foo) @{x + y, 'a', 0@});
1615 This is equivalent to writing the following:
1619 struct foo temp = @{x + y, 'a', 0@};
1624 You can also construct an array. If all the elements of the compound literal
1625 are (made up of) simple constant expressions, suitable for use in
1626 initializers of objects of static storage duration, then the compound
1627 literal can be coerced to a pointer to its first element and used in
1628 such an initializer, as shown here:
1631 char **foo = (char *[]) @{ "x", "y", "z" @};
1634 Compound literals for scalar types and union types are is
1635 also allowed, but then the compound literal is equivalent
1638 As a GNU extension, GCC allows initialization of objects with static storage
1639 duration by compound literals (which is not possible in ISO C99, because
1640 the initializer is not a constant).
1641 It is handled as if the object was initialized only with the bracket
1642 enclosed list if compound literal's and object types match.
1643 The initializer list of the compound literal must be constant.
1644 If the object being initialized has array type of unknown size, the size is
1645 determined by compound literal size.
1648 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1649 static int y[] = (int []) @{1, 2, 3@};
1650 static int z[] = (int [3]) @{1@};
1654 The above lines are equivalent to the following:
1656 static struct foo x = @{1, 'a', 'b'@};
1657 static int y[] = @{1, 2, 3@};
1658 static int z[] = @{1, 0, 0@};
1661 @node Designated Inits
1662 @section Designated Initializers
1663 @cindex initializers with labeled elements
1664 @cindex labeled elements in initializers
1665 @cindex case labels in initializers
1666 @cindex designated initializers
1668 Standard C89 requires the elements of an initializer to appear in a fixed
1669 order, the same as the order of the elements in the array or structure
1672 In ISO C99 you can give the elements in any order, specifying the array
1673 indices or structure field names they apply to, and GNU C allows this as
1674 an extension in C89 mode as well. This extension is not
1675 implemented in GNU C++.
1677 To specify an array index, write
1678 @samp{[@var{index}] =} before the element value. For example,
1681 int a[6] = @{ [4] = 29, [2] = 15 @};
1688 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1692 The index values must be constant expressions, even if the array being
1693 initialized is automatic.
1695 An alternative syntax for this which has been obsolete since GCC 2.5 but
1696 GCC still accepts is to write @samp{[@var{index}]} before the element
1697 value, with no @samp{=}.
1699 To initialize a range of elements to the same value, write
1700 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1701 extension. For example,
1704 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1708 If the value in it has side-effects, the side-effects will happen only once,
1709 not for each initialized field by the range initializer.
1712 Note that the length of the array is the highest value specified
1715 In a structure initializer, specify the name of a field to initialize
1716 with @samp{.@var{fieldname} =} before the element value. For example,
1717 given the following structure,
1720 struct point @{ int x, y; @};
1724 the following initialization
1727 struct point p = @{ .y = yvalue, .x = xvalue @};
1734 struct point p = @{ xvalue, yvalue @};
1737 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1738 @samp{@var{fieldname}:}, as shown here:
1741 struct point p = @{ y: yvalue, x: xvalue @};
1745 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1746 @dfn{designator}. You can also use a designator (or the obsolete colon
1747 syntax) when initializing a union, to specify which element of the union
1748 should be used. For example,
1751 union foo @{ int i; double d; @};
1753 union foo f = @{ .d = 4 @};
1757 will convert 4 to a @code{double} to store it in the union using
1758 the second element. By contrast, casting 4 to type @code{union foo}
1759 would store it into the union as the integer @code{i}, since it is
1760 an integer. (@xref{Cast to Union}.)
1762 You can combine this technique of naming elements with ordinary C
1763 initialization of successive elements. Each initializer element that
1764 does not have a designator applies to the next consecutive element of the
1765 array or structure. For example,
1768 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1775 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1778 Labeling the elements of an array initializer is especially useful
1779 when the indices are characters or belong to an @code{enum} type.
1784 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1785 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1788 @cindex designator lists
1789 You can also write a series of @samp{.@var{fieldname}} and
1790 @samp{[@var{index}]} designators before an @samp{=} to specify a
1791 nested subobject to initialize; the list is taken relative to the
1792 subobject corresponding to the closest surrounding brace pair. For
1793 example, with the @samp{struct point} declaration above:
1796 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1800 If the same field is initialized multiple times, it will have value from
1801 the last initialization. If any such overridden initialization has
1802 side-effect, it is unspecified whether the side-effect happens or not.
1803 Currently, gcc will discard them and issue a warning.
1806 @section Case Ranges
1808 @cindex ranges in case statements
1810 You can specify a range of consecutive values in a single @code{case} label,
1814 case @var{low} ... @var{high}:
1818 This has the same effect as the proper number of individual @code{case}
1819 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1821 This feature is especially useful for ranges of ASCII character codes:
1827 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1828 it may be parsed wrong when you use it with integer values. For example,
1843 @section Cast to a Union Type
1844 @cindex cast to a union
1845 @cindex union, casting to a
1847 A cast to union type is similar to other casts, except that the type
1848 specified is a union type. You can specify the type either with
1849 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1850 a constructor though, not a cast, and hence does not yield an lvalue like
1851 normal casts. (@xref{Compound Literals}.)
1853 The types that may be cast to the union type are those of the members
1854 of the union. Thus, given the following union and variables:
1857 union foo @{ int i; double d; @};
1863 both @code{x} and @code{y} can be cast to type @code{union foo}.
1865 Using the cast as the right-hand side of an assignment to a variable of
1866 union type is equivalent to storing in a member of the union:
1871 u = (union foo) x @equiv{} u.i = x
1872 u = (union foo) y @equiv{} u.d = y
1875 You can also use the union cast as a function argument:
1878 void hack (union foo);
1880 hack ((union foo) x);
1883 @node Mixed Declarations
1884 @section Mixed Declarations and Code
1885 @cindex mixed declarations and code
1886 @cindex declarations, mixed with code
1887 @cindex code, mixed with declarations
1889 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1890 within compound statements. As an extension, GCC also allows this in
1891 C89 mode. For example, you could do:
1900 Each identifier is visible from where it is declared until the end of
1901 the enclosing block.
1903 @node Function Attributes
1904 @section Declaring Attributes of Functions
1905 @cindex function attributes
1906 @cindex declaring attributes of functions
1907 @cindex functions that never return
1908 @cindex functions that have no side effects
1909 @cindex functions in arbitrary sections
1910 @cindex functions that behave like malloc
1911 @cindex @code{volatile} applied to function
1912 @cindex @code{const} applied to function
1913 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1914 @cindex functions with non-null pointer arguments
1915 @cindex functions that are passed arguments in registers on the 386
1916 @cindex functions that pop the argument stack on the 386
1917 @cindex functions that do not pop the argument stack on the 386
1919 In GNU C, you declare certain things about functions called in your program
1920 which help the compiler optimize function calls and check your code more
1923 The keyword @code{__attribute__} allows you to specify special
1924 attributes when making a declaration. This keyword is followed by an
1925 attribute specification inside double parentheses. The following
1926 attributes are currently defined for functions on all targets:
1927 @code{noreturn}, @code{noinline}, @code{always_inline},
1928 @code{pure}, @code{const}, @code{nothrow},
1929 @code{format}, @code{format_arg}, @code{no_instrument_function},
1930 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1931 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1932 @code{alias}, and @code{nonnull}. Several other attributes are defined
1933 for functions on particular target systems. Other attributes, including
1934 @code{section} are supported for variables declarations
1935 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1937 You may also specify attributes with @samp{__} preceding and following
1938 each keyword. This allows you to use them in header files without
1939 being concerned about a possible macro of the same name. For example,
1940 you may use @code{__noreturn__} instead of @code{noreturn}.
1942 @xref{Attribute Syntax}, for details of the exact syntax for using
1946 @cindex @code{noreturn} function attribute
1948 A few standard library functions, such as @code{abort} and @code{exit},
1949 cannot return. GCC knows this automatically. Some programs define
1950 their own functions that never return. You can declare them
1951 @code{noreturn} to tell the compiler this fact. For example,
1955 void fatal () __attribute__ ((noreturn));
1958 fatal (/* @r{@dots{}} */)
1960 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
1966 The @code{noreturn} keyword tells the compiler to assume that
1967 @code{fatal} cannot return. It can then optimize without regard to what
1968 would happen if @code{fatal} ever did return. This makes slightly
1969 better code. More importantly, it helps avoid spurious warnings of
1970 uninitialized variables.
1972 Do not assume that registers saved by the calling function are
1973 restored before calling the @code{noreturn} function.
1975 It does not make sense for a @code{noreturn} function to have a return
1976 type other than @code{void}.
1978 The attribute @code{noreturn} is not implemented in GCC versions
1979 earlier than 2.5. An alternative way to declare that a function does
1980 not return, which works in the current version and in some older
1981 versions, is as follows:
1984 typedef void voidfn ();
1986 volatile voidfn fatal;
1989 @cindex @code{noinline} function attribute
1991 This function attribute prevents a function from being considered for
1994 @cindex @code{always_inline} function attribute
1996 Generally, functions are not inlined unless optimization is specified.
1997 For functions declared inline, this attribute inlines the function even
1998 if no optimization level was specified.
2000 @cindex @code{pure} function attribute
2002 Many functions have no effects except the return value and their
2003 return value depends only on the parameters and/or global variables.
2004 Such a function can be subject
2005 to common subexpression elimination and loop optimization just as an
2006 arithmetic operator would be. These functions should be declared
2007 with the attribute @code{pure}. For example,
2010 int square (int) __attribute__ ((pure));
2014 says that the hypothetical function @code{square} is safe to call
2015 fewer times than the program says.
2017 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2018 Interesting non-pure functions are functions with infinite loops or those
2019 depending on volatile memory or other system resource, that may change between
2020 two consecutive calls (such as @code{feof} in a multithreading environment).
2022 The attribute @code{pure} is not implemented in GCC versions earlier
2024 @cindex @code{const} function attribute
2026 Many functions do not examine any values except their arguments, and
2027 have no effects except the return value. Basically this is just slightly
2028 more strict class than the @code{pure} attribute above, since function is not
2029 allowed to read global memory.
2031 @cindex pointer arguments
2032 Note that a function that has pointer arguments and examines the data
2033 pointed to must @emph{not} be declared @code{const}. Likewise, a
2034 function that calls a non-@code{const} function usually must not be
2035 @code{const}. It does not make sense for a @code{const} function to
2038 The attribute @code{const} is not implemented in GCC versions earlier
2039 than 2.5. An alternative way to declare that a function has no side
2040 effects, which works in the current version and in some older versions,
2044 typedef int intfn ();
2046 extern const intfn square;
2049 This approach does not work in GNU C++ from 2.6.0 on, since the language
2050 specifies that the @samp{const} must be attached to the return value.
2052 @cindex @code{nothrow} function attribute
2054 The @code{nothrow} attribute is used to inform the compiler that a
2055 function cannot throw an exception. For example, most functions in
2056 the standard C library can be guaranteed not to throw an exception
2057 with the notable exceptions of @code{qsort} and @code{bsearch} that
2058 take function pointer arguments. The @code{nothrow} attribute is not
2059 implemented in GCC versions earlier than 3.2.
2061 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2062 @cindex @code{format} function attribute
2064 The @code{format} attribute specifies that a function takes @code{printf},
2065 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2066 should be type-checked against a format string. For example, the
2071 my_printf (void *my_object, const char *my_format, ...)
2072 __attribute__ ((format (printf, 2, 3)));
2076 causes the compiler to check the arguments in calls to @code{my_printf}
2077 for consistency with the @code{printf} style format string argument
2080 The parameter @var{archetype} determines how the format string is
2081 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
2082 or @code{strfmon}. (You can also use @code{__printf__},
2083 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
2084 parameter @var{string-index} specifies which argument is the format
2085 string argument (starting from 1), while @var{first-to-check} is the
2086 number of the first argument to check against the format string. For
2087 functions where the arguments are not available to be checked (such as
2088 @code{vprintf}), specify the third parameter as zero. In this case the
2089 compiler only checks the format string for consistency. For
2090 @code{strftime} formats, the third parameter is required to be zero.
2092 In the example above, the format string (@code{my_format}) is the second
2093 argument of the function @code{my_print}, and the arguments to check
2094 start with the third argument, so the correct parameters for the format
2095 attribute are 2 and 3.
2097 @opindex ffreestanding
2098 The @code{format} attribute allows you to identify your own functions
2099 which take format strings as arguments, so that GCC can check the
2100 calls to these functions for errors. The compiler always (unless
2101 @option{-ffreestanding} is used) checks formats
2102 for the standard library functions @code{printf}, @code{fprintf},
2103 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2104 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2105 warnings are requested (using @option{-Wformat}), so there is no need to
2106 modify the header file @file{stdio.h}. In C99 mode, the functions
2107 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2108 @code{vsscanf} are also checked. Except in strictly conforming C
2109 standard modes, the X/Open function @code{strfmon} is also checked as
2110 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2111 @xref{C Dialect Options,,Options Controlling C Dialect}.
2113 @item format_arg (@var{string-index})
2114 @cindex @code{format_arg} function attribute
2115 @opindex Wformat-nonliteral
2116 The @code{format_arg} attribute specifies that a function takes a format
2117 string for a @code{printf}, @code{scanf}, @code{strftime} or
2118 @code{strfmon} style function and modifies it (for example, to translate
2119 it into another language), so the result can be passed to a
2120 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2121 function (with the remaining arguments to the format function the same
2122 as they would have been for the unmodified string). For example, the
2127 my_dgettext (char *my_domain, const char *my_format)
2128 __attribute__ ((format_arg (2)));
2132 causes the compiler to check the arguments in calls to a @code{printf},
2133 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2134 format string argument is a call to the @code{my_dgettext} function, for
2135 consistency with the format string argument @code{my_format}. If the
2136 @code{format_arg} attribute had not been specified, all the compiler
2137 could tell in such calls to format functions would be that the format
2138 string argument is not constant; this would generate a warning when
2139 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2140 without the attribute.
2142 The parameter @var{string-index} specifies which argument is the format
2143 string argument (starting from 1).
2145 The @code{format-arg} attribute allows you to identify your own
2146 functions which modify format strings, so that GCC can check the
2147 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2148 type function whose operands are a call to one of your own function.
2149 The compiler always treats @code{gettext}, @code{dgettext}, and
2150 @code{dcgettext} in this manner except when strict ISO C support is
2151 requested by @option{-ansi} or an appropriate @option{-std} option, or
2152 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
2153 Controlling C Dialect}.
2155 @item nonnull (@var{arg-index}, @dots{})
2156 @cindex @code{nonnull} function attribute
2157 The @code{nonnull} attribute specifies that some function parameters should
2158 be non-null pointers. For instance, the declaration:
2162 my_memcpy (void *dest, const void *src, size_t len)
2163 __attribute__((nonnull (1, 2)));
2167 causes the compiler to check that, in calls to @code{my_memcpy},
2168 arguments @var{dest} and @var{src} are non-null. If the compiler
2169 determines that a null pointer is passed in an argument slot marked
2170 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2171 is issued. The compiler may also choose to make optimizations based
2172 on the knowledge that certain function arguments will not be null.
2174 If no argument index list is given to the @code{nonnull} attribute,
2175 all pointer arguments are marked as non-null. To illustrate, the
2176 following declaration is equivalent to the previous example:
2180 my_memcpy (void *dest, const void *src, size_t len)
2181 __attribute__((nonnull));
2184 @item no_instrument_function
2185 @cindex @code{no_instrument_function} function attribute
2186 @opindex finstrument-functions
2187 If @option{-finstrument-functions} is given, profiling function calls will
2188 be generated at entry and exit of most user-compiled functions.
2189 Functions with this attribute will not be so instrumented.
2191 @item section ("@var{section-name}")
2192 @cindex @code{section} function attribute
2193 Normally, the compiler places the code it generates in the @code{text} section.
2194 Sometimes, however, you need additional sections, or you need certain
2195 particular functions to appear in special sections. The @code{section}
2196 attribute specifies that a function lives in a particular section.
2197 For example, the declaration:
2200 extern void foobar (void) __attribute__ ((section ("bar")));
2204 puts the function @code{foobar} in the @code{bar} section.
2206 Some file formats do not support arbitrary sections so the @code{section}
2207 attribute is not available on all platforms.
2208 If you need to map the entire contents of a module to a particular
2209 section, consider using the facilities of the linker instead.
2213 @cindex @code{constructor} function attribute
2214 @cindex @code{destructor} function attribute
2215 The @code{constructor} attribute causes the function to be called
2216 automatically before execution enters @code{main ()}. Similarly, the
2217 @code{destructor} attribute causes the function to be called
2218 automatically after @code{main ()} has completed or @code{exit ()} has
2219 been called. Functions with these attributes are useful for
2220 initializing data that will be used implicitly during the execution of
2223 These attributes are not currently implemented for Objective-C@.
2225 @cindex @code{unused} attribute.
2227 This attribute, attached to a function, means that the function is meant
2228 to be possibly unused. GCC will not produce a warning for this
2229 function. GNU C++ does not currently support this attribute as
2230 definitions without parameters are valid in C++.
2232 @cindex @code{used} attribute.
2234 This attribute, attached to a function, means that code must be emitted
2235 for the function even if it appears that the function is not referenced.
2236 This is useful, for example, when the function is referenced only in
2239 @cindex @code{deprecated} attribute.
2241 The @code{deprecated} attribute results in a warning if the function
2242 is used anywhere in the source file. This is useful when identifying
2243 functions that are expected to be removed in a future version of a
2244 program. The warning also includes the location of the declaration
2245 of the deprecated function, to enable users to easily find further
2246 information about why the function is deprecated, or what they should
2247 do instead. Note that the warnings only occurs for uses:
2250 int old_fn () __attribute__ ((deprecated));
2252 int (*fn_ptr)() = old_fn;
2255 results in a warning on line 3 but not line 2.
2257 The @code{deprecated} attribute can also be used for variables and
2258 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2261 @cindex @code{weak} attribute
2262 The @code{weak} attribute causes the declaration to be emitted as a weak
2263 symbol rather than a global. This is primarily useful in defining
2264 library functions which can be overridden in user code, though it can
2265 also be used with non-function declarations. Weak symbols are supported
2266 for ELF targets, and also for a.out targets when using the GNU assembler
2270 @cindex @code{malloc} attribute
2271 The @code{malloc} attribute is used to tell the compiler that a function
2272 may be treated as if it were the malloc function. The compiler assumes
2273 that calls to malloc result in a pointers that cannot alias anything.
2274 This will often improve optimization.
2276 @item alias ("@var{target}")
2277 @cindex @code{alias} attribute
2278 The @code{alias} attribute causes the declaration to be emitted as an
2279 alias for another symbol, which must be specified. For instance,
2282 void __f () @{ /* @r{Do something.} */; @}
2283 void f () __attribute__ ((weak, alias ("__f")));
2286 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
2287 mangled name for the target must be used.
2289 Not all target machines support this attribute.
2291 @item visibility ("@var{visibility_type}")
2292 @cindex @code{visibility} attribute
2293 The @code{visibility} attribute on ELF targets causes the declaration
2294 to be emitted with hidden, protected or internal visibility.
2297 void __attribute__ ((visibility ("protected")))
2298 f () @{ /* @r{Do something.} */; @}
2299 int i __attribute__ ((visibility ("hidden")));
2302 See the ELF gABI for complete details, but the short story is
2306 Hidden visibility indicates that the symbol will not be placed into
2307 the dynamic symbol table, so no other @dfn{module} (executable or
2308 shared library) can reference it directly.
2311 Protected visibility indicates that the symbol will be placed in the
2312 dynamic symbol table, but that references within the defining module
2313 will bind to the local symbol. That is, the symbol cannot be overridden
2317 Internal visibility is like hidden visibility, but with additional
2318 processor specific semantics. Unless otherwise specified by the psABI,
2319 gcc defines internal visibility to mean that the function is @emph{never}
2320 called from another module. Note that hidden symbols, while then cannot
2321 be referenced directly by other modules, can be referenced indirectly via
2322 function pointers. By indicating that a symbol cannot be called from
2323 outside the module, gcc may for instance omit the load of a PIC register
2324 since it is known that the calling function loaded the correct value.
2327 Not all ELF targets support this attribute.
2329 @item regparm (@var{number})
2330 @cindex functions that are passed arguments in registers on the 386
2331 On the Intel 386, the @code{regparm} attribute causes the compiler to
2332 pass up to @var{number} integer arguments in registers EAX,
2333 EDX, and ECX instead of on the stack. Functions that take a
2334 variable number of arguments will continue to be passed all of their
2335 arguments on the stack.
2338 @cindex functions that pop the argument stack on the 386
2339 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2340 assume that the called function will pop off the stack space used to
2341 pass arguments, unless it takes a variable number of arguments.
2343 The PowerPC compiler for Windows NT currently ignores the @code{stdcall}
2347 @cindex functions that do pop the argument stack on the 386
2349 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2350 assume that the calling function will pop off the stack space used to
2351 pass arguments. This is
2352 useful to override the effects of the @option{-mrtd} switch.
2354 The PowerPC compiler for Windows NT currently ignores the @code{cdecl}
2357 @item longcall/shortcall
2358 @cindex functions called via pointer on the RS/6000 and PowerPC
2359 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
2360 compiler to always call this function via a pointer, just as it would if
2361 the @option{-mlongcall} option had been specified. The @code{shortcall}
2362 attribute causes the compiler not to do this. These attributes override
2363 both the @option{-mlongcall} switch and the @code{#pragma longcall}
2366 @xref{RS/6000 and PowerPC Options}, for more information on when long
2367 calls are and are not necessary.
2369 @item long_call/short_call
2370 @cindex indirect calls on ARM
2371 This attribute allows to specify how to call a particular function on
2372 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2373 command line switch and @code{#pragma long_calls} settings. The
2374 @code{long_call} attribute causes the compiler to always call the
2375 function by first loading its address into a register and then using the
2376 contents of that register. The @code{short_call} attribute always places
2377 the offset to the function from the call site into the @samp{BL}
2378 instruction directly.
2381 @cindex functions which are imported from a dll on PowerPC Windows NT
2382 On the PowerPC running Windows NT, the @code{dllimport} attribute causes
2383 the compiler to call the function via a global pointer to the function
2384 pointer that is set up by the Windows NT dll library. The pointer name
2385 is formed by combining @code{__imp_} and the function name.
2388 @cindex functions which are exported from a dll on PowerPC Windows NT
2389 On the PowerPC running Windows NT, the @code{dllexport} attribute causes
2390 the compiler to provide a global pointer to the function pointer, so
2391 that it can be called with the @code{dllimport} attribute. The pointer
2392 name is formed by combining @code{__imp_} and the function name.
2394 @item exception (@var{except-func} [, @var{except-arg}])
2395 @cindex functions which specify exception handling on PowerPC Windows NT
2396 On the PowerPC running Windows NT, the @code{exception} attribute causes
2397 the compiler to modify the structured exception table entry it emits for
2398 the declared function. The string or identifier @var{except-func} is
2399 placed in the third entry of the structured exception table. It
2400 represents a function, which is called by the exception handling
2401 mechanism if an exception occurs. If it was specified, the string or
2402 identifier @var{except-arg} is placed in the fourth entry of the
2403 structured exception table.
2405 @item function_vector
2406 @cindex calling functions through the function vector on the H8/300 processors
2407 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2408 function should be called through the function vector. Calling a
2409 function through the function vector will reduce code size, however;
2410 the function vector has a limited size (maximum 128 entries on the H8/300
2411 and 64 entries on the H8/300H) and shares space with the interrupt vector.
2413 You must use GAS and GLD from GNU binutils version 2.7 or later for
2414 this attribute to work correctly.
2417 @cindex interrupt handler functions
2418 Use this attribute on the ARM, AVR, M32R/D and Xstormy16 ports to indicate
2419 that the specified function is an interrupt handler. The compiler will
2420 generate function entry and exit sequences suitable for use in an
2421 interrupt handler when this attribute is present.
2423 Note, interrupt handlers for the H8/300, H8/300H and SH processors can
2424 be specified via the @code{interrupt_handler} attribute.
2426 Note, on the AVR interrupts will be enabled inside the function.
2428 Note, for the ARM you can specify the kind of interrupt to be handled by
2429 adding an optional parameter to the interrupt attribute like this:
2432 void f () __attribute__ ((interrupt ("IRQ")));
2435 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2437 @item interrupt_handler
2438 @cindex interrupt handler functions on the H8/300 and SH processors
2439 Use this attribute on the H8/300, H8/300H and SH to indicate that the
2440 specified function is an interrupt handler. The compiler will generate
2441 function entry and exit sequences suitable for use in an interrupt
2442 handler when this attribute is present.
2445 Use this attribute on the SH to indicate an @code{interrupt_handler}
2446 function should switch to an alternate stack. It expects a string
2447 argument that names a global variable holding the address of the
2452 void f () __attribute__ ((interrupt_handler,
2453 sp_switch ("alt_stack")));
2457 Use this attribute on the SH for an @code{interrupt_handle} to return using
2458 @code{trapa} instead of @code{rte}. This attribute expects an integer
2459 argument specifying the trap number to be used.
2462 @cindex eight bit data on the H8/300 and H8/300H
2463 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2464 variable should be placed into the eight bit data section.
2465 The compiler will generate more efficient code for certain operations
2466 on data in the eight bit data area. Note the eight bit data area is limited to
2469 You must use GAS and GLD from GNU binutils version 2.7 or later for
2470 this attribute to work correctly.
2473 @cindex tiny data section on the H8/300H
2474 Use this attribute on the H8/300H to indicate that the specified
2475 variable should be placed into the tiny data section.
2476 The compiler will generate more efficient code for loads and stores
2477 on data in the tiny data section. Note the tiny data area is limited to
2478 slightly under 32kbytes of data.
2481 @cindex signal handler functions on the AVR processors
2482 Use this attribute on the AVR to indicate that the specified
2483 function is an signal handler. The compiler will generate function
2484 entry and exit sequences suitable for use in an signal handler when this
2485 attribute is present. Interrupts will be disabled inside function.
2488 @cindex function without a prologue/epilogue code
2489 Use this attribute on the ARM, AVR and IP2K ports to indicate that the
2490 specified function do not need prologue/epilogue sequences generated by
2491 the compiler. It is up to the programmer to provide these sequences.
2493 @item model (@var{model-name})
2494 @cindex function addressability on the M32R/D
2495 Use this attribute on the M32R/D to set the addressability of an object,
2496 and the code generated for a function.
2497 The identifier @var{model-name} is one of @code{small}, @code{medium},
2498 or @code{large}, representing each of the code models.
2500 Small model objects live in the lower 16MB of memory (so that their
2501 addresses can be loaded with the @code{ld24} instruction), and are
2502 callable with the @code{bl} instruction.
2504 Medium model objects may live anywhere in the 32-bit address space (the
2505 compiler will generate @code{seth/add3} instructions to load their addresses),
2506 and are callable with the @code{bl} instruction.
2508 Large model objects may live anywhere in the 32-bit address space (the
2509 compiler will generate @code{seth/add3} instructions to load their addresses),
2510 and may not be reachable with the @code{bl} instruction (the compiler will
2511 generate the much slower @code{seth/add3/jl} instruction sequence).
2515 You can specify multiple attributes in a declaration by separating them
2516 by commas within the double parentheses or by immediately following an
2517 attribute declaration with another attribute declaration.
2519 @cindex @code{#pragma}, reason for not using
2520 @cindex pragma, reason for not using
2521 Some people object to the @code{__attribute__} feature, suggesting that
2522 ISO C's @code{#pragma} should be used instead. At the time
2523 @code{__attribute__} was designed, there were two reasons for not doing
2528 It is impossible to generate @code{#pragma} commands from a macro.
2531 There is no telling what the same @code{#pragma} might mean in another
2535 These two reasons applied to almost any application that might have been
2536 proposed for @code{#pragma}. It was basically a mistake to use
2537 @code{#pragma} for @emph{anything}.
2539 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2540 to be generated from macros. In addition, a @code{#pragma GCC}
2541 namespace is now in use for GCC-specific pragmas. However, it has been
2542 found convenient to use @code{__attribute__} to achieve a natural
2543 attachment of attributes to their corresponding declarations, whereas
2544 @code{#pragma GCC} is of use for constructs that do not naturally form
2545 part of the grammar. @xref{Other Directives,,Miscellaneous
2546 Preprocessing Directives, cpp, The C Preprocessor}.
2548 @node Attribute Syntax
2549 @section Attribute Syntax
2550 @cindex attribute syntax
2552 This section describes the syntax with which @code{__attribute__} may be
2553 used, and the constructs to which attribute specifiers bind, for the C
2554 language. Some details may vary for C++ and Objective-C@. Because of
2555 infelicities in the grammar for attributes, some forms described here
2556 may not be successfully parsed in all cases.
2558 There are some problems with the semantics of attributes in C++. For
2559 example, there are no manglings for attributes, although they may affect
2560 code generation, so problems may arise when attributed types are used in
2561 conjunction with templates or overloading. Similarly, @code{typeid}
2562 does not distinguish between types with different attributes. Support
2563 for attributes in C++ may be restricted in future to attributes on
2564 declarations only, but not on nested declarators.
2566 @xref{Function Attributes}, for details of the semantics of attributes
2567 applying to functions. @xref{Variable Attributes}, for details of the
2568 semantics of attributes applying to variables. @xref{Type Attributes},
2569 for details of the semantics of attributes applying to structure, union
2570 and enumerated types.
2572 An @dfn{attribute specifier} is of the form
2573 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2574 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2575 each attribute is one of the following:
2579 Empty. Empty attributes are ignored.
2582 A word (which may be an identifier such as @code{unused}, or a reserved
2583 word such as @code{const}).
2586 A word, followed by, in parentheses, parameters for the attribute.
2587 These parameters take one of the following forms:
2591 An identifier. For example, @code{mode} attributes use this form.
2594 An identifier followed by a comma and a non-empty comma-separated list
2595 of expressions. For example, @code{format} attributes use this form.
2598 A possibly empty comma-separated list of expressions. For example,
2599 @code{format_arg} attributes use this form with the list being a single
2600 integer constant expression, and @code{alias} attributes use this form
2601 with the list being a single string constant.
2605 An @dfn{attribute specifier list} is a sequence of one or more attribute
2606 specifiers, not separated by any other tokens.
2608 An attribute specifier list may appear after the colon following a
2609 label, other than a @code{case} or @code{default} label. The only
2610 attribute it makes sense to use after a label is @code{unused}. This
2611 feature is intended for code generated by programs which contains labels
2612 that may be unused but which is compiled with @option{-Wall}. It would
2613 not normally be appropriate to use in it human-written code, though it
2614 could be useful in cases where the code that jumps to the label is
2615 contained within an @code{#ifdef} conditional.
2617 An attribute specifier list may appear as part of a @code{struct},
2618 @code{union} or @code{enum} specifier. It may go either immediately
2619 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2620 the closing brace. It is ignored if the content of the structure, union
2621 or enumerated type is not defined in the specifier in which the
2622 attribute specifier list is used---that is, in usages such as
2623 @code{struct __attribute__((foo)) bar} with no following opening brace.
2624 Where attribute specifiers follow the closing brace, they are considered
2625 to relate to the structure, union or enumerated type defined, not to any
2626 enclosing declaration the type specifier appears in, and the type
2627 defined is not complete until after the attribute specifiers.
2628 @c Otherwise, there would be the following problems: a shift/reduce
2629 @c conflict between attributes binding the struct/union/enum and
2630 @c binding to the list of specifiers/qualifiers; and "aligned"
2631 @c attributes could use sizeof for the structure, but the size could be
2632 @c changed later by "packed" attributes.
2634 Otherwise, an attribute specifier appears as part of a declaration,
2635 counting declarations of unnamed parameters and type names, and relates
2636 to that declaration (which may be nested in another declaration, for
2637 example in the case of a parameter declaration), or to a particular declarator
2638 within a declaration. Where an
2639 attribute specifier is applied to a parameter declared as a function or
2640 an array, it should apply to the function or array rather than the
2641 pointer to which the parameter is implicitly converted, but this is not
2642 yet correctly implemented.
2644 Any list of specifiers and qualifiers at the start of a declaration may
2645 contain attribute specifiers, whether or not such a list may in that
2646 context contain storage class specifiers. (Some attributes, however,
2647 are essentially in the nature of storage class specifiers, and only make
2648 sense where storage class specifiers may be used; for example,
2649 @code{section}.) There is one necessary limitation to this syntax: the
2650 first old-style parameter declaration in a function definition cannot
2651 begin with an attribute specifier, because such an attribute applies to
2652 the function instead by syntax described below (which, however, is not
2653 yet implemented in this case). In some other cases, attribute
2654 specifiers are permitted by this grammar but not yet supported by the
2655 compiler. All attribute specifiers in this place relate to the
2656 declaration as a whole. In the obsolescent usage where a type of
2657 @code{int} is implied by the absence of type specifiers, such a list of
2658 specifiers and qualifiers may be an attribute specifier list with no
2659 other specifiers or qualifiers.
2661 An attribute specifier list may appear immediately before a declarator
2662 (other than the first) in a comma-separated list of declarators in a
2663 declaration of more than one identifier using a single list of
2664 specifiers and qualifiers. Such attribute specifiers apply
2665 only to the identifier before whose declarator they appear. For
2669 __attribute__((noreturn)) void d0 (void),
2670 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2675 the @code{noreturn} attribute applies to all the functions
2676 declared; the @code{format} attribute only applies to @code{d1}.
2678 An attribute specifier list may appear immediately before the comma,
2679 @code{=} or semicolon terminating the declaration of an identifier other
2680 than a function definition. At present, such attribute specifiers apply
2681 to the declared object or function, but in future they may attach to the
2682 outermost adjacent declarator. In simple cases there is no difference,
2683 but, for example, in
2686 void (****f)(void) __attribute__((noreturn));
2690 at present the @code{noreturn} attribute applies to @code{f}, which
2691 causes a warning since @code{f} is not a function, but in future it may
2692 apply to the function @code{****f}. The precise semantics of what
2693 attributes in such cases will apply to are not yet specified. Where an
2694 assembler name for an object or function is specified (@pxref{Asm
2695 Labels}), at present the attribute must follow the @code{asm}
2696 specification; in future, attributes before the @code{asm} specification
2697 may apply to the adjacent declarator, and those after it to the declared
2700 An attribute specifier list may, in future, be permitted to appear after
2701 the declarator in a function definition (before any old-style parameter
2702 declarations or the function body).
2704 Attribute specifiers may be mixed with type qualifiers appearing inside
2705 the @code{[]} of a parameter array declarator, in the C99 construct by
2706 which such qualifiers are applied to the pointer to which the array is
2707 implicitly converted. Such attribute specifiers apply to the pointer,
2708 not to the array, but at present this is not implemented and they are
2711 An attribute specifier list may appear at the start of a nested
2712 declarator. At present, there are some limitations in this usage: the
2713 attributes correctly apply to the declarator, but for most individual
2714 attributes the semantics this implies are not implemented.
2715 When attribute specifiers follow the @code{*} of a pointer
2716 declarator, they may be mixed with any type qualifiers present.
2717 The following describes the formal semantics of this syntax. It will make the
2718 most sense if you are familiar with the formal specification of
2719 declarators in the ISO C standard.
2721 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2722 D1}, where @code{T} contains declaration specifiers that specify a type
2723 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2724 contains an identifier @var{ident}. The type specified for @var{ident}
2725 for derived declarators whose type does not include an attribute
2726 specifier is as in the ISO C standard.
2728 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2729 and the declaration @code{T D} specifies the type
2730 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2731 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2732 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2734 If @code{D1} has the form @code{*
2735 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2736 declaration @code{T D} specifies the type
2737 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2738 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2739 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2745 void (__attribute__((noreturn)) ****f) (void);
2749 specifies the type ``pointer to pointer to pointer to pointer to
2750 non-returning function returning @code{void}''. As another example,
2753 char *__attribute__((aligned(8))) *f;
2757 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2758 Note again that this does not work with most attributes; for example,
2759 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2760 is not yet supported.
2762 For compatibility with existing code written for compiler versions that
2763 did not implement attributes on nested declarators, some laxity is
2764 allowed in the placing of attributes. If an attribute that only applies
2765 to types is applied to a declaration, it will be treated as applying to
2766 the type of that declaration. If an attribute that only applies to
2767 declarations is applied to the type of a declaration, it will be treated
2768 as applying to that declaration; and, for compatibility with code
2769 placing the attributes immediately before the identifier declared, such
2770 an attribute applied to a function return type will be treated as
2771 applying to the function type, and such an attribute applied to an array
2772 element type will be treated as applying to the array type. If an
2773 attribute that only applies to function types is applied to a
2774 pointer-to-function type, it will be treated as applying to the pointer
2775 target type; if such an attribute is applied to a function return type
2776 that is not a pointer-to-function type, it will be treated as applying
2777 to the function type.
2779 @node Function Prototypes
2780 @section Prototypes and Old-Style Function Definitions
2781 @cindex function prototype declarations
2782 @cindex old-style function definitions
2783 @cindex promotion of formal parameters
2785 GNU C extends ISO C to allow a function prototype to override a later
2786 old-style non-prototype definition. Consider the following example:
2789 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2796 /* @r{Prototype function declaration.} */
2797 int isroot P((uid_t));
2799 /* @r{Old-style function definition.} */
2801 isroot (x) /* ??? lossage here ??? */
2808 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2809 not allow this example, because subword arguments in old-style
2810 non-prototype definitions are promoted. Therefore in this example the
2811 function definition's argument is really an @code{int}, which does not
2812 match the prototype argument type of @code{short}.
2814 This restriction of ISO C makes it hard to write code that is portable
2815 to traditional C compilers, because the programmer does not know
2816 whether the @code{uid_t} type is @code{short}, @code{int}, or
2817 @code{long}. Therefore, in cases like these GNU C allows a prototype
2818 to override a later old-style definition. More precisely, in GNU C, a
2819 function prototype argument type overrides the argument type specified
2820 by a later old-style definition if the former type is the same as the
2821 latter type before promotion. Thus in GNU C the above example is
2822 equivalent to the following:
2835 GNU C++ does not support old-style function definitions, so this
2836 extension is irrelevant.
2839 @section C++ Style Comments
2841 @cindex C++ comments
2842 @cindex comments, C++ style
2844 In GNU C, you may use C++ style comments, which start with @samp{//} and
2845 continue until the end of the line. Many other C implementations allow
2846 such comments, and they are included in the 1999 C standard. However,
2847 C++ style comments are not recognized if you specify an @option{-std}
2848 option specifying a version of ISO C before C99, or @option{-ansi}
2849 (equivalent to @option{-std=c89}).
2852 @section Dollar Signs in Identifier Names
2854 @cindex dollar signs in identifier names
2855 @cindex identifier names, dollar signs in
2857 In GNU C, you may normally use dollar signs in identifier names.
2858 This is because many traditional C implementations allow such identifiers.
2859 However, dollar signs in identifiers are not supported on a few target
2860 machines, typically because the target assembler does not allow them.
2862 @node Character Escapes
2863 @section The Character @key{ESC} in Constants
2865 You can use the sequence @samp{\e} in a string or character constant to
2866 stand for the ASCII character @key{ESC}.
2869 @section Inquiring on Alignment of Types or Variables
2871 @cindex type alignment
2872 @cindex variable alignment
2874 The keyword @code{__alignof__} allows you to inquire about how an object
2875 is aligned, or the minimum alignment usually required by a type. Its
2876 syntax is just like @code{sizeof}.
2878 For example, if the target machine requires a @code{double} value to be
2879 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2880 This is true on many RISC machines. On more traditional machine
2881 designs, @code{__alignof__ (double)} is 4 or even 2.
2883 Some machines never actually require alignment; they allow reference to any
2884 data type even at an odd addresses. For these machines, @code{__alignof__}
2885 reports the @emph{recommended} alignment of a type.
2887 If the operand of @code{__alignof__} is an lvalue rather than a type,
2888 its value is the required alignment for its type, taking into account
2889 any minimum alignment specified with GCC's @code{__attribute__}
2890 extension (@pxref{Variable Attributes}). For example, after this
2894 struct foo @{ int x; char y; @} foo1;
2898 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
2899 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
2901 It is an error to ask for the alignment of an incomplete type.
2903 @node Variable Attributes
2904 @section Specifying Attributes of Variables
2905 @cindex attribute of variables
2906 @cindex variable attributes
2908 The keyword @code{__attribute__} allows you to specify special
2909 attributes of variables or structure fields. This keyword is followed
2910 by an attribute specification inside double parentheses. Ten
2911 attributes are currently defined for variables: @code{aligned},
2912 @code{mode}, @code{nocommon}, @code{packed}, @code{section},
2913 @code{transparent_union}, @code{unused}, @code{deprecated},
2914 @code{vector_size}, and @code{weak}. Some other attributes are defined
2915 for variables on particular target systems. Other attributes are
2916 available for functions (@pxref{Function Attributes}) and for types
2917 (@pxref{Type Attributes}). Other front ends might define more
2918 attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2920 You may also specify attributes with @samp{__} preceding and following
2921 each keyword. This allows you to use them in header files without
2922 being concerned about a possible macro of the same name. For example,
2923 you may use @code{__aligned__} instead of @code{aligned}.
2925 @xref{Attribute Syntax}, for details of the exact syntax for using
2929 @cindex @code{aligned} attribute
2930 @item aligned (@var{alignment})
2931 This attribute specifies a minimum alignment for the variable or
2932 structure field, measured in bytes. For example, the declaration:
2935 int x __attribute__ ((aligned (16))) = 0;
2939 causes the compiler to allocate the global variable @code{x} on a
2940 16-byte boundary. On a 68040, this could be used in conjunction with
2941 an @code{asm} expression to access the @code{move16} instruction which
2942 requires 16-byte aligned operands.
2944 You can also specify the alignment of structure fields. For example, to
2945 create a double-word aligned @code{int} pair, you could write:
2948 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2952 This is an alternative to creating a union with a @code{double} member
2953 that forces the union to be double-word aligned.
2955 As in the preceding examples, you can explicitly specify the alignment
2956 (in bytes) that you wish the compiler to use for a given variable or
2957 structure field. Alternatively, you can leave out the alignment factor
2958 and just ask the compiler to align a variable or field to the maximum
2959 useful alignment for the target machine you are compiling for. For
2960 example, you could write:
2963 short array[3] __attribute__ ((aligned));
2966 Whenever you leave out the alignment factor in an @code{aligned} attribute
2967 specification, the compiler automatically sets the alignment for the declared
2968 variable or field to the largest alignment which is ever used for any data
2969 type on the target machine you are compiling for. Doing this can often make
2970 copy operations more efficient, because the compiler can use whatever
2971 instructions copy the biggest chunks of memory when performing copies to
2972 or from the variables or fields that you have aligned this way.
2974 The @code{aligned} attribute can only increase the alignment; but you
2975 can decrease it by specifying @code{packed} as well. See below.
2977 Note that the effectiveness of @code{aligned} attributes may be limited
2978 by inherent limitations in your linker. On many systems, the linker is
2979 only able to arrange for variables to be aligned up to a certain maximum
2980 alignment. (For some linkers, the maximum supported alignment may
2981 be very very small.) If your linker is only able to align variables
2982 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2983 in an @code{__attribute__} will still only provide you with 8 byte
2984 alignment. See your linker documentation for further information.
2986 @item mode (@var{mode})
2987 @cindex @code{mode} attribute
2988 This attribute specifies the data type for the declaration---whichever
2989 type corresponds to the mode @var{mode}. This in effect lets you
2990 request an integer or floating point type according to its width.
2992 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2993 indicate the mode corresponding to a one-byte integer, @samp{word} or
2994 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2995 or @samp{__pointer__} for the mode used to represent pointers.
2998 @cindex @code{nocommon} attribute
3000 This attribute specifies requests GCC not to place a variable
3001 ``common'' but instead to allocate space for it directly. If you
3002 specify the @option{-fno-common} flag, GCC will do this for all
3005 Specifying the @code{nocommon} attribute for a variable provides an
3006 initialization of zeros. A variable may only be initialized in one
3010 @cindex @code{packed} attribute
3011 The @code{packed} attribute specifies that a variable or structure field
3012 should have the smallest possible alignment---one byte for a variable,
3013 and one bit for a field, unless you specify a larger value with the
3014 @code{aligned} attribute.
3016 Here is a structure in which the field @code{x} is packed, so that it
3017 immediately follows @code{a}:
3023 int x[2] __attribute__ ((packed));
3027 @item section ("@var{section-name}")
3028 @cindex @code{section} variable attribute
3029 Normally, the compiler places the objects it generates in sections like
3030 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
3031 or you need certain particular variables to appear in special sections,
3032 for example to map to special hardware. The @code{section}
3033 attribute specifies that a variable (or function) lives in a particular
3034 section. For example, this small program uses several specific section names:
3037 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
3038 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
3039 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
3040 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3044 /* Initialize stack pointer */
3045 init_sp (stack + sizeof (stack));
3047 /* Initialize initialized data */
3048 memcpy (&init_data, &data, &edata - &data);
3050 /* Turn on the serial ports */
3057 Use the @code{section} attribute with an @emph{initialized} definition
3058 of a @emph{global} variable, as shown in the example. GCC issues
3059 a warning and otherwise ignores the @code{section} attribute in
3060 uninitialized variable declarations.
3062 You may only use the @code{section} attribute with a fully initialized
3063 global definition because of the way linkers work. The linker requires
3064 each object be defined once, with the exception that uninitialized
3065 variables tentatively go in the @code{common} (or @code{bss}) section
3066 and can be multiply ``defined''. You can force a variable to be
3067 initialized with the @option{-fno-common} flag or the @code{nocommon}
3070 Some file formats do not support arbitrary sections so the @code{section}
3071 attribute is not available on all platforms.
3072 If you need to map the entire contents of a module to a particular
3073 section, consider using the facilities of the linker instead.
3076 @cindex @code{shared} variable attribute
3077 On Windows NT, in addition to putting variable definitions in a named
3078 section, the section can also be shared among all running copies of an
3079 executable or DLL@. For example, this small program defines shared data
3080 by putting it in a named section @code{shared} and marking the section
3084 int foo __attribute__((section ("shared"), shared)) = 0;
3089 /* Read and write foo. All running
3090 copies see the same value. */
3096 You may only use the @code{shared} attribute along with @code{section}
3097 attribute with a fully initialized global definition because of the way
3098 linkers work. See @code{section} attribute for more information.
3100 The @code{shared} attribute is only available on Windows NT@.
3102 @item transparent_union
3103 This attribute, attached to a function parameter which is a union, means
3104 that the corresponding argument may have the type of any union member,
3105 but the argument is passed as if its type were that of the first union
3106 member. For more details see @xref{Type Attributes}. You can also use
3107 this attribute on a @code{typedef} for a union data type; then it
3108 applies to all function parameters with that type.
3111 This attribute, attached to a variable, means that the variable is meant
3112 to be possibly unused. GCC will not produce a warning for this
3116 The @code{deprecated} attribute results in a warning if the variable
3117 is used anywhere in the source file. This is useful when identifying
3118 variables that are expected to be removed in a future version of a
3119 program. The warning also includes the location of the declaration
3120 of the deprecated variable, to enable users to easily find further
3121 information about why the variable is deprecated, or what they should
3122 do instead. Note that the warnings only occurs for uses:
3125 extern int old_var __attribute__ ((deprecated));
3127 int new_fn () @{ return old_var; @}
3130 results in a warning on line 3 but not line 2.
3132 The @code{deprecated} attribute can also be used for functions and
3133 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
3135 @item vector_size (@var{bytes})
3136 This attribute specifies the vector size for the variable, measured in
3137 bytes. For example, the declaration:
3140 int foo __attribute__ ((vector_size (16)));
3144 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3145 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3146 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3148 This attribute is only applicable to integral and float scalars,
3149 although arrays, pointers, and function return values are allowed in
3150 conjunction with this construct.
3152 Aggregates with this attribute are invalid, even if they are of the same
3153 size as a corresponding scalar. For example, the declaration:
3156 struct S @{ int a; @};
3157 struct S __attribute__ ((vector_size (16))) foo;
3161 is invalid even if the size of the structure is the same as the size of
3165 The @code{weak} attribute is described in @xref{Function Attributes}.
3167 @item model (@var{model-name})
3168 @cindex variable addressability on the M32R/D
3169 Use this attribute on the M32R/D to set the addressability of an object.
3170 The identifier @var{model-name} is one of @code{small}, @code{medium},
3171 or @code{large}, representing each of the code models.
3173 Small model objects live in the lower 16MB of memory (so that their
3174 addresses can be loaded with the @code{ld24} instruction).
3176 Medium and large model objects may live anywhere in the 32-bit address space
3177 (the compiler will generate @code{seth/add3} instructions to load their
3182 To specify multiple attributes, separate them by commas within the
3183 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3186 @node Type Attributes
3187 @section Specifying Attributes of Types
3188 @cindex attribute of types
3189 @cindex type attributes
3191 The keyword @code{__attribute__} allows you to specify special
3192 attributes of @code{struct} and @code{union} types when you define such
3193 types. This keyword is followed by an attribute specification inside
3194 double parentheses. Six attributes are currently defined for types:
3195 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
3196 @code{deprecated} and @code{may_alias}. Other attributes are defined for
3197 functions (@pxref{Function Attributes}) and for variables
3198 (@pxref{Variable Attributes}).
3200 You may also specify any one of these attributes with @samp{__}
3201 preceding and following its keyword. This allows you to use these
3202 attributes in header files without being concerned about a possible
3203 macro of the same name. For example, you may use @code{__aligned__}
3204 instead of @code{aligned}.
3206 You may specify the @code{aligned} and @code{transparent_union}
3207 attributes either in a @code{typedef} declaration or just past the
3208 closing curly brace of a complete enum, struct or union type
3209 @emph{definition} and the @code{packed} attribute only past the closing
3210 brace of a definition.
3212 You may also specify attributes between the enum, struct or union
3213 tag and the name of the type rather than after the closing brace.
3215 @xref{Attribute Syntax}, for details of the exact syntax for using
3219 @cindex @code{aligned} attribute
3220 @item aligned (@var{alignment})
3221 This attribute specifies a minimum alignment (in bytes) for variables
3222 of the specified type. For example, the declarations:
3225 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3226 typedef int more_aligned_int __attribute__ ((aligned (8)));
3230 force the compiler to insure (as far as it can) that each variable whose
3231 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3232 aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all
3233 variables of type @code{struct S} aligned to 8-byte boundaries allows
3234 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3235 store) instructions when copying one variable of type @code{struct S} to
3236 another, thus improving run-time efficiency.
3238 Note that the alignment of any given @code{struct} or @code{union} type
3239 is required by the ISO C standard to be at least a perfect multiple of
3240 the lowest common multiple of the alignments of all of the members of
3241 the @code{struct} or @code{union} in question. This means that you @emph{can}
3242 effectively adjust the alignment of a @code{struct} or @code{union}
3243 type by attaching an @code{aligned} attribute to any one of the members
3244 of such a type, but the notation illustrated in the example above is a
3245 more obvious, intuitive, and readable way to request the compiler to
3246 adjust the alignment of an entire @code{struct} or @code{union} type.
3248 As in the preceding example, you can explicitly specify the alignment
3249 (in bytes) that you wish the compiler to use for a given @code{struct}
3250 or @code{union} type. Alternatively, you can leave out the alignment factor
3251 and just ask the compiler to align a type to the maximum
3252 useful alignment for the target machine you are compiling for. For
3253 example, you could write:
3256 struct S @{ short f[3]; @} __attribute__ ((aligned));
3259 Whenever you leave out the alignment factor in an @code{aligned}
3260 attribute specification, the compiler automatically sets the alignment
3261 for the type to the largest alignment which is ever used for any data
3262 type on the target machine you are compiling for. Doing this can often
3263 make copy operations more efficient, because the compiler can use
3264 whatever instructions copy the biggest chunks of memory when performing
3265 copies to or from the variables which have types that you have aligned
3268 In the example above, if the size of each @code{short} is 2 bytes, then
3269 the size of the entire @code{struct S} type is 6 bytes. The smallest
3270 power of two which is greater than or equal to that is 8, so the
3271 compiler sets the alignment for the entire @code{struct S} type to 8
3274 Note that although you can ask the compiler to select a time-efficient
3275 alignment for a given type and then declare only individual stand-alone
3276 objects of that type, the compiler's ability to select a time-efficient
3277 alignment is primarily useful only when you plan to create arrays of
3278 variables having the relevant (efficiently aligned) type. If you
3279 declare or use arrays of variables of an efficiently-aligned type, then
3280 it is likely that your program will also be doing pointer arithmetic (or
3281 subscripting, which amounts to the same thing) on pointers to the
3282 relevant type, and the code that the compiler generates for these
3283 pointer arithmetic operations will often be more efficient for
3284 efficiently-aligned types than for other types.
3286 The @code{aligned} attribute can only increase the alignment; but you
3287 can decrease it by specifying @code{packed} as well. See below.
3289 Note that the effectiveness of @code{aligned} attributes may be limited
3290 by inherent limitations in your linker. On many systems, the linker is
3291 only able to arrange for variables to be aligned up to a certain maximum
3292 alignment. (For some linkers, the maximum supported alignment may
3293 be very very small.) If your linker is only able to align variables
3294 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3295 in an @code{__attribute__} will still only provide you with 8 byte
3296 alignment. See your linker documentation for further information.
3299 This attribute, attached to an @code{enum}, @code{struct}, or
3300 @code{union} type definition, specified that the minimum required memory
3301 be used to represent the type.
3303 @opindex fshort-enums
3304 Specifying this attribute for @code{struct} and @code{union} types is
3305 equivalent to specifying the @code{packed} attribute on each of the
3306 structure or union members. Specifying the @option{-fshort-enums}
3307 flag on the line is equivalent to specifying the @code{packed}
3308 attribute on all @code{enum} definitions.
3310 You may only specify this attribute after a closing curly brace on an
3311 @code{enum} definition, not in a @code{typedef} declaration, unless that
3312 declaration also contains the definition of the @code{enum}.
3314 @item transparent_union
3315 This attribute, attached to a @code{union} type definition, indicates
3316 that any function parameter having that union type causes calls to that
3317 function to be treated in a special way.
3319 First, the argument corresponding to a transparent union type can be of
3320 any type in the union; no cast is required. Also, if the union contains
3321 a pointer type, the corresponding argument can be a null pointer
3322 constant or a void pointer expression; and if the union contains a void
3323 pointer type, the corresponding argument can be any pointer expression.
3324 If the union member type is a pointer, qualifiers like @code{const} on
3325 the referenced type must be respected, just as with normal pointer
3328 Second, the argument is passed to the function using the calling
3329 conventions of first member of the transparent union, not the calling
3330 conventions of the union itself. All members of the union must have the
3331 same machine representation; this is necessary for this argument passing
3334 Transparent unions are designed for library functions that have multiple
3335 interfaces for compatibility reasons. For example, suppose the
3336 @code{wait} function must accept either a value of type @code{int *} to
3337 comply with Posix, or a value of type @code{union wait *} to comply with
3338 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3339 @code{wait} would accept both kinds of arguments, but it would also
3340 accept any other pointer type and this would make argument type checking
3341 less useful. Instead, @code{<sys/wait.h>} might define the interface
3349 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3351 pid_t wait (wait_status_ptr_t);
3354 This interface allows either @code{int *} or @code{union wait *}
3355 arguments to be passed, using the @code{int *} calling convention.
3356 The program can call @code{wait} with arguments of either type:
3359 int w1 () @{ int w; return wait (&w); @}
3360 int w2 () @{ union wait w; return wait (&w); @}
3363 With this interface, @code{wait}'s implementation might look like this:
3366 pid_t wait (wait_status_ptr_t p)
3368 return waitpid (-1, p.__ip, 0);
3373 When attached to a type (including a @code{union} or a @code{struct}),
3374 this attribute means that variables of that type are meant to appear
3375 possibly unused. GCC will not produce a warning for any variables of
3376 that type, even if the variable appears to do nothing. This is often
3377 the case with lock or thread classes, which are usually defined and then
3378 not referenced, but contain constructors and destructors that have
3379 nontrivial bookkeeping functions.
3382 The @code{deprecated} attribute results in a warning if the type
3383 is used anywhere in the source file. This is useful when identifying
3384 types that are expected to be removed in a future version of a program.
3385 If possible, the warning also includes the location of the declaration
3386 of the deprecated type, to enable users to easily find further
3387 information about why the type is deprecated, or what they should do
3388 instead. Note that the warnings only occur for uses and then only
3389 if the type is being applied to an identifier that itself is not being
3390 declared as deprecated.
3393 typedef int T1 __attribute__ ((deprecated));
3397 typedef T1 T3 __attribute__ ((deprecated));
3398 T3 z __attribute__ ((deprecated));
3401 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3402 warning is issued for line 4 because T2 is not explicitly
3403 deprecated. Line 5 has no warning because T3 is explicitly
3404 deprecated. Similarly for line 6.
3406 The @code{deprecated} attribute can also be used for functions and
3407 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3410 Accesses to objects with types with this attribute are not subjected to
3411 type-based alias analysis, but are instead assumed to be able to alias
3412 any other type of objects, just like the @code{char} type. See
3413 @option{-fstrict-aliasing} for more information on aliasing issues.
3418 typedef short __attribute__((__may_alias__)) short_a;
3424 short_a *b = (short_a *) &a;
3428 if (a == 0x12345678)
3435 If you replaced @code{short_a} with @code{short} in the variable
3436 declaration, the above program would abort when compiled with
3437 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3438 above in recent GCC versions.
3441 To specify multiple attributes, separate them by commas within the
3442 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3446 @section An Inline Function is As Fast As a Macro
3447 @cindex inline functions
3448 @cindex integrating function code
3450 @cindex macros, inline alternative
3452 By declaring a function @code{inline}, you can direct GCC to
3453 integrate that function's code into the code for its callers. This
3454 makes execution faster by eliminating the function-call overhead; in
3455 addition, if any of the actual argument values are constant, their known
3456 values may permit simplifications at compile time so that not all of the
3457 inline function's code needs to be included. The effect on code size is
3458 less predictable; object code may be larger or smaller with function
3459 inlining, depending on the particular case. Inlining of functions is an
3460 optimization and it really ``works'' only in optimizing compilation. If
3461 you don't use @option{-O}, no function is really inline.
3463 Inline functions are included in the ISO C99 standard, but there are
3464 currently substantial differences between what GCC implements and what
3465 the ISO C99 standard requires.
3467 To declare a function inline, use the @code{inline} keyword in its
3468 declaration, like this:
3478 (If you are writing a header file to be included in ISO C programs, write
3479 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3480 You can also make all ``simple enough'' functions inline with the option
3481 @option{-finline-functions}.
3484 Note that certain usages in a function definition can make it unsuitable
3485 for inline substitution. Among these usages are: use of varargs, use of
3486 alloca, use of variable sized data types (@pxref{Variable Length}),
3487 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3488 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3489 will warn when a function marked @code{inline} could not be substituted,
3490 and will give the reason for the failure.
3492 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3493 does not affect the linkage of the function.
3495 @cindex automatic @code{inline} for C++ member fns
3496 @cindex @code{inline} automatic for C++ member fns
3497 @cindex member fns, automatically @code{inline}
3498 @cindex C++ member fns, automatically @code{inline}
3499 @opindex fno-default-inline
3500 GCC automatically inlines member functions defined within the class
3501 body of C++ programs even if they are not explicitly declared
3502 @code{inline}. (You can override this with @option{-fno-default-inline};
3503 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3505 @cindex inline functions, omission of
3506 @opindex fkeep-inline-functions
3507 When a function is both inline and @code{static}, if all calls to the
3508 function are integrated into the caller, and the function's address is
3509 never used, then the function's own assembler code is never referenced.
3510 In this case, GCC does not actually output assembler code for the
3511 function, unless you specify the option @option{-fkeep-inline-functions}.
3512 Some calls cannot be integrated for various reasons (in particular,
3513 calls that precede the function's definition cannot be integrated, and
3514 neither can recursive calls within the definition). If there is a
3515 nonintegrated call, then the function is compiled to assembler code as
3516 usual. The function must also be compiled as usual if the program
3517 refers to its address, because that can't be inlined.
3519 @cindex non-static inline function
3520 When an inline function is not @code{static}, then the compiler must assume
3521 that there may be calls from other source files; since a global symbol can
3522 be defined only once in any program, the function must not be defined in
3523 the other source files, so the calls therein cannot be integrated.
3524 Therefore, a non-@code{static} inline function is always compiled on its
3525 own in the usual fashion.
3527 If you specify both @code{inline} and @code{extern} in the function
3528 definition, then the definition is used only for inlining. In no case
3529 is the function compiled on its own, not even if you refer to its
3530 address explicitly. Such an address becomes an external reference, as
3531 if you had only declared the function, and had not defined it.
3533 This combination of @code{inline} and @code{extern} has almost the
3534 effect of a macro. The way to use it is to put a function definition in
3535 a header file with these keywords, and put another copy of the
3536 definition (lacking @code{inline} and @code{extern}) in a library file.
3537 The definition in the header file will cause most calls to the function
3538 to be inlined. If any uses of the function remain, they will refer to
3539 the single copy in the library.
3541 For future compatibility with when GCC implements ISO C99 semantics for
3542 inline functions, it is best to use @code{static inline} only. (The
3543 existing semantics will remain available when @option{-std=gnu89} is
3544 specified, but eventually the default will be @option{-std=gnu99} and
3545 that will implement the C99 semantics, though it does not do so yet.)
3547 GCC does not inline any functions when not optimizing unless you specify
3548 the @samp{always_inline} attribute for the function, like this:
3552 inline void foo (const char) __attribute__((always_inline));
3556 @section Assembler Instructions with C Expression Operands
3557 @cindex extended @code{asm}
3558 @cindex @code{asm} expressions
3559 @cindex assembler instructions
3562 In an assembler instruction using @code{asm}, you can specify the
3563 operands of the instruction using C expressions. This means you need not
3564 guess which registers or memory locations will contain the data you want
3567 You must specify an assembler instruction template much like what
3568 appears in a machine description, plus an operand constraint string for
3571 For example, here is how to use the 68881's @code{fsinx} instruction:
3574 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3578 Here @code{angle} is the C expression for the input operand while
3579 @code{result} is that of the output operand. Each has @samp{"f"} as its
3580 operand constraint, saying that a floating point register is required.
3581 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3582 output operands' constraints must use @samp{=}. The constraints use the
3583 same language used in the machine description (@pxref{Constraints}).
3585 Each operand is described by an operand-constraint string followed by
3586 the C expression in parentheses. A colon separates the assembler
3587 template from the first output operand and another separates the last
3588 output operand from the first input, if any. Commas separate the
3589 operands within each group. The total number of operands is currently
3590 limited to 30; this limitation may be lifted in some future version of
3593 If there are no output operands but there are input operands, you must
3594 place two consecutive colons surrounding the place where the output
3597 As of GCC version 3.1, it is also possible to specify input and output
3598 operands using symbolic names which can be referenced within the
3599 assembler code. These names are specified inside square brackets
3600 preceding the constraint string, and can be referenced inside the
3601 assembler code using @code{%[@var{name}]} instead of a percentage sign
3602 followed by the operand number. Using named operands the above example
3606 asm ("fsinx %[angle],%[output]"
3607 : [output] "=f" (result)
3608 : [angle] "f" (angle));
3612 Note that the symbolic operand names have no relation whatsoever to
3613 other C identifiers. You may use any name you like, even those of
3614 existing C symbols, but must ensure that no two operands within the same
3615 assembler construct use the same symbolic name.
3617 Output operand expressions must be lvalues; the compiler can check this.
3618 The input operands need not be lvalues. The compiler cannot check
3619 whether the operands have data types that are reasonable for the
3620 instruction being executed. It does not parse the assembler instruction
3621 template and does not know what it means or even whether it is valid
3622 assembler input. The extended @code{asm} feature is most often used for
3623 machine instructions the compiler itself does not know exist. If
3624 the output expression cannot be directly addressed (for example, it is a
3625 bit-field), your constraint must allow a register. In that case, GCC
3626 will use the register as the output of the @code{asm}, and then store
3627 that register into the output.
3629 The ordinary output operands must be write-only; GCC will assume that
3630 the values in these operands before the instruction are dead and need
3631 not be generated. Extended asm supports input-output or read-write
3632 operands. Use the constraint character @samp{+} to indicate such an
3633 operand and list it with the output operands.
3635 When the constraints for the read-write operand (or the operand in which
3636 only some of the bits are to be changed) allows a register, you may, as
3637 an alternative, logically split its function into two separate operands,
3638 one input operand and one write-only output operand. The connection
3639 between them is expressed by constraints which say they need to be in
3640 the same location when the instruction executes. You can use the same C
3641 expression for both operands, or different expressions. For example,
3642 here we write the (fictitious) @samp{combine} instruction with
3643 @code{bar} as its read-only source operand and @code{foo} as its
3644 read-write destination:
3647 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3651 The constraint @samp{"0"} for operand 1 says that it must occupy the
3652 same location as operand 0. A number in constraint is allowed only in
3653 an input operand and it must refer to an output operand.
3655 Only a number in the constraint can guarantee that one operand will be in
3656 the same place as another. The mere fact that @code{foo} is the value
3657 of both operands is not enough to guarantee that they will be in the
3658 same place in the generated assembler code. The following would not
3662 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3665 Various optimizations or reloading could cause operands 0 and 1 to be in
3666 different registers; GCC knows no reason not to do so. For example, the
3667 compiler might find a copy of the value of @code{foo} in one register and
3668 use it for operand 1, but generate the output operand 0 in a different
3669 register (copying it afterward to @code{foo}'s own address). Of course,
3670 since the register for operand 1 is not even mentioned in the assembler
3671 code, the result will not work, but GCC can't tell that.
3673 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3674 the operand number for a matching constraint. For example:
3677 asm ("cmoveq %1,%2,%[result]"
3678 : [result] "=r"(result)
3679 : "r" (test), "r"(new), "[result]"(old));
3682 Some instructions clobber specific hard registers. To describe this,
3683 write a third colon after the input operands, followed by the names of
3684 the clobbered hard registers (given as strings). Here is a realistic
3685 example for the VAX:
3688 asm volatile ("movc3 %0,%1,%2"
3690 : "g" (from), "g" (to), "g" (count)
3691 : "r0", "r1", "r2", "r3", "r4", "r5");
3694 You may not write a clobber description in a way that overlaps with an
3695 input or output operand. For example, you may not have an operand
3696 describing a register class with one member if you mention that register
3697 in the clobber list. There is no way for you to specify that an input
3698 operand is modified without also specifying it as an output
3699 operand. Note that if all the output operands you specify are for this
3700 purpose (and hence unused), you will then also need to specify
3701 @code{volatile} for the @code{asm} construct, as described below, to
3702 prevent GCC from deleting the @code{asm} statement as unused.
3704 If you refer to a particular hardware register from the assembler code,
3705 you will probably have to list the register after the third colon to
3706 tell the compiler the register's value is modified. In some assemblers,
3707 the register names begin with @samp{%}; to produce one @samp{%} in the
3708 assembler code, you must write @samp{%%} in the input.
3710 If your assembler instruction can alter the condition code register, add
3711 @samp{cc} to the list of clobbered registers. GCC on some machines
3712 represents the condition codes as a specific hardware register;
3713 @samp{cc} serves to name this register. On other machines, the
3714 condition code is handled differently, and specifying @samp{cc} has no
3715 effect. But it is valid no matter what the machine.
3717 If your assembler instruction modifies memory in an unpredictable
3718 fashion, add @samp{memory} to the list of clobbered registers. This
3719 will cause GCC to not keep memory values cached in registers across
3720 the assembler instruction. You will also want to add the
3721 @code{volatile} keyword if the memory affected is not listed in the
3722 inputs or outputs of the @code{asm}, as the @samp{memory} clobber does
3723 not count as a side-effect of the @code{asm}.
3725 You can put multiple assembler instructions together in a single
3726 @code{asm} template, separated by the characters normally used in assembly
3727 code for the system. A combination that works in most places is a newline
3728 to break the line, plus a tab character to move to the instruction field
3729 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3730 assembler allows semicolons as a line-breaking character. Note that some
3731 assembler dialects use semicolons to start a comment.
3732 The input operands are guaranteed not to use any of the clobbered
3733 registers, and neither will the output operands' addresses, so you can
3734 read and write the clobbered registers as many times as you like. Here
3735 is an example of multiple instructions in a template; it assumes the
3736 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3739 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3741 : "g" (from), "g" (to)
3745 Unless an output operand has the @samp{&} constraint modifier, GCC
3746 may allocate it in the same register as an unrelated input operand, on
3747 the assumption the inputs are consumed before the outputs are produced.
3748 This assumption may be false if the assembler code actually consists of
3749 more than one instruction. In such a case, use @samp{&} for each output
3750 operand that may not overlap an input. @xref{Modifiers}.
3752 If you want to test the condition code produced by an assembler
3753 instruction, you must include a branch and a label in the @code{asm}
3754 construct, as follows:
3757 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3763 This assumes your assembler supports local labels, as the GNU assembler
3764 and most Unix assemblers do.
3766 Speaking of labels, jumps from one @code{asm} to another are not
3767 supported. The compiler's optimizers do not know about these jumps, and
3768 therefore they cannot take account of them when deciding how to
3771 @cindex macros containing @code{asm}
3772 Usually the most convenient way to use these @code{asm} instructions is to
3773 encapsulate them in macros that look like functions. For example,
3777 (@{ double __value, __arg = (x); \
3778 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3783 Here the variable @code{__arg} is used to make sure that the instruction
3784 operates on a proper @code{double} value, and to accept only those
3785 arguments @code{x} which can convert automatically to a @code{double}.
3787 Another way to make sure the instruction operates on the correct data
3788 type is to use a cast in the @code{asm}. This is different from using a
3789 variable @code{__arg} in that it converts more different types. For
3790 example, if the desired type were @code{int}, casting the argument to
3791 @code{int} would accept a pointer with no complaint, while assigning the
3792 argument to an @code{int} variable named @code{__arg} would warn about
3793 using a pointer unless the caller explicitly casts it.
3795 If an @code{asm} has output operands, GCC assumes for optimization
3796 purposes the instruction has no side effects except to change the output
3797 operands. This does not mean instructions with a side effect cannot be
3798 used, but you must be careful, because the compiler may eliminate them
3799 if the output operands aren't used, or move them out of loops, or
3800 replace two with one if they constitute a common subexpression. Also,
3801 if your instruction does have a side effect on a variable that otherwise
3802 appears not to change, the old value of the variable may be reused later
3803 if it happens to be found in a register.
3805 You can prevent an @code{asm} instruction from being deleted, moved
3806 significantly, or combined, by writing the keyword @code{volatile} after
3807 the @code{asm}. For example:
3810 #define get_and_set_priority(new) \
3812 asm volatile ("get_and_set_priority %0, %1" \
3813 : "=g" (__old) : "g" (new)); \
3818 If you write an @code{asm} instruction with no outputs, GCC will know
3819 the instruction has side-effects and will not delete the instruction or
3820 move it outside of loops.
3822 The @code{volatile} keyword indicates that the instruction has
3823 important side-effects. GCC will not delete a volatile @code{asm} if
3824 it is reachable. (The instruction can still be deleted if GCC can
3825 prove that control-flow will never reach the location of the
3826 instruction.) In addition, GCC will not reschedule instructions
3827 across a volatile @code{asm} instruction. For example:
3830 *(volatile int *)addr = foo;
3831 asm volatile ("eieio" : : );
3835 Assume @code{addr} contains the address of a memory mapped device
3836 register. The PowerPC @code{eieio} instruction (Enforce In-order
3837 Execution of I/O) tells the CPU to make sure that the store to that
3838 device register happens before it issues any other I/O@.
3840 Note that even a volatile @code{asm} instruction can be moved in ways
3841 that appear insignificant to the compiler, such as across jump
3842 instructions. You can't expect a sequence of volatile @code{asm}
3843 instructions to remain perfectly consecutive. If you want consecutive
3844 output, use a single @code{asm}. Also, GCC will perform some
3845 optimizations across a volatile @code{asm} instruction; GCC does not
3846 ``forget everything'' when it encounters a volatile @code{asm}
3847 instruction the way some other compilers do.
3849 An @code{asm} instruction without any operands or clobbers (an ``old
3850 style'' @code{asm}) will be treated identically to a volatile
3851 @code{asm} instruction.
3853 It is a natural idea to look for a way to give access to the condition
3854 code left by the assembler instruction. However, when we attempted to
3855 implement this, we found no way to make it work reliably. The problem
3856 is that output operands might need reloading, which would result in
3857 additional following ``store'' instructions. On most machines, these
3858 instructions would alter the condition code before there was time to
3859 test it. This problem doesn't arise for ordinary ``test'' and
3860 ``compare'' instructions because they don't have any output operands.
3862 For reasons similar to those described above, it is not possible to give
3863 an assembler instruction access to the condition code left by previous
3866 If you are writing a header file that should be includable in ISO C
3867 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3870 @subsection i386 floating point asm operands
3872 There are several rules on the usage of stack-like regs in
3873 asm_operands insns. These rules apply only to the operands that are
3878 Given a set of input regs that die in an asm_operands, it is
3879 necessary to know which are implicitly popped by the asm, and
3880 which must be explicitly popped by gcc.
3882 An input reg that is implicitly popped by the asm must be
3883 explicitly clobbered, unless it is constrained to match an
3887 For any input reg that is implicitly popped by an asm, it is
3888 necessary to know how to adjust the stack to compensate for the pop.
3889 If any non-popped input is closer to the top of the reg-stack than
3890 the implicitly popped reg, it would not be possible to know what the
3891 stack looked like---it's not clear how the rest of the stack ``slides
3894 All implicitly popped input regs must be closer to the top of
3895 the reg-stack than any input that is not implicitly popped.
3897 It is possible that if an input dies in an insn, reload might
3898 use the input reg for an output reload. Consider this example:
3901 asm ("foo" : "=t" (a) : "f" (b));
3904 This asm says that input B is not popped by the asm, and that
3905 the asm pushes a result onto the reg-stack, i.e., the stack is one
3906 deeper after the asm than it was before. But, it is possible that
3907 reload will think that it can use the same reg for both the input and
3908 the output, if input B dies in this insn.
3910 If any input operand uses the @code{f} constraint, all output reg
3911 constraints must use the @code{&} earlyclobber.
3913 The asm above would be written as
3916 asm ("foo" : "=&t" (a) : "f" (b));
3920 Some operands need to be in particular places on the stack. All
3921 output operands fall in this category---there is no other way to
3922 know which regs the outputs appear in unless the user indicates
3923 this in the constraints.
3925 Output operands must specifically indicate which reg an output
3926 appears in after an asm. @code{=f} is not allowed: the operand
3927 constraints must select a class with a single reg.
3930 Output operands may not be ``inserted'' between existing stack regs.
3931 Since no 387 opcode uses a read/write operand, all output operands
3932 are dead before the asm_operands, and are pushed by the asm_operands.
3933 It makes no sense to push anywhere but the top of the reg-stack.
3935 Output operands must start at the top of the reg-stack: output
3936 operands may not ``skip'' a reg.
3939 Some asm statements may need extra stack space for internal
3940 calculations. This can be guaranteed by clobbering stack registers
3941 unrelated to the inputs and outputs.
3945 Here are a couple of reasonable asms to want to write. This asm
3946 takes one input, which is internally popped, and produces two outputs.
3949 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
3952 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
3953 and replaces them with one output. The user must code the @code{st(1)}
3954 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
3957 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
3963 @section Controlling Names Used in Assembler Code
3964 @cindex assembler names for identifiers
3965 @cindex names used in assembler code
3966 @cindex identifiers, names in assembler code
3968 You can specify the name to be used in the assembler code for a C
3969 function or variable by writing the @code{asm} (or @code{__asm__})
3970 keyword after the declarator as follows:
3973 int foo asm ("myfoo") = 2;
3977 This specifies that the name to be used for the variable @code{foo} in
3978 the assembler code should be @samp{myfoo} rather than the usual
3981 On systems where an underscore is normally prepended to the name of a C
3982 function or variable, this feature allows you to define names for the
3983 linker that do not start with an underscore.
3985 It does not make sense to use this feature with a non-static local
3986 variable since such variables do not have assembler names. If you are
3987 trying to put the variable in a particular register, see @ref{Explicit
3988 Reg Vars}. GCC presently accepts such code with a warning, but will
3989 probably be changed to issue an error, rather than a warning, in the
3992 You cannot use @code{asm} in this way in a function @emph{definition}; but
3993 you can get the same effect by writing a declaration for the function
3994 before its definition and putting @code{asm} there, like this:
3997 extern func () asm ("FUNC");
4004 It is up to you to make sure that the assembler names you choose do not
4005 conflict with any other assembler symbols. Also, you must not use a
4006 register name; that would produce completely invalid assembler code. GCC
4007 does not as yet have the ability to store static variables in registers.
4008 Perhaps that will be added.
4010 @node Explicit Reg Vars
4011 @section Variables in Specified Registers
4012 @cindex explicit register variables
4013 @cindex variables in specified registers
4014 @cindex specified registers
4015 @cindex registers, global allocation
4017 GNU C allows you to put a few global variables into specified hardware
4018 registers. You can also specify the register in which an ordinary
4019 register variable should be allocated.
4023 Global register variables reserve registers throughout the program.
4024 This may be useful in programs such as programming language
4025 interpreters which have a couple of global variables that are accessed
4029 Local register variables in specific registers do not reserve the
4030 registers. The compiler's data flow analysis is capable of determining
4031 where the specified registers contain live values, and where they are
4032 available for other uses. Stores into local register variables may be deleted
4033 when they appear to be dead according to dataflow analysis. References
4034 to local register variables may be deleted or moved or simplified.
4036 These local variables are sometimes convenient for use with the extended
4037 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
4038 output of the assembler instruction directly into a particular register.
4039 (This will work provided the register you specify fits the constraints
4040 specified for that operand in the @code{asm}.)
4048 @node Global Reg Vars
4049 @subsection Defining Global Register Variables
4050 @cindex global register variables
4051 @cindex registers, global variables in
4053 You can define a global register variable in GNU C like this:
4056 register int *foo asm ("a5");
4060 Here @code{a5} is the name of the register which should be used. Choose a
4061 register which is normally saved and restored by function calls on your
4062 machine, so that library routines will not clobber it.
4064 Naturally the register name is cpu-dependent, so you would need to
4065 conditionalize your program according to cpu type. The register
4066 @code{a5} would be a good choice on a 68000 for a variable of pointer
4067 type. On machines with register windows, be sure to choose a ``global''
4068 register that is not affected magically by the function call mechanism.
4070 In addition, operating systems on one type of cpu may differ in how they
4071 name the registers; then you would need additional conditionals. For
4072 example, some 68000 operating systems call this register @code{%a5}.
4074 Eventually there may be a way of asking the compiler to choose a register
4075 automatically, but first we need to figure out how it should choose and
4076 how to enable you to guide the choice. No solution is evident.
4078 Defining a global register variable in a certain register reserves that
4079 register entirely for this use, at least within the current compilation.
4080 The register will not be allocated for any other purpose in the functions
4081 in the current compilation. The register will not be saved and restored by
4082 these functions. Stores into this register are never deleted even if they
4083 would appear to be dead, but references may be deleted or moved or
4086 It is not safe to access the global register variables from signal
4087 handlers, or from more than one thread of control, because the system
4088 library routines may temporarily use the register for other things (unless
4089 you recompile them specially for the task at hand).
4091 @cindex @code{qsort}, and global register variables
4092 It is not safe for one function that uses a global register variable to
4093 call another such function @code{foo} by way of a third function
4094 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4095 different source file in which the variable wasn't declared). This is
4096 because @code{lose} might save the register and put some other value there.
4097 For example, you can't expect a global register variable to be available in
4098 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4099 might have put something else in that register. (If you are prepared to
4100 recompile @code{qsort} with the same global register variable, you can
4101 solve this problem.)
4103 If you want to recompile @code{qsort} or other source files which do not
4104 actually use your global register variable, so that they will not use that
4105 register for any other purpose, then it suffices to specify the compiler
4106 option @option{-ffixed-@var{reg}}. You need not actually add a global
4107 register declaration to their source code.
4109 A function which can alter the value of a global register variable cannot
4110 safely be called from a function compiled without this variable, because it
4111 could clobber the value the caller expects to find there on return.
4112 Therefore, the function which is the entry point into the part of the
4113 program that uses the global register variable must explicitly save and
4114 restore the value which belongs to its caller.
4116 @cindex register variable after @code{longjmp}
4117 @cindex global register after @code{longjmp}
4118 @cindex value after @code{longjmp}
4121 On most machines, @code{longjmp} will restore to each global register
4122 variable the value it had at the time of the @code{setjmp}. On some
4123 machines, however, @code{longjmp} will not change the value of global
4124 register variables. To be portable, the function that called @code{setjmp}
4125 should make other arrangements to save the values of the global register
4126 variables, and to restore them in a @code{longjmp}. This way, the same
4127 thing will happen regardless of what @code{longjmp} does.
4129 All global register variable declarations must precede all function
4130 definitions. If such a declaration could appear after function
4131 definitions, the declaration would be too late to prevent the register from
4132 being used for other purposes in the preceding functions.
4134 Global register variables may not have initial values, because an
4135 executable file has no means to supply initial contents for a register.
4137 On the Sparc, there are reports that g3 @dots{} g7 are suitable
4138 registers, but certain library functions, such as @code{getwd}, as well
4139 as the subroutines for division and remainder, modify g3 and g4. g1 and
4140 g2 are local temporaries.
4142 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4143 Of course, it will not do to use more than a few of those.
4145 @node Local Reg Vars
4146 @subsection Specifying Registers for Local Variables
4147 @cindex local variables, specifying registers
4148 @cindex specifying registers for local variables
4149 @cindex registers for local variables
4151 You can define a local register variable with a specified register
4155 register int *foo asm ("a5");
4159 Here @code{a5} is the name of the register which should be used. Note
4160 that this is the same syntax used for defining global register
4161 variables, but for a local variable it would appear within a function.
4163 Naturally the register name is cpu-dependent, but this is not a
4164 problem, since specific registers are most often useful with explicit
4165 assembler instructions (@pxref{Extended Asm}). Both of these things
4166 generally require that you conditionalize your program according to
4169 In addition, operating systems on one type of cpu may differ in how they
4170 name the registers; then you would need additional conditionals. For
4171 example, some 68000 operating systems call this register @code{%a5}.
4173 Defining such a register variable does not reserve the register; it
4174 remains available for other uses in places where flow control determines
4175 the variable's value is not live. However, these registers are made
4176 unavailable for use in the reload pass; excessive use of this feature
4177 leaves the compiler too few available registers to compile certain
4180 This option does not guarantee that GCC will generate code that has
4181 this variable in the register you specify at all times. You may not
4182 code an explicit reference to this register in an @code{asm} statement
4183 and assume it will always refer to this variable.
4185 Stores into local register variables may be deleted when they appear to be dead
4186 according to dataflow analysis. References to local register variables may
4187 be deleted or moved or simplified.
4189 @node Alternate Keywords
4190 @section Alternate Keywords
4191 @cindex alternate keywords
4192 @cindex keywords, alternate
4194 @option{-ansi} and the various @option{-std} options disable certain
4195 keywords. This causes trouble when you want to use GNU C extensions, or
4196 a general-purpose header file that should be usable by all programs,
4197 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4198 @code{inline} are not available in programs compiled with
4199 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4200 program compiled with @option{-std=c99}). The ISO C99 keyword
4201 @code{restrict} is only available when @option{-std=gnu99} (which will
4202 eventually be the default) or @option{-std=c99} (or the equivalent
4203 @option{-std=iso9899:1999}) is used.
4205 The way to solve these problems is to put @samp{__} at the beginning and
4206 end of each problematical keyword. For example, use @code{__asm__}
4207 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4209 Other C compilers won't accept these alternative keywords; if you want to
4210 compile with another compiler, you can define the alternate keywords as
4211 macros to replace them with the customary keywords. It looks like this:
4219 @findex __extension__
4221 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4223 prevent such warnings within one expression by writing
4224 @code{__extension__} before the expression. @code{__extension__} has no
4225 effect aside from this.
4227 @node Incomplete Enums
4228 @section Incomplete @code{enum} Types
4230 You can define an @code{enum} tag without specifying its possible values.
4231 This results in an incomplete type, much like what you get if you write
4232 @code{struct foo} without describing the elements. A later declaration
4233 which does specify the possible values completes the type.
4235 You can't allocate variables or storage using the type while it is
4236 incomplete. However, you can work with pointers to that type.
4238 This extension may not be very useful, but it makes the handling of
4239 @code{enum} more consistent with the way @code{struct} and @code{union}
4242 This extension is not supported by GNU C++.
4244 @node Function Names
4245 @section Function Names as Strings
4246 @cindex @code{__FUNCTION__} identifier
4247 @cindex @code{__PRETTY_FUNCTION__} identifier
4248 @cindex @code{__func__} identifier
4250 GCC predefines two magic identifiers to hold the name of the current
4251 function. The identifier @code{__FUNCTION__} holds the name of the function
4252 as it appears in the source. The identifier @code{__PRETTY_FUNCTION__}
4253 holds the name of the function pretty printed in a language specific
4256 These names are always the same in a C function, but in a C++ function
4257 they may be different. For example, this program:
4261 extern int printf (char *, ...);
4268 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4269 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4287 __PRETTY_FUNCTION__ = int a::sub (int)
4290 The compiler automagically replaces the identifiers with a string
4291 literal containing the appropriate name. Thus, they are neither
4292 preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor
4293 variables. This means that they catenate with other string literals, and
4294 that they can be used to initialize char arrays. For example
4297 char here[] = "Function " __FUNCTION__ " in " __FILE__;
4300 On the other hand, @samp{#ifdef __FUNCTION__} does not have any special
4301 meaning inside a function, since the preprocessor does not do anything
4302 special with the identifier @code{__FUNCTION__}.
4304 Note that these semantics are deprecated, and that GCC 3.2 will handle
4305 @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} the same way as
4306 @code{__func__}. @code{__func__} is defined by the ISO standard C99:
4309 The identifier @code{__func__} is implicitly declared by the translator
4310 as if, immediately following the opening brace of each function
4311 definition, the declaration
4314 static const char __func__[] = "function-name";
4317 appeared, where function-name is the name of the lexically-enclosing
4318 function. This name is the unadorned name of the function.
4321 By this definition, @code{__func__} is a variable, not a string literal.
4322 In particular, @code{__func__} does not catenate with other string
4325 In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are
4326 variables, declared in the same way as @code{__func__}.
4328 @node Return Address
4329 @section Getting the Return or Frame Address of a Function
4331 These functions may be used to get information about the callers of a
4334 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4335 This function returns the return address of the current function, or of
4336 one of its callers. The @var{level} argument is number of frames to
4337 scan up the call stack. A value of @code{0} yields the return address
4338 of the current function, a value of @code{1} yields the return address
4339 of the caller of the current function, and so forth. When inlining
4340 the expected behavior is that the function will return the address of
4341 the function that will be returned to. To work around this behavior use
4342 the @code{noinline} function attribute.
4344 The @var{level} argument must be a constant integer.
4346 On some machines it may be impossible to determine the return address of
4347 any function other than the current one; in such cases, or when the top
4348 of the stack has been reached, this function will return @code{0} or a
4349 random value. In addition, @code{__builtin_frame_address} may be used
4350 to determine if the top of the stack has been reached.
4352 This function should only be used with a nonzero argument for debugging
4356 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4357 This function is similar to @code{__builtin_return_address}, but it
4358 returns the address of the function frame rather than the return address
4359 of the function. Calling @code{__builtin_frame_address} with a value of
4360 @code{0} yields the frame address of the current function, a value of
4361 @code{1} yields the frame address of the caller of the current function,
4364 The frame is the area on the stack which holds local variables and saved
4365 registers. The frame address is normally the address of the first word
4366 pushed on to the stack by the function. However, the exact definition
4367 depends upon the processor and the calling convention. If the processor
4368 has a dedicated frame pointer register, and the function has a frame,
4369 then @code{__builtin_frame_address} will return the value of the frame
4372 On some machines it may be impossible to determine the frame address of
4373 any function other than the current one; in such cases, or when the top
4374 of the stack has been reached, this function will return @code{0} if
4375 the first frame pointer is properly initialized by the startup code.
4377 This function should only be used with a nonzero argument for debugging
4381 @node Vector Extensions
4382 @section Using vector instructions through built-in functions
4384 On some targets, the instruction set contains SIMD vector instructions that
4385 operate on multiple values contained in one large register at the same time.
4386 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4389 The first step in using these extensions is to provide the necessary data
4390 types. This should be done using an appropriate @code{typedef}:
4393 typedef int v4si __attribute__ ((mode(V4SI)));
4396 The base type @code{int} is effectively ignored by the compiler, the
4397 actual properties of the new type @code{v4si} are defined by the
4398 @code{__attribute__}. It defines the machine mode to be used; for vector
4399 types these have the form @code{V@var{n}@var{B}}; @var{n} should be the
4400 number of elements in the vector, and @var{B} should be the base mode of the
4401 individual elements. The following can be used as base modes:
4405 An integer that is as wide as the smallest addressable unit, usually 8 bits.
4407 An integer, twice as wide as a QI mode integer, usually 16 bits.
4409 An integer, four times as wide as a QI mode integer, usually 32 bits.
4411 An integer, eight times as wide as a QI mode integer, usually 64 bits.
4413 A floating point value, as wide as a SI mode integer, usually 32 bits.
4415 A floating point value, as wide as a DI mode integer, usually 64 bits.
4418 There are no @code{V1xx} vector modes - they would be identical to the
4419 corresponding base mode.
4421 Specifying a combination that is not valid for the current architecture
4422 will cause gcc to synthesize the instructions using a narrower mode.
4423 For example, if you specify a variable of type @code{V4SI} and your
4424 architecture does not allow for this specific SIMD type, gcc will
4425 produce code that uses 4 @code{SIs}.
4427 The types defined in this manner can be used with a subset of normal C
4428 operations. Currently, gcc will allow using the following operators on
4429 these types: @code{+, -, *, /, unary minus}@.
4431 The operations behave like C++ @code{valarrays}. Addition is defined as
4432 the addition of the corresponding elements of the operands. For
4433 example, in the code below, each of the 4 elements in @var{a} will be
4434 added to the corresponding 4 elements in @var{b} and the resulting
4435 vector will be stored in @var{c}.
4438 typedef int v4si __attribute__ ((mode(V4SI)));
4445 Subtraction, multiplication, and division operate in a similar manner.
4446 Likewise, the result of using the unary minus operator on a vector type
4447 is a vector whose elements are the negative value of the corresponding
4448 elements in the operand.
4450 You can declare variables and use them in function calls and returns, as
4451 well as in assignments and some casts. You can specify a vector type as
4452 a return type for a function. Vector types can also be used as function
4453 arguments. It is possible to cast from one vector type to another,
4454 provided they are of the same size (in fact, you can also cast vectors
4455 to and from other datatypes of the same size).
4457 You cannot operate between vectors of different lengths or different
4458 signness without a cast.
4460 A port that supports hardware vector operations, usually provides a set
4461 of built-in functions that can be used to operate on vectors. For
4462 example, a function to add two vectors and multiply the result by a
4463 third could look like this:
4466 v4si f (v4si a, v4si b, v4si c)
4468 v4si tmp = __builtin_addv4si (a, b);
4469 return __builtin_mulv4si (tmp, c);
4474 @node Other Builtins
4475 @section Other built-in functions provided by GCC
4476 @cindex built-in functions
4477 @findex __builtin_isgreater
4478 @findex __builtin_isgreaterequal
4479 @findex __builtin_isless
4480 @findex __builtin_islessequal
4481 @findex __builtin_islessgreater
4482 @findex __builtin_isunordered
4508 @findex fprintf_unlocked
4510 @findex fputs_unlocked
4519 @findex printf_unlocked
4541 GCC provides a large number of built-in functions other than the ones
4542 mentioned above. Some of these are for internal use in the processing
4543 of exceptions or variable-length argument lists and will not be
4544 documented here because they may change from time to time; we do not
4545 recommend general use of these functions.
4547 The remaining functions are provided for optimization purposes.
4549 @opindex fno-builtin
4550 GCC includes built-in versions of many of the functions in the standard
4551 C library. The versions prefixed with @code{__builtin_} will always be
4552 treated as having the same meaning as the C library function even if you
4553 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
4554 Many of these functions are only optimized in certain cases; if they are
4555 not optimized in a particular case, a call to the library function will
4560 The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit}
4561 are recognized and presumed not to return, but otherwise are not built
4562 in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi},
4563 @option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in
4564 strict C89 mode (@option{-ansi} or @option{-std=c89}). All these functions
4565 have corresponding versions prefixed with @code{__builtin_}, which may be
4566 used even in strict C89 mode.
4568 Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp},
4569 @code{bzero}, @code{index}, @code{rindex}, @code{ffs}, @code{fputs_unlocked},
4570 @code{printf_unlocked} and @code{fprintf_unlocked} may be handled as
4571 built-in functions. All these functions have corresponding versions
4572 prefixed with @code{__builtin_}, which may be used even in strict C89
4575 The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl},
4576 @code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf},
4577 @code{cimagl}, @code{llabs} and @code{imaxabs} are handled as built-in
4578 functions except in strict ISO C90 mode. There are also built-in
4579 versions of the ISO C99 functions @code{cosf}, @code{cosl},
4580 @code{fabsf}, @code{fabsl}, @code{sinf}, @code{sinl}, @code{sqrtf}, and
4581 @code{sqrtl}, that are recognized in any mode since ISO C90 reserves
4582 these names for the purpose to which ISO C99 puts them. All these
4583 functions have corresponding versions prefixed with @code{__builtin_}.
4585 The ISO C90 functions @code{abs}, @code{cos}, @code{fabs},
4586 @code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy},
4587 @code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat},
4588 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
4589 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
4590 @code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr} are all
4591 recognized as built-in functions unless @option{-fno-builtin} is
4592 specified (or @option{-fno-builtin-@var{function}} is specified for an
4593 individual function). All of these functions have corresponding
4594 versions prefixed with @code{__builtin_}.
4596 GCC provides built-in versions of the ISO C99 floating point comparison
4597 macros that avoid raising exceptions for unordered operands. They have
4598 the same names as the standard macros ( @code{isgreater},
4599 @code{isgreaterequal}, @code{isless}, @code{islessequal},
4600 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
4601 prefixed. We intend for a library implementor to be able to simply
4602 @code{#define} each standard macro to its built-in equivalent.
4604 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
4606 You can use the built-in function @code{__builtin_types_compatible_p} to
4607 determine whether two types are the same.
4609 This built-in function returns 1 if the unqualified versions of the
4610 types @var{type1} and @var{type2} (which are types, not expressions) are
4611 compatible, 0 otherwise. The result of this built-in function can be
4612 used in integer constant expressions.
4614 This built-in function ignores top level qualifiers (e.g., @code{const},
4615 @code{volatile}). For example, @code{int} is equivalent to @code{const
4618 The type @code{int[]} and @code{int[5]} are compatible. On the other
4619 hand, @code{int} and @code{char *} are not compatible, even if the size
4620 of their types, on the particular architecture are the same. Also, the
4621 amount of pointer indirection is taken into account when determining
4622 similarity. Consequently, @code{short *} is not similar to
4623 @code{short **}. Furthermore, two types that are typedefed are
4624 considered compatible if their underlying types are compatible.
4626 An @code{enum} type is considered to be compatible with another
4627 @code{enum} type. For example, @code{enum @{foo, bar@}} is similar to
4628 @code{enum @{hot, dog@}}.
4630 You would typically use this function in code whose execution varies
4631 depending on the arguments' types. For example:
4637 if (__builtin_types_compatible_p (typeof (x), long double)) \
4638 tmp = foo_long_double (tmp); \
4639 else if (__builtin_types_compatible_p (typeof (x), double)) \
4640 tmp = foo_double (tmp); \
4641 else if (__builtin_types_compatible_p (typeof (x), float)) \
4642 tmp = foo_float (tmp); \
4649 @emph{Note:} This construct is only available for C.
4653 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
4655 You can use the built-in function @code{__builtin_choose_expr} to
4656 evaluate code depending on the value of a constant expression. This
4657 built-in function returns @var{exp1} if @var{const_exp}, which is a
4658 constant expression that must be able to be determined at compile time,
4659 is nonzero. Otherwise it returns 0.
4661 This built-in function is analogous to the @samp{? :} operator in C,
4662 except that the expression returned has its type unaltered by promotion
4663 rules. Also, the built-in function does not evaluate the expression
4664 that was not chosen. For example, if @var{const_exp} evaluates to true,
4665 @var{exp2} is not evaluated even if it has side-effects.
4667 This built-in function can return an lvalue if the chosen argument is an
4670 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
4671 type. Similarly, if @var{exp2} is returned, its return type is the same
4678 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), double), \
4680 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), float), \
4682 /* @r{The void expression results in a compile-time error} \
4683 @r{when assigning the result to something.} */ \
4687 @emph{Note:} This construct is only available for C. Furthermore, the
4688 unused expression (@var{exp1} or @var{exp2} depending on the value of
4689 @var{const_exp}) may still generate syntax errors. This may change in
4694 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
4695 You can use the built-in function @code{__builtin_constant_p} to
4696 determine if a value is known to be constant at compile-time and hence
4697 that GCC can perform constant-folding on expressions involving that
4698 value. The argument of the function is the value to test. The function
4699 returns the integer 1 if the argument is known to be a compile-time
4700 constant and 0 if it is not known to be a compile-time constant. A
4701 return of 0 does not indicate that the value is @emph{not} a constant,
4702 but merely that GCC cannot prove it is a constant with the specified
4703 value of the @option{-O} option.
4705 You would typically use this function in an embedded application where
4706 memory was a critical resource. If you have some complex calculation,
4707 you may want it to be folded if it involves constants, but need to call
4708 a function if it does not. For example:
4711 #define Scale_Value(X) \
4712 (__builtin_constant_p (X) \
4713 ? ((X) * SCALE + OFFSET) : Scale (X))
4716 You may use this built-in function in either a macro or an inline
4717 function. However, if you use it in an inlined function and pass an
4718 argument of the function as the argument to the built-in, GCC will
4719 never return 1 when you call the inline function with a string constant
4720 or compound literal (@pxref{Compound Literals}) and will not return 1
4721 when you pass a constant numeric value to the inline function unless you
4722 specify the @option{-O} option.
4724 You may also use @code{__builtin_constant_p} in initializers for static
4725 data. For instance, you can write
4728 static const int table[] = @{
4729 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
4735 This is an acceptable initializer even if @var{EXPRESSION} is not a
4736 constant expression. GCC must be more conservative about evaluating the
4737 built-in in this case, because it has no opportunity to perform
4740 Previous versions of GCC did not accept this built-in in data
4741 initializers. The earliest version where it is completely safe is
4745 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
4746 @opindex fprofile-arcs
4747 You may use @code{__builtin_expect} to provide the compiler with
4748 branch prediction information. In general, you should prefer to
4749 use actual profile feedback for this (@option{-fprofile-arcs}), as
4750 programmers are notoriously bad at predicting how their programs
4751 actually perform. However, there are applications in which this
4752 data is hard to collect.
4754 The return value is the value of @var{exp}, which should be an
4755 integral expression. The value of @var{c} must be a compile-time
4756 constant. The semantics of the built-in are that it is expected
4757 that @var{exp} == @var{c}. For example:
4760 if (__builtin_expect (x, 0))
4765 would indicate that we do not expect to call @code{foo}, since
4766 we expect @code{x} to be zero. Since you are limited to integral
4767 expressions for @var{exp}, you should use constructions such as
4770 if (__builtin_expect (ptr != NULL, 1))
4775 when testing pointer or floating-point values.
4778 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
4779 This function is used to minimize cache-miss latency by moving data into
4780 a cache before it is accessed.
4781 You can insert calls to @code{__builtin_prefetch} into code for which
4782 you know addresses of data in memory that is likely to be accessed soon.
4783 If the target supports them, data prefetch instructions will be generated.
4784 If the prefetch is done early enough before the access then the data will
4785 be in the cache by the time it is accessed.
4787 The value of @var{addr} is the address of the memory to prefetch.
4788 There are two optional arguments, @var{rw} and @var{locality}.
4789 The value of @var{rw} is a compile-time constant one or zero; one
4790 means that the prefetch is preparing for a write to the memory address
4791 and zero, the default, means that the prefetch is preparing for a read.
4792 The value @var{locality} must be a compile-time constant integer between
4793 zero and three. A value of zero means that the data has no temporal
4794 locality, so it need not be left in the cache after the access. A value
4795 of three means that the data has a high degree of temporal locality and
4796 should be left in all levels of cache possible. Values of one and two
4797 mean, respectively, a low or moderate degree of temporal locality. The
4801 for (i = 0; i < n; i++)
4804 __builtin_prefetch (&a[i+j], 1, 1);
4805 __builtin_prefetch (&b[i+j], 0, 1);
4810 Data prefetch does not generate faults if @var{addr} is invalid, but
4811 the address expression itself must be valid. For example, a prefetch
4812 of @code{p->next} will not fault if @code{p->next} is not a valid
4813 address, but evaluation will fault if @code{p} is not a valid address.
4815 If the target does not support data prefetch, the address expression
4816 is evaluated if it includes side effects but no other code is generated
4817 and GCC does not issue a warning.
4820 @node Target Builtins
4821 @section Built-in Functions Specific to Particular Target Machines
4823 On some target machines, GCC supports many built-in functions specific
4824 to those machines. Generally these generate calls to specific machine
4825 instructions, but allow the compiler to schedule those calls.
4828 * Alpha Built-in Functions::
4829 * X86 Built-in Functions::
4830 * PowerPC AltiVec Built-in Functions::
4833 @node Alpha Built-in Functions
4834 @subsection Alpha Built-in Functions
4836 These built-in functions are available for the Alpha family of
4837 processors, depending on the command-line switches used.
4839 The following built-in functions are always available. They
4840 all generate the machine instruction that is part of the name.
4843 long __builtin_alpha_implver (void)
4844 long __builtin_alpha_rpcc (void)
4845 long __builtin_alpha_amask (long)
4846 long __builtin_alpha_cmpbge (long, long)
4847 long __builtin_alpha_extbl (long, long)
4848 long __builtin_alpha_extwl (long, long)
4849 long __builtin_alpha_extll (long, long)
4850 long __builtin_alpha_extql (long, long)
4851 long __builtin_alpha_extwh (long, long)
4852 long __builtin_alpha_extlh (long, long)
4853 long __builtin_alpha_extqh (long, long)
4854 long __builtin_alpha_insbl (long, long)
4855 long __builtin_alpha_inswl (long, long)
4856 long __builtin_alpha_insll (long, long)
4857 long __builtin_alpha_insql (long, long)
4858 long __builtin_alpha_inswh (long, long)
4859 long __builtin_alpha_inslh (long, long)
4860 long __builtin_alpha_insqh (long, long)
4861 long __builtin_alpha_mskbl (long, long)
4862 long __builtin_alpha_mskwl (long, long)
4863 long __builtin_alpha_mskll (long, long)
4864 long __builtin_alpha_mskql (long, long)
4865 long __builtin_alpha_mskwh (long, long)
4866 long __builtin_alpha_msklh (long, long)
4867 long __builtin_alpha_mskqh (long, long)
4868 long __builtin_alpha_umulh (long, long)
4869 long __builtin_alpha_zap (long, long)
4870 long __builtin_alpha_zapnot (long, long)
4873 The following built-in functions are always with @option{-mmax}
4874 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
4875 later. They all generate the machine instruction that is part
4879 long __builtin_alpha_pklb (long)
4880 long __builtin_alpha_pkwb (long)
4881 long __builtin_alpha_unpkbl (long)
4882 long __builtin_alpha_unpkbw (long)
4883 long __builtin_alpha_minub8 (long, long)
4884 long __builtin_alpha_minsb8 (long, long)
4885 long __builtin_alpha_minuw4 (long, long)
4886 long __builtin_alpha_minsw4 (long, long)
4887 long __builtin_alpha_maxub8 (long, long)
4888 long __builtin_alpha_maxsb8 (long, long)
4889 long __builtin_alpha_maxuw4 (long, long)
4890 long __builtin_alpha_maxsw4 (long, long)
4891 long __builtin_alpha_perr (long, long)
4894 The following built-in functions are always with @option{-mcix}
4895 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
4896 later. They all generate the machine instruction that is part
4900 long __builtin_alpha_cttz (long)
4901 long __builtin_alpha_ctlz (long)
4902 long __builtin_alpha_ctpop (long)
4905 The following builtins are available on systems that use the OSF/1
4906 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
4907 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
4908 @code{rdval} and @code{wrval}.
4911 void *__builtin_thread_pointer (void)
4912 void __builtin_set_thread_pointer (void *)
4915 @node X86 Built-in Functions
4916 @subsection X86 Built-in Functions
4918 These built-in functions are available for the i386 and x86-64 family
4919 of computers, depending on the command-line switches used.
4921 The following machine modes are available for use with MMX built-in functions
4922 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
4923 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
4924 vector of eight 8-bit integers. Some of the built-in functions operate on
4925 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
4927 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
4928 of two 32-bit floating point values.
4930 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
4931 floating point values. Some instructions use a vector of four 32-bit
4932 integers, these use @code{V4SI}. Finally, some instructions operate on an
4933 entire vector register, interpreting it as a 128-bit integer, these use mode
4936 The following built-in functions are made available by @option{-mmmx}.
4937 All of them generate the machine instruction that is part of the name.
4940 v8qi __builtin_ia32_paddb (v8qi, v8qi)
4941 v4hi __builtin_ia32_paddw (v4hi, v4hi)
4942 v2si __builtin_ia32_paddd (v2si, v2si)
4943 v8qi __builtin_ia32_psubb (v8qi, v8qi)
4944 v4hi __builtin_ia32_psubw (v4hi, v4hi)
4945 v2si __builtin_ia32_psubd (v2si, v2si)
4946 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
4947 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
4948 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
4949 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
4950 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
4951 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
4952 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
4953 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
4954 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
4955 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
4956 di __builtin_ia32_pand (di, di)
4957 di __builtin_ia32_pandn (di,di)
4958 di __builtin_ia32_por (di, di)
4959 di __builtin_ia32_pxor (di, di)
4960 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
4961 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
4962 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
4963 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
4964 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
4965 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
4966 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
4967 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
4968 v2si __builtin_ia32_punpckhdq (v2si, v2si)
4969 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
4970 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
4971 v2si __builtin_ia32_punpckldq (v2si, v2si)
4972 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
4973 v4hi __builtin_ia32_packssdw (v2si, v2si)
4974 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
4977 The following built-in functions are made available either with
4978 @option{-msse}, or with a combination of @option{-m3dnow} and
4979 @option{-march=athlon}. All of them generate the machine
4980 instruction that is part of the name.
4983 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
4984 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
4985 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
4986 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
4987 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
4988 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
4989 v8qi __builtin_ia32_pminub (v8qi, v8qi)
4990 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
4991 int __builtin_ia32_pextrw (v4hi, int)
4992 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
4993 int __builtin_ia32_pmovmskb (v8qi)
4994 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
4995 void __builtin_ia32_movntq (di *, di)
4996 void __builtin_ia32_sfence (void)
4999 The following built-in functions are available when @option{-msse} is used.
5000 All of them generate the machine instruction that is part of the name.
5003 int __builtin_ia32_comieq (v4sf, v4sf)
5004 int __builtin_ia32_comineq (v4sf, v4sf)
5005 int __builtin_ia32_comilt (v4sf, v4sf)
5006 int __builtin_ia32_comile (v4sf, v4sf)
5007 int __builtin_ia32_comigt (v4sf, v4sf)
5008 int __builtin_ia32_comige (v4sf, v4sf)
5009 int __builtin_ia32_ucomieq (v4sf, v4sf)
5010 int __builtin_ia32_ucomineq (v4sf, v4sf)
5011 int __builtin_ia32_ucomilt (v4sf, v4sf)
5012 int __builtin_ia32_ucomile (v4sf, v4sf)
5013 int __builtin_ia32_ucomigt (v4sf, v4sf)
5014 int __builtin_ia32_ucomige (v4sf, v4sf)
5015 v4sf __builtin_ia32_addps (v4sf, v4sf)
5016 v4sf __builtin_ia32_subps (v4sf, v4sf)
5017 v4sf __builtin_ia32_mulps (v4sf, v4sf)
5018 v4sf __builtin_ia32_divps (v4sf, v4sf)
5019 v4sf __builtin_ia32_addss (v4sf, v4sf)
5020 v4sf __builtin_ia32_subss (v4sf, v4sf)
5021 v4sf __builtin_ia32_mulss (v4sf, v4sf)
5022 v4sf __builtin_ia32_divss (v4sf, v4sf)
5023 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
5024 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
5025 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
5026 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
5027 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
5028 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
5029 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
5030 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
5031 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
5032 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
5033 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
5034 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
5035 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
5036 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
5037 v4si __builtin_ia32_cmpless (v4sf, v4sf)
5038 v4si __builtin_ia32_cmpgtss (v4sf, v4sf)
5039 v4si __builtin_ia32_cmpgess (v4sf, v4sf)
5040 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
5041 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
5042 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
5043 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
5044 v4si __builtin_ia32_cmpngtss (v4sf, v4sf)
5045 v4si __builtin_ia32_cmpngess (v4sf, v4sf)
5046 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
5047 v4sf __builtin_ia32_maxps (v4sf, v4sf)
5048 v4sf __builtin_ia32_maxss (v4sf, v4sf)
5049 v4sf __builtin_ia32_minps (v4sf, v4sf)
5050 v4sf __builtin_ia32_minss (v4sf, v4sf)
5051 v4sf __builtin_ia32_andps (v4sf, v4sf)
5052 v4sf __builtin_ia32_andnps (v4sf, v4sf)
5053 v4sf __builtin_ia32_orps (v4sf, v4sf)
5054 v4sf __builtin_ia32_xorps (v4sf, v4sf)
5055 v4sf __builtin_ia32_movss (v4sf, v4sf)
5056 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
5057 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
5058 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
5059 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
5060 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
5061 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
5062 v2si __builtin_ia32_cvtps2pi (v4sf)
5063 int __builtin_ia32_cvtss2si (v4sf)
5064 v2si __builtin_ia32_cvttps2pi (v4sf)
5065 int __builtin_ia32_cvttss2si (v4sf)
5066 v4sf __builtin_ia32_rcpps (v4sf)
5067 v4sf __builtin_ia32_rsqrtps (v4sf)
5068 v4sf __builtin_ia32_sqrtps (v4sf)
5069 v4sf __builtin_ia32_rcpss (v4sf)
5070 v4sf __builtin_ia32_rsqrtss (v4sf)
5071 v4sf __builtin_ia32_sqrtss (v4sf)
5072 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
5073 void __builtin_ia32_movntps (float *, v4sf)
5074 int __builtin_ia32_movmskps (v4sf)
5077 The following built-in functions are available when @option{-msse} is used.
5080 @item v4sf __builtin_ia32_loadaps (float *)
5081 Generates the @code{movaps} machine instruction as a load from memory.
5082 @item void __builtin_ia32_storeaps (float *, v4sf)
5083 Generates the @code{movaps} machine instruction as a store to memory.
5084 @item v4sf __builtin_ia32_loadups (float *)
5085 Generates the @code{movups} machine instruction as a load from memory.
5086 @item void __builtin_ia32_storeups (float *, v4sf)
5087 Generates the @code{movups} machine instruction as a store to memory.
5088 @item v4sf __builtin_ia32_loadsss (float *)
5089 Generates the @code{movss} machine instruction as a load from memory.
5090 @item void __builtin_ia32_storess (float *, v4sf)
5091 Generates the @code{movss} machine instruction as a store to memory.
5092 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
5093 Generates the @code{movhps} machine instruction as a load from memory.
5094 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
5095 Generates the @code{movlps} machine instruction as a load from memory
5096 @item void __builtin_ia32_storehps (v4sf, v2si *)
5097 Generates the @code{movhps} machine instruction as a store to memory.
5098 @item void __builtin_ia32_storelps (v4sf, v2si *)
5099 Generates the @code{movlps} machine instruction as a store to memory.
5102 The following built-in functions are available when @option{-m3dnow} is used.
5103 All of them generate the machine instruction that is part of the name.
5106 void __builtin_ia32_femms (void)
5107 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
5108 v2si __builtin_ia32_pf2id (v2sf)
5109 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
5110 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
5111 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
5112 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
5113 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
5114 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
5115 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
5116 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
5117 v2sf __builtin_ia32_pfrcp (v2sf)
5118 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
5119 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
5120 v2sf __builtin_ia32_pfrsqrt (v2sf)
5121 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
5122 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
5123 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
5124 v2sf __builtin_ia32_pi2fd (v2si)
5125 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
5128 The following built-in functions are available when both @option{-m3dnow}
5129 and @option{-march=athlon} are used. All of them generate the machine
5130 instruction that is part of the name.
5133 v2si __builtin_ia32_pf2iw (v2sf)
5134 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
5135 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
5136 v2sf __builtin_ia32_pi2fw (v2si)
5137 v2sf __builtin_ia32_pswapdsf (v2sf)
5138 v2si __builtin_ia32_pswapdsi (v2si)
5141 @node PowerPC AltiVec Built-in Functions
5142 @subsection PowerPC AltiVec Built-in Functions
5144 These built-in functions are available for the PowerPC family
5145 of computers, depending on the command-line switches used.
5147 The following machine modes are available for use with AltiVec built-in
5148 functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four
5149 32-bit integers, @code{V4SF} for a vector of four 32-bit floating point
5150 numbers, @code{V8HI} for a vector of eight 16-bit integers, and
5151 @code{V16QI} for a vector of sixteen 8-bit integers.
5153 The following functions are made available by including
5154 @code{<altivec.h>} and using @option{-maltivec} and
5155 @option{-mabi=altivec}. The functions implement the functionality
5156 described in Motorola's AltiVec Programming Interface Manual.
5158 There are a few differences from Motorola's documentation and GCC's
5159 implementation. Vector constants are done with curly braces (not
5160 parentheses). Vector initializers require no casts if the vector
5161 constant is of the same type as the variable it is initializing. The
5162 @code{vector bool} type is deprecated and will be discontinued in
5163 further revisions. Use @code{vector signed} instead. If @code{signed}
5164 or @code{unsigned} is omitted, the vector type will default to
5165 @code{signed}. Lastly, all overloaded functions are implemented with macros
5166 for the C implementation. So code the following example will not work:
5169 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
5172 Since vec_add is a macro, the vector constant in the above example will
5173 be treated as four different arguments. Wrap the entire argument in
5174 parentheses for this to work. The C++ implementation does not use
5177 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
5178 Internally, GCC uses built-in functions to achieve the functionality in
5179 the aforementioned header file, but they are not supported and are
5180 subject to change without notice.
5183 vector signed char vec_abs (vector signed char, vector signed char);
5184 vector signed short vec_abs (vector signed short, vector signed short);
5185 vector signed int vec_abs (vector signed int, vector signed int);
5186 vector signed float vec_abs (vector signed float, vector signed float);
5188 vector signed char vec_abss (vector signed char, vector signed char);
5189 vector signed short vec_abss (vector signed short, vector signed short);
5191 vector signed char vec_add (vector signed char, vector signed char);
5192 vector unsigned char vec_add (vector signed char, vector unsigned char);
5194 vector unsigned char vec_add (vector unsigned char, vector signed char);
5196 vector unsigned char vec_add (vector unsigned char,
5197 vector unsigned char);
5198 vector signed short vec_add (vector signed short, vector signed short);
5199 vector unsigned short vec_add (vector signed short,
5200 vector unsigned short);
5201 vector unsigned short vec_add (vector unsigned short,
5202 vector signed short);
5203 vector unsigned short vec_add (vector unsigned short,
5204 vector unsigned short);
5205 vector signed int vec_add (vector signed int, vector signed int);
5206 vector unsigned int vec_add (vector signed int, vector unsigned int);
5207 vector unsigned int vec_add (vector unsigned int, vector signed int);
5208 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
5209 vector float vec_add (vector float, vector float);
5211 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
5213 vector unsigned char vec_adds (vector signed char,
5214 vector unsigned char);
5215 vector unsigned char vec_adds (vector unsigned char,
5216 vector signed char);
5217 vector unsigned char vec_adds (vector unsigned char,
5218 vector unsigned char);
5219 vector signed char vec_adds (vector signed char, vector signed char);
5220 vector unsigned short vec_adds (vector signed short,
5221 vector unsigned short);
5222 vector unsigned short vec_adds (vector unsigned short,
5223 vector signed short);
5224 vector unsigned short vec_adds (vector unsigned short,
5225 vector unsigned short);
5226 vector signed short vec_adds (vector signed short, vector signed short);
5228 vector unsigned int vec_adds (vector signed int, vector unsigned int);
5229 vector unsigned int vec_adds (vector unsigned int, vector signed int);
5230 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
5232 vector signed int vec_adds (vector signed int, vector signed int);
5234 vector float vec_and (vector float, vector float);
5235 vector float vec_and (vector float, vector signed int);
5236 vector float vec_and (vector signed int, vector float);
5237 vector signed int vec_and (vector signed int, vector signed int);
5238 vector unsigned int vec_and (vector signed int, vector unsigned int);
5239 vector unsigned int vec_and (vector unsigned int, vector signed int);
5240 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
5241 vector signed short vec_and (vector signed short, vector signed short);
5242 vector unsigned short vec_and (vector signed short,
5243 vector unsigned short);
5244 vector unsigned short vec_and (vector unsigned short,
5245 vector signed short);
5246 vector unsigned short vec_and (vector unsigned short,
5247 vector unsigned short);
5248 vector signed char vec_and (vector signed char, vector signed char);
5249 vector unsigned char vec_and (vector signed char, vector unsigned char);
5251 vector unsigned char vec_and (vector unsigned char, vector signed char);
5253 vector unsigned char vec_and (vector unsigned char,
5254 vector unsigned char);
5256 vector float vec_andc (vector float, vector float);
5257 vector float vec_andc (vector float, vector signed int);
5258 vector float vec_andc (vector signed int, vector float);
5259 vector signed int vec_andc (vector signed int, vector signed int);
5260 vector unsigned int vec_andc (vector signed int, vector unsigned int);
5261 vector unsigned int vec_andc (vector unsigned int, vector signed int);
5262 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
5264 vector signed short vec_andc (vector signed short, vector signed short);
5266 vector unsigned short vec_andc (vector signed short,
5267 vector unsigned short);
5268 vector unsigned short vec_andc (vector unsigned short,
5269 vector signed short);
5270 vector unsigned short vec_andc (vector unsigned short,
5271 vector unsigned short);
5272 vector signed char vec_andc (vector signed char, vector signed char);
5273 vector unsigned char vec_andc (vector signed char,
5274 vector unsigned char);
5275 vector unsigned char vec_andc (vector unsigned char,
5276 vector signed char);
5277 vector unsigned char vec_andc (vector unsigned char,
5278 vector unsigned char);
5280 vector unsigned char vec_avg (vector unsigned char,
5281 vector unsigned char);
5282 vector signed char vec_avg (vector signed char, vector signed char);
5283 vector unsigned short vec_avg (vector unsigned short,
5284 vector unsigned short);
5285 vector signed short vec_avg (vector signed short, vector signed short);
5286 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
5287 vector signed int vec_avg (vector signed int, vector signed int);
5289 vector float vec_ceil (vector float);
5291 vector signed int vec_cmpb (vector float, vector float);
5293 vector signed char vec_cmpeq (vector signed char, vector signed char);
5294 vector signed char vec_cmpeq (vector unsigned char,
5295 vector unsigned char);
5296 vector signed short vec_cmpeq (vector signed short,
5297 vector signed short);
5298 vector signed short vec_cmpeq (vector unsigned short,
5299 vector unsigned short);
5300 vector signed int vec_cmpeq (vector signed int, vector signed int);
5301 vector signed int vec_cmpeq (vector unsigned int, vector unsigned int);
5302 vector signed int vec_cmpeq (vector float, vector float);
5304 vector signed int vec_cmpge (vector float, vector float);
5306 vector signed char vec_cmpgt (vector unsigned char,
5307 vector unsigned char);
5308 vector signed char vec_cmpgt (vector signed char, vector signed char);
5309 vector signed short vec_cmpgt (vector unsigned short,
5310 vector unsigned short);
5311 vector signed short vec_cmpgt (vector signed short,
5312 vector signed short);
5313 vector signed int vec_cmpgt (vector unsigned int, vector unsigned int);
5314 vector signed int vec_cmpgt (vector signed int, vector signed int);
5315 vector signed int vec_cmpgt (vector float, vector float);
5317 vector signed int vec_cmple (vector float, vector float);
5319 vector signed char vec_cmplt (vector unsigned char,
5320 vector unsigned char);
5321 vector signed char vec_cmplt (vector signed char, vector signed char);
5322 vector signed short vec_cmplt (vector unsigned short,
5323 vector unsigned short);
5324 vector signed short vec_cmplt (vector signed short,
5325 vector signed short);
5326 vector signed int vec_cmplt (vector unsigned int, vector unsigned int);
5327 vector signed int vec_cmplt (vector signed int, vector signed int);
5328 vector signed int vec_cmplt (vector float, vector float);
5330 vector float vec_ctf (vector unsigned int, const char);
5331 vector float vec_ctf (vector signed int, const char);
5333 vector signed int vec_cts (vector float, const char);
5335 vector unsigned int vec_ctu (vector float, const char);
5337 void vec_dss (const char);
5339 void vec_dssall (void);
5341 void vec_dst (void *, int, const char);
5343 void vec_dstst (void *, int, const char);
5345 void vec_dststt (void *, int, const char);
5347 void vec_dstt (void *, int, const char);
5349 vector float vec_expte (vector float, vector float);
5351 vector float vec_floor (vector float, vector float);
5353 vector float vec_ld (int, vector float *);
5354 vector float vec_ld (int, float *):
5355 vector signed int vec_ld (int, int *);
5356 vector signed int vec_ld (int, vector signed int *);
5357 vector unsigned int vec_ld (int, vector unsigned int *);
5358 vector unsigned int vec_ld (int, unsigned int *);
5359 vector signed short vec_ld (int, short *, vector signed short *);
5360 vector unsigned short vec_ld (int, unsigned short *,
5361 vector unsigned short *);
5362 vector signed char vec_ld (int, signed char *);
5363 vector signed char vec_ld (int, vector signed char *);
5364 vector unsigned char vec_ld (int, unsigned char *);
5365 vector unsigned char vec_ld (int, vector unsigned char *);
5367 vector signed char vec_lde (int, signed char *);
5368 vector unsigned char vec_lde (int, unsigned char *);
5369 vector signed short vec_lde (int, short *);
5370 vector unsigned short vec_lde (int, unsigned short *);
5371 vector float vec_lde (int, float *);
5372 vector signed int vec_lde (int, int *);
5373 vector unsigned int vec_lde (int, unsigned int *);
5375 void float vec_ldl (int, float *);
5376 void float vec_ldl (int, vector float *);
5377 void signed int vec_ldl (int, vector signed int *);
5378 void signed int vec_ldl (int, int *);
5379 void unsigned int vec_ldl (int, unsigned int *);
5380 void unsigned int vec_ldl (int, vector unsigned int *);
5381 void signed short vec_ldl (int, vector signed short *);
5382 void signed short vec_ldl (int, short *);
5383 void unsigned short vec_ldl (int, vector unsigned short *);
5384 void unsigned short vec_ldl (int, unsigned short *);
5385 void signed char vec_ldl (int, vector signed char *);
5386 void signed char vec_ldl (int, signed char *);
5387 void unsigned char vec_ldl (int, vector unsigned char *);
5388 void unsigned char vec_ldl (int, unsigned char *);
5390 vector float vec_loge (vector float);
5392 vector unsigned char vec_lvsl (int, void *, int *);
5394 vector unsigned char vec_lvsr (int, void *, int *);
5396 vector float vec_madd (vector float, vector float, vector float);
5398 vector signed short vec_madds (vector signed short, vector signed short,
5399 vector signed short);
5401 vector unsigned char vec_max (vector signed char, vector unsigned char);
5403 vector unsigned char vec_max (vector unsigned char, vector signed char);
5405 vector unsigned char vec_max (vector unsigned char,
5406 vector unsigned char);
5407 vector signed char vec_max (vector signed char, vector signed char);
5408 vector unsigned short vec_max (vector signed short,
5409 vector unsigned short);
5410 vector unsigned short vec_max (vector unsigned short,
5411 vector signed short);
5412 vector unsigned short vec_max (vector unsigned short,
5413 vector unsigned short);
5414 vector signed short vec_max (vector signed short, vector signed short);
5415 vector unsigned int vec_max (vector signed int, vector unsigned int);
5416 vector unsigned int vec_max (vector unsigned int, vector signed int);
5417 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
5418 vector signed int vec_max (vector signed int, vector signed int);
5419 vector float vec_max (vector float, vector float);
5421 vector signed char vec_mergeh (vector signed char, vector signed char);
5422 vector unsigned char vec_mergeh (vector unsigned char,
5423 vector unsigned char);
5424 vector signed short vec_mergeh (vector signed short,
5425 vector signed short);
5426 vector unsigned short vec_mergeh (vector unsigned short,
5427 vector unsigned short);
5428 vector float vec_mergeh (vector float, vector float);
5429 vector signed int vec_mergeh (vector signed int, vector signed int);
5430 vector unsigned int vec_mergeh (vector unsigned int,
5431 vector unsigned int);
5433 vector signed char vec_mergel (vector signed char, vector signed char);
5434 vector unsigned char vec_mergel (vector unsigned char,
5435 vector unsigned char);
5436 vector signed short vec_mergel (vector signed short,
5437 vector signed short);
5438 vector unsigned short vec_mergel (vector unsigned short,
5439 vector unsigned short);
5440 vector float vec_mergel (vector float, vector float);
5441 vector signed int vec_mergel (vector signed int, vector signed int);
5442 vector unsigned int vec_mergel (vector unsigned int,
5443 vector unsigned int);
5445 vector unsigned short vec_mfvscr (void);
5447 vector unsigned char vec_min (vector signed char, vector unsigned char);
5449 vector unsigned char vec_min (vector unsigned char, vector signed char);
5451 vector unsigned char vec_min (vector unsigned char,
5452 vector unsigned char);
5453 vector signed char vec_min (vector signed char, vector signed char);
5454 vector unsigned short vec_min (vector signed short,
5455 vector unsigned short);
5456 vector unsigned short vec_min (vector unsigned short,
5457 vector signed short);
5458 vector unsigned short vec_min (vector unsigned short,
5459 vector unsigned short);
5460 vector signed short vec_min (vector signed short, vector signed short);
5461 vector unsigned int vec_min (vector signed int, vector unsigned int);
5462 vector unsigned int vec_min (vector unsigned int, vector signed int);
5463 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
5464 vector signed int vec_min (vector signed int, vector signed int);
5465 vector float vec_min (vector float, vector float);
5467 vector signed short vec_mladd (vector signed short, vector signed short,
5468 vector signed short);
5469 vector signed short vec_mladd (vector signed short,
5470 vector unsigned short,
5471 vector unsigned short);
5472 vector signed short vec_mladd (vector unsigned short,
5473 vector signed short,
5474 vector signed short);
5475 vector unsigned short vec_mladd (vector unsigned short,
5476 vector unsigned short,
5477 vector unsigned short);
5479 vector signed short vec_mradds (vector signed short,
5480 vector signed short,
5481 vector signed short);
5483 vector unsigned int vec_msum (vector unsigned char,
5484 vector unsigned char,
5485 vector unsigned int);
5486 vector signed int vec_msum (vector signed char, vector unsigned char,
5488 vector unsigned int vec_msum (vector unsigned short,
5489 vector unsigned short,
5490 vector unsigned int);
5491 vector signed int vec_msum (vector signed short, vector signed short,
5494 vector unsigned int vec_msums (vector unsigned short,
5495 vector unsigned short,
5496 vector unsigned int);
5497 vector signed int vec_msums (vector signed short, vector signed short,
5500 void vec_mtvscr (vector signed int);
5501 void vec_mtvscr (vector unsigned int);
5502 void vec_mtvscr (vector signed short);
5503 void vec_mtvscr (vector unsigned short);
5504 void vec_mtvscr (vector signed char);
5505 void vec_mtvscr (vector unsigned char);
5507 vector unsigned short vec_mule (vector unsigned char,
5508 vector unsigned char);
5509 vector signed short vec_mule (vector signed char, vector signed char);
5510 vector unsigned int vec_mule (vector unsigned short,
5511 vector unsigned short);
5512 vector signed int vec_mule (vector signed short, vector signed short);
5514 vector unsigned short vec_mulo (vector unsigned char,
5515 vector unsigned char);
5516 vector signed short vec_mulo (vector signed char, vector signed char);
5517 vector unsigned int vec_mulo (vector unsigned short,
5518 vector unsigned short);
5519 vector signed int vec_mulo (vector signed short, vector signed short);
5521 vector float vec_nmsub (vector float, vector float, vector float);
5523 vector float vec_nor (vector float, vector float);
5524 vector signed int vec_nor (vector signed int, vector signed int);
5525 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
5526 vector signed short vec_nor (vector signed short, vector signed short);
5527 vector unsigned short vec_nor (vector unsigned short,
5528 vector unsigned short);
5529 vector signed char vec_nor (vector signed char, vector signed char);
5530 vector unsigned char vec_nor (vector unsigned char,
5531 vector unsigned char);
5533 vector float vec_or (vector float, vector float);
5534 vector float vec_or (vector float, vector signed int);
5535 vector float vec_or (vector signed int, vector float);
5536 vector signed int vec_or (vector signed int, vector signed int);
5537 vector unsigned int vec_or (vector signed int, vector unsigned int);
5538 vector unsigned int vec_or (vector unsigned int, vector signed int);
5539 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
5540 vector signed short vec_or (vector signed short, vector signed short);
5541 vector unsigned short vec_or (vector signed short,
5542 vector unsigned short);
5543 vector unsigned short vec_or (vector unsigned short,
5544 vector signed short);
5545 vector unsigned short vec_or (vector unsigned short,
5546 vector unsigned short);
5547 vector signed char vec_or (vector signed char, vector signed char);
5548 vector unsigned char vec_or (vector signed char, vector unsigned char);
5549 vector unsigned char vec_or (vector unsigned char, vector signed char);
5550 vector unsigned char vec_or (vector unsigned char,
5551 vector unsigned char);
5553 vector signed char vec_pack (vector signed short, vector signed short);
5554 vector unsigned char vec_pack (vector unsigned short,
5555 vector unsigned short);
5556 vector signed short vec_pack (vector signed int, vector signed int);
5557 vector unsigned short vec_pack (vector unsigned int,
5558 vector unsigned int);
5560 vector signed short vec_packpx (vector unsigned int,
5561 vector unsigned int);
5563 vector unsigned char vec_packs (vector unsigned short,
5564 vector unsigned short);
5565 vector signed char vec_packs (vector signed short, vector signed short);
5567 vector unsigned short vec_packs (vector unsigned int,
5568 vector unsigned int);
5569 vector signed short vec_packs (vector signed int, vector signed int);
5571 vector unsigned char vec_packsu (vector unsigned short,
5572 vector unsigned short);
5573 vector unsigned char vec_packsu (vector signed short,
5574 vector signed short);
5575 vector unsigned short vec_packsu (vector unsigned int,
5576 vector unsigned int);
5577 vector unsigned short vec_packsu (vector signed int, vector signed int);
5579 vector float vec_perm (vector float, vector float,
5580 vector unsigned char);
5581 vector signed int vec_perm (vector signed int, vector signed int,
5582 vector unsigned char);
5583 vector unsigned int vec_perm (vector unsigned int, vector unsigned int,
5584 vector unsigned char);
5585 vector signed short vec_perm (vector signed short, vector signed short,
5586 vector unsigned char);
5587 vector unsigned short vec_perm (vector unsigned short,
5588 vector unsigned short,
5589 vector unsigned char);
5590 vector signed char vec_perm (vector signed char, vector signed char,
5591 vector unsigned char);
5592 vector unsigned char vec_perm (vector unsigned char,
5593 vector unsigned char,
5594 vector unsigned char);
5596 vector float vec_re (vector float);
5598 vector signed char vec_rl (vector signed char, vector unsigned char);
5599 vector unsigned char vec_rl (vector unsigned char,
5600 vector unsigned char);
5601 vector signed short vec_rl (vector signed short, vector unsigned short);
5603 vector unsigned short vec_rl (vector unsigned short,
5604 vector unsigned short);
5605 vector signed int vec_rl (vector signed int, vector unsigned int);
5606 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
5608 vector float vec_round (vector float);
5610 vector float vec_rsqrte (vector float);
5612 vector float vec_sel (vector float, vector float, vector signed int);
5613 vector float vec_sel (vector float, vector float, vector unsigned int);
5614 vector signed int vec_sel (vector signed int, vector signed int,
5616 vector signed int vec_sel (vector signed int, vector signed int,
5617 vector unsigned int);
5618 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
5620 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
5621 vector unsigned int);
5622 vector signed short vec_sel (vector signed short, vector signed short,
5623 vector signed short);
5624 vector signed short vec_sel (vector signed short, vector signed short,
5625 vector unsigned short);
5626 vector unsigned short vec_sel (vector unsigned short,
5627 vector unsigned short,
5628 vector signed short);
5629 vector unsigned short vec_sel (vector unsigned short,
5630 vector unsigned short,
5631 vector unsigned short);
5632 vector signed char vec_sel (vector signed char, vector signed char,
5633 vector signed char);
5634 vector signed char vec_sel (vector signed char, vector signed char,
5635 vector unsigned char);
5636 vector unsigned char vec_sel (vector unsigned char,
5637 vector unsigned char,
5638 vector signed char);
5639 vector unsigned char vec_sel (vector unsigned char,
5640 vector unsigned char,
5641 vector unsigned char);
5643 vector signed char vec_sl (vector signed char, vector unsigned char);
5644 vector unsigned char vec_sl (vector unsigned char,
5645 vector unsigned char);
5646 vector signed short vec_sl (vector signed short, vector unsigned short);
5648 vector unsigned short vec_sl (vector unsigned short,
5649 vector unsigned short);
5650 vector signed int vec_sl (vector signed int, vector unsigned int);
5651 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
5653 vector float vec_sld (vector float, vector float, const char);
5654 vector signed int vec_sld (vector signed int, vector signed int,
5656 vector unsigned int vec_sld (vector unsigned int, vector unsigned int,
5658 vector signed short vec_sld (vector signed short, vector signed short,
5660 vector unsigned short vec_sld (vector unsigned short,
5661 vector unsigned short, const char);
5662 vector signed char vec_sld (vector signed char, vector signed char,
5664 vector unsigned char vec_sld (vector unsigned char,
5665 vector unsigned char,
5668 vector signed int vec_sll (vector signed int, vector unsigned int);
5669 vector signed int vec_sll (vector signed int, vector unsigned short);
5670 vector signed int vec_sll (vector signed int, vector unsigned char);
5671 vector unsigned int vec_sll (vector unsigned int, vector unsigned int);
5672 vector unsigned int vec_sll (vector unsigned int,
5673 vector unsigned short);
5674 vector unsigned int vec_sll (vector unsigned int, vector unsigned char);
5676 vector signed short vec_sll (vector signed short, vector unsigned int);
5677 vector signed short vec_sll (vector signed short,
5678 vector unsigned short);
5679 vector signed short vec_sll (vector signed short, vector unsigned char);
5681 vector unsigned short vec_sll (vector unsigned short,
5682 vector unsigned int);
5683 vector unsigned short vec_sll (vector unsigned short,
5684 vector unsigned short);
5685 vector unsigned short vec_sll (vector unsigned short,
5686 vector unsigned char);
5687 vector signed char vec_sll (vector signed char, vector unsigned int);
5688 vector signed char vec_sll (vector signed char, vector unsigned short);
5689 vector signed char vec_sll (vector signed char, vector unsigned char);
5690 vector unsigned char vec_sll (vector unsigned char,
5691 vector unsigned int);
5692 vector unsigned char vec_sll (vector unsigned char,
5693 vector unsigned short);
5694 vector unsigned char vec_sll (vector unsigned char,
5695 vector unsigned char);
5697 vector float vec_slo (vector float, vector signed char);
5698 vector float vec_slo (vector float, vector unsigned char);
5699 vector signed int vec_slo (vector signed int, vector signed char);
5700 vector signed int vec_slo (vector signed int, vector unsigned char);
5701 vector unsigned int vec_slo (vector unsigned int, vector signed char);
5702 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
5704 vector signed short vec_slo (vector signed short, vector signed char);
5705 vector signed short vec_slo (vector signed short, vector unsigned char);
5707 vector unsigned short vec_slo (vector unsigned short,
5708 vector signed char);
5709 vector unsigned short vec_slo (vector unsigned short,
5710 vector unsigned char);
5711 vector signed char vec_slo (vector signed char, vector signed char);
5712 vector signed char vec_slo (vector signed char, vector unsigned char);
5713 vector unsigned char vec_slo (vector unsigned char, vector signed char);
5715 vector unsigned char vec_slo (vector unsigned char,
5716 vector unsigned char);
5718 vector signed char vec_splat (vector signed char, const char);
5719 vector unsigned char vec_splat (vector unsigned char, const char);
5720 vector signed short vec_splat (vector signed short, const char);
5721 vector unsigned short vec_splat (vector unsigned short, const char);
5722 vector float vec_splat (vector float, const char);
5723 vector signed int vec_splat (vector signed int, const char);
5724 vector unsigned int vec_splat (vector unsigned int, const char);
5726 vector signed char vec_splat_s8 (const char);
5728 vector signed short vec_splat_s16 (const char);
5730 vector signed int vec_splat_s32 (const char);
5732 vector unsigned char vec_splat_u8 (const char);
5734 vector unsigned short vec_splat_u16 (const char);
5736 vector unsigned int vec_splat_u32 (const char);
5738 vector signed char vec_sr (vector signed char, vector unsigned char);
5739 vector unsigned char vec_sr (vector unsigned char,
5740 vector unsigned char);
5741 vector signed short vec_sr (vector signed short, vector unsigned short);
5743 vector unsigned short vec_sr (vector unsigned short,
5744 vector unsigned short);
5745 vector signed int vec_sr (vector signed int, vector unsigned int);
5746 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
5748 vector signed char vec_sra (vector signed char, vector unsigned char);
5749 vector unsigned char vec_sra (vector unsigned char,
5750 vector unsigned char);
5751 vector signed short vec_sra (vector signed short,
5752 vector unsigned short);
5753 vector unsigned short vec_sra (vector unsigned short,
5754 vector unsigned short);
5755 vector signed int vec_sra (vector signed int, vector unsigned int);
5756 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
5758 vector signed int vec_srl (vector signed int, vector unsigned int);
5759 vector signed int vec_srl (vector signed int, vector unsigned short);
5760 vector signed int vec_srl (vector signed int, vector unsigned char);
5761 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
5762 vector unsigned int vec_srl (vector unsigned int,
5763 vector unsigned short);
5764 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
5766 vector signed short vec_srl (vector signed short, vector unsigned int);
5767 vector signed short vec_srl (vector signed short,
5768 vector unsigned short);
5769 vector signed short vec_srl (vector signed short, vector unsigned char);
5771 vector unsigned short vec_srl (vector unsigned short,
5772 vector unsigned int);
5773 vector unsigned short vec_srl (vector unsigned short,
5774 vector unsigned short);
5775 vector unsigned short vec_srl (vector unsigned short,
5776 vector unsigned char);
5777 vector signed char vec_srl (vector signed char, vector unsigned int);
5778 vector signed char vec_srl (vector signed char, vector unsigned short);
5779 vector signed char vec_srl (vector signed char, vector unsigned char);
5780 vector unsigned char vec_srl (vector unsigned char,
5781 vector unsigned int);
5782 vector unsigned char vec_srl (vector unsigned char,
5783 vector unsigned short);
5784 vector unsigned char vec_srl (vector unsigned char,
5785 vector unsigned char);
5787 vector float vec_sro (vector float, vector signed char);
5788 vector float vec_sro (vector float, vector unsigned char);
5789 vector signed int vec_sro (vector signed int, vector signed char);
5790 vector signed int vec_sro (vector signed int, vector unsigned char);
5791 vector unsigned int vec_sro (vector unsigned int, vector signed char);
5792 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
5794 vector signed short vec_sro (vector signed short, vector signed char);
5795 vector signed short vec_sro (vector signed short, vector unsigned char);
5797 vector unsigned short vec_sro (vector unsigned short,
5798 vector signed char);
5799 vector unsigned short vec_sro (vector unsigned short,
5800 vector unsigned char);
5801 vector signed char vec_sro (vector signed char, vector signed char);
5802 vector signed char vec_sro (vector signed char, vector unsigned char);
5803 vector unsigned char vec_sro (vector unsigned char, vector signed char);
5805 vector unsigned char vec_sro (vector unsigned char,
5806 vector unsigned char);
5808 void vec_st (vector float, int, float *);
5809 void vec_st (vector float, int, vector float *);
5810 void vec_st (vector signed int, int, int *);
5811 void vec_st (vector signed int, int, unsigned int *);
5812 void vec_st (vector unsigned int, int, unsigned int *);
5813 void vec_st (vector unsigned int, int, vector unsigned int *);
5814 void vec_st (vector signed short, int, short *);
5815 void vec_st (vector signed short, int, vector unsigned short *);
5816 void vec_st (vector signed short, int, vector signed short *);
5817 void vec_st (vector unsigned short, int, unsigned short *);
5818 void vec_st (vector unsigned short, int, vector unsigned short *);
5819 void vec_st (vector signed char, int, signed char *);
5820 void vec_st (vector signed char, int, unsigned char *);
5821 void vec_st (vector signed char, int, vector signed char *);
5822 void vec_st (vector unsigned char, int, unsigned char *);
5823 void vec_st (vector unsigned char, int, vector unsigned char *);
5825 void vec_ste (vector signed char, int, unsigned char *);
5826 void vec_ste (vector signed char, int, signed char *);
5827 void vec_ste (vector unsigned char, int, unsigned char *);
5828 void vec_ste (vector signed short, int, short *);
5829 void vec_ste (vector signed short, int, unsigned short *);
5830 void vec_ste (vector unsigned short, int, void *);
5831 void vec_ste (vector signed int, int, unsigned int *);
5832 void vec_ste (vector signed int, int, int *);
5833 void vec_ste (vector unsigned int, int, unsigned int *);
5834 void vec_ste (vector float, int, float *);
5836 void vec_stl (vector float, int, vector float *);
5837 void vec_stl (vector float, int, float *);
5838 void vec_stl (vector signed int, int, vector signed int *);
5839 void vec_stl (vector signed int, int, int *);
5840 void vec_stl (vector signed int, int, unsigned int *);
5841 void vec_stl (vector unsigned int, int, vector unsigned int *);
5842 void vec_stl (vector unsigned int, int, unsigned int *);
5843 void vec_stl (vector signed short, int, short *);
5844 void vec_stl (vector signed short, int, unsigned short *);
5845 void vec_stl (vector signed short, int, vector signed short *);
5846 void vec_stl (vector unsigned short, int, unsigned short *);
5847 void vec_stl (vector unsigned short, int, vector signed short *);
5848 void vec_stl (vector signed char, int, signed char *);
5849 void vec_stl (vector signed char, int, unsigned char *);
5850 void vec_stl (vector signed char, int, vector signed char *);
5851 void vec_stl (vector unsigned char, int, unsigned char *);
5852 void vec_stl (vector unsigned char, int, vector unsigned char *);
5854 vector signed char vec_sub (vector signed char, vector signed char);
5855 vector unsigned char vec_sub (vector signed char, vector unsigned char);
5857 vector unsigned char vec_sub (vector unsigned char, vector signed char);
5859 vector unsigned char vec_sub (vector unsigned char,
5860 vector unsigned char);
5861 vector signed short vec_sub (vector signed short, vector signed short);
5862 vector unsigned short vec_sub (vector signed short,
5863 vector unsigned short);
5864 vector unsigned short vec_sub (vector unsigned short,
5865 vector signed short);
5866 vector unsigned short vec_sub (vector unsigned short,
5867 vector unsigned short);
5868 vector signed int vec_sub (vector signed int, vector signed int);
5869 vector unsigned int vec_sub (vector signed int, vector unsigned int);
5870 vector unsigned int vec_sub (vector unsigned int, vector signed int);
5871 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
5872 vector float vec_sub (vector float, vector float);
5874 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
5876 vector unsigned char vec_subs (vector signed char,
5877 vector unsigned char);
5878 vector unsigned char vec_subs (vector unsigned char,
5879 vector signed char);
5880 vector unsigned char vec_subs (vector unsigned char,
5881 vector unsigned char);
5882 vector signed char vec_subs (vector signed char, vector signed char);
5883 vector unsigned short vec_subs (vector signed short,
5884 vector unsigned short);
5885 vector unsigned short vec_subs (vector unsigned short,
5886 vector signed short);
5887 vector unsigned short vec_subs (vector unsigned short,
5888 vector unsigned short);
5889 vector signed short vec_subs (vector signed short, vector signed short);
5891 vector unsigned int vec_subs (vector signed int, vector unsigned int);
5892 vector unsigned int vec_subs (vector unsigned int, vector signed int);
5893 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
5895 vector signed int vec_subs (vector signed int, vector signed int);
5897 vector unsigned int vec_sum4s (vector unsigned char,
5898 vector unsigned int);
5899 vector signed int vec_sum4s (vector signed char, vector signed int);
5900 vector signed int vec_sum4s (vector signed short, vector signed int);
5902 vector signed int vec_sum2s (vector signed int, vector signed int);
5904 vector signed int vec_sums (vector signed int, vector signed int);
5906 vector float vec_trunc (vector float);
5908 vector signed short vec_unpackh (vector signed char);
5909 vector unsigned int vec_unpackh (vector signed short);
5910 vector signed int vec_unpackh (vector signed short);
5912 vector signed short vec_unpackl (vector signed char);
5913 vector unsigned int vec_unpackl (vector signed short);
5914 vector signed int vec_unpackl (vector signed short);
5916 vector float vec_xor (vector float, vector float);
5917 vector float vec_xor (vector float, vector signed int);
5918 vector float vec_xor (vector signed int, vector float);
5919 vector signed int vec_xor (vector signed int, vector signed int);
5920 vector unsigned int vec_xor (vector signed int, vector unsigned int);
5921 vector unsigned int vec_xor (vector unsigned int, vector signed int);
5922 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
5923 vector signed short vec_xor (vector signed short, vector signed short);
5924 vector unsigned short vec_xor (vector signed short,
5925 vector unsigned short);
5926 vector unsigned short vec_xor (vector unsigned short,
5927 vector signed short);
5928 vector unsigned short vec_xor (vector unsigned short,
5929 vector unsigned short);
5930 vector signed char vec_xor (vector signed char, vector signed char);
5931 vector unsigned char vec_xor (vector signed char, vector unsigned char);
5933 vector unsigned char vec_xor (vector unsigned char, vector signed char);
5935 vector unsigned char vec_xor (vector unsigned char,
5936 vector unsigned char);
5938 vector signed int vec_all_eq (vector signed char, vector unsigned char);
5940 vector signed int vec_all_eq (vector signed char, vector signed char);
5941 vector signed int vec_all_eq (vector unsigned char, vector signed char);
5943 vector signed int vec_all_eq (vector unsigned char,
5944 vector unsigned char);
5945 vector signed int vec_all_eq (vector signed short,
5946 vector unsigned short);
5947 vector signed int vec_all_eq (vector signed short, vector signed short);
5949 vector signed int vec_all_eq (vector unsigned short,
5950 vector signed short);
5951 vector signed int vec_all_eq (vector unsigned short,
5952 vector unsigned short);
5953 vector signed int vec_all_eq (vector signed int, vector unsigned int);
5954 vector signed int vec_all_eq (vector signed int, vector signed int);
5955 vector signed int vec_all_eq (vector unsigned int, vector signed int);
5956 vector signed int vec_all_eq (vector unsigned int, vector unsigned int);
5958 vector signed int vec_all_eq (vector float, vector float);
5960 vector signed int vec_all_ge (vector signed char, vector unsigned char);
5962 vector signed int vec_all_ge (vector unsigned char, vector signed char);
5964 vector signed int vec_all_ge (vector unsigned char,
5965 vector unsigned char);
5966 vector signed int vec_all_ge (vector signed char, vector signed char);
5967 vector signed int vec_all_ge (vector signed short,
5968 vector unsigned short);
5969 vector signed int vec_all_ge (vector unsigned short,
5970 vector signed short);
5971 vector signed int vec_all_ge (vector unsigned short,
5972 vector unsigned short);
5973 vector signed int vec_all_ge (vector signed short, vector signed short);
5975 vector signed int vec_all_ge (vector signed int, vector unsigned int);
5976 vector signed int vec_all_ge (vector unsigned int, vector signed int);
5977 vector signed int vec_all_ge (vector unsigned int, vector unsigned int);
5979 vector signed int vec_all_ge (vector signed int, vector signed int);
5980 vector signed int vec_all_ge (vector float, vector float);
5982 vector signed int vec_all_gt (vector signed char, vector unsigned char);
5984 vector signed int vec_all_gt (vector unsigned char, vector signed char);
5986 vector signed int vec_all_gt (vector unsigned char,
5987 vector unsigned char);
5988 vector signed int vec_all_gt (vector signed char, vector signed char);
5989 vector signed int vec_all_gt (vector signed short,
5990 vector unsigned short);
5991 vector signed int vec_all_gt (vector unsigned short,
5992 vector signed short);
5993 vector signed int vec_all_gt (vector unsigned short,
5994 vector unsigned short);
5995 vector signed int vec_all_gt (vector signed short, vector signed short);
5997 vector signed int vec_all_gt (vector signed int, vector unsigned int);
5998 vector signed int vec_all_gt (vector unsigned int, vector signed int);
5999 vector signed int vec_all_gt (vector unsigned int, vector unsigned int);
6001 vector signed int vec_all_gt (vector signed int, vector signed int);
6002 vector signed int vec_all_gt (vector float, vector float);
6004 vector signed int vec_all_in (vector float, vector float);
6006 vector signed int vec_all_le (vector signed char, vector unsigned char);
6008 vector signed int vec_all_le (vector unsigned char, vector signed char);
6010 vector signed int vec_all_le (vector unsigned char,
6011 vector unsigned char);
6012 vector signed int vec_all_le (vector signed char, vector signed char);
6013 vector signed int vec_all_le (vector signed short,
6014 vector unsigned short);
6015 vector signed int vec_all_le (vector unsigned short,
6016 vector signed short);
6017 vector signed int vec_all_le (vector unsigned short,
6018 vector unsigned short);
6019 vector signed int vec_all_le (vector signed short, vector signed short);
6021 vector signed int vec_all_le (vector signed int, vector unsigned int);
6022 vector signed int vec_all_le (vector unsigned int, vector signed int);
6023 vector signed int vec_all_le (vector unsigned int, vector unsigned int);
6025 vector signed int vec_all_le (vector signed int, vector signed int);
6026 vector signed int vec_all_le (vector float, vector float);
6028 vector signed int vec_all_lt (vector signed char, vector unsigned char);
6030 vector signed int vec_all_lt (vector unsigned char, vector signed char);
6032 vector signed int vec_all_lt (vector unsigned char,
6033 vector unsigned char);
6034 vector signed int vec_all_lt (vector signed char, vector signed char);
6035 vector signed int vec_all_lt (vector signed short,
6036 vector unsigned short);
6037 vector signed int vec_all_lt (vector unsigned short,
6038 vector signed short);
6039 vector signed int vec_all_lt (vector unsigned short,
6040 vector unsigned short);
6041 vector signed int vec_all_lt (vector signed short, vector signed short);
6043 vector signed int vec_all_lt (vector signed int, vector unsigned int);
6044 vector signed int vec_all_lt (vector unsigned int, vector signed int);
6045 vector signed int vec_all_lt (vector unsigned int, vector unsigned int);
6047 vector signed int vec_all_lt (vector signed int, vector signed int);
6048 vector signed int vec_all_lt (vector float, vector float);
6050 vector signed int vec_all_nan (vector float);
6052 vector signed int vec_all_ne (vector signed char, vector unsigned char);
6054 vector signed int vec_all_ne (vector signed char, vector signed char);
6055 vector signed int vec_all_ne (vector unsigned char, vector signed char);
6057 vector signed int vec_all_ne (vector unsigned char,
6058 vector unsigned char);
6059 vector signed int vec_all_ne (vector signed short,
6060 vector unsigned short);
6061 vector signed int vec_all_ne (vector signed short, vector signed short);
6063 vector signed int vec_all_ne (vector unsigned short,
6064 vector signed short);
6065 vector signed int vec_all_ne (vector unsigned short,
6066 vector unsigned short);
6067 vector signed int vec_all_ne (vector signed int, vector unsigned int);
6068 vector signed int vec_all_ne (vector signed int, vector signed int);
6069 vector signed int vec_all_ne (vector unsigned int, vector signed int);
6070 vector signed int vec_all_ne (vector unsigned int, vector unsigned int);
6072 vector signed int vec_all_ne (vector float, vector float);
6074 vector signed int vec_all_nge (vector float, vector float);
6076 vector signed int vec_all_ngt (vector float, vector float);
6078 vector signed int vec_all_nle (vector float, vector float);
6080 vector signed int vec_all_nlt (vector float, vector float);
6082 vector signed int vec_all_numeric (vector float);
6084 vector signed int vec_any_eq (vector signed char, vector unsigned char);
6086 vector signed int vec_any_eq (vector signed char, vector signed char);
6087 vector signed int vec_any_eq (vector unsigned char, vector signed char);
6089 vector signed int vec_any_eq (vector unsigned char,
6090 vector unsigned char);
6091 vector signed int vec_any_eq (vector signed short,
6092 vector unsigned short);
6093 vector signed int vec_any_eq (vector signed short, vector signed short);
6095 vector signed int vec_any_eq (vector unsigned short,
6096 vector signed short);
6097 vector signed int vec_any_eq (vector unsigned short,
6098 vector unsigned short);
6099 vector signed int vec_any_eq (vector signed int, vector unsigned int);
6100 vector signed int vec_any_eq (vector signed int, vector signed int);
6101 vector signed int vec_any_eq (vector unsigned int, vector signed int);
6102 vector signed int vec_any_eq (vector unsigned int, vector unsigned int);
6104 vector signed int vec_any_eq (vector float, vector float);
6106 vector signed int vec_any_ge (vector signed char, vector unsigned char);
6108 vector signed int vec_any_ge (vector unsigned char, vector signed char);
6110 vector signed int vec_any_ge (vector unsigned char,
6111 vector unsigned char);
6112 vector signed int vec_any_ge (vector signed char, vector signed char);
6113 vector signed int vec_any_ge (vector signed short,
6114 vector unsigned short);
6115 vector signed int vec_any_ge (vector unsigned short,
6116 vector signed short);
6117 vector signed int vec_any_ge (vector unsigned short,
6118 vector unsigned short);
6119 vector signed int vec_any_ge (vector signed short, vector signed short);
6121 vector signed int vec_any_ge (vector signed int, vector unsigned int);
6122 vector signed int vec_any_ge (vector unsigned int, vector signed int);
6123 vector signed int vec_any_ge (vector unsigned int, vector unsigned int);
6125 vector signed int vec_any_ge (vector signed int, vector signed int);
6126 vector signed int vec_any_ge (vector float, vector float);
6128 vector signed int vec_any_gt (vector signed char, vector unsigned char);
6130 vector signed int vec_any_gt (vector unsigned char, vector signed char);
6132 vector signed int vec_any_gt (vector unsigned char,
6133 vector unsigned char);
6134 vector signed int vec_any_gt (vector signed char, vector signed char);
6135 vector signed int vec_any_gt (vector signed short,
6136 vector unsigned short);
6137 vector signed int vec_any_gt (vector unsigned short,
6138 vector signed short);
6139 vector signed int vec_any_gt (vector unsigned short,
6140 vector unsigned short);
6141 vector signed int vec_any_gt (vector signed short, vector signed short);
6143 vector signed int vec_any_gt (vector signed int, vector unsigned int);
6144 vector signed int vec_any_gt (vector unsigned int, vector signed int);
6145 vector signed int vec_any_gt (vector unsigned int, vector unsigned int);
6147 vector signed int vec_any_gt (vector signed int, vector signed int);
6148 vector signed int vec_any_gt (vector float, vector float);
6150 vector signed int vec_any_le (vector signed char, vector unsigned char);
6152 vector signed int vec_any_le (vector unsigned char, vector signed char);
6154 vector signed int vec_any_le (vector unsigned char,
6155 vector unsigned char);
6156 vector signed int vec_any_le (vector signed char, vector signed char);
6157 vector signed int vec_any_le (vector signed short,
6158 vector unsigned short);
6159 vector signed int vec_any_le (vector unsigned short,
6160 vector signed short);
6161 vector signed int vec_any_le (vector unsigned short,
6162 vector unsigned short);
6163 vector signed int vec_any_le (vector signed short, vector signed short);
6165 vector signed int vec_any_le (vector signed int, vector unsigned int);
6166 vector signed int vec_any_le (vector unsigned int, vector signed int);
6167 vector signed int vec_any_le (vector unsigned int, vector unsigned int);
6169 vector signed int vec_any_le (vector signed int, vector signed int);
6170 vector signed int vec_any_le (vector float, vector float);
6172 vector signed int vec_any_lt (vector signed char, vector unsigned char);
6174 vector signed int vec_any_lt (vector unsigned char, vector signed char);
6176 vector signed int vec_any_lt (vector unsigned char,
6177 vector unsigned char);
6178 vector signed int vec_any_lt (vector signed char, vector signed char);
6179 vector signed int vec_any_lt (vector signed short,
6180 vector unsigned short);
6181 vector signed int vec_any_lt (vector unsigned short,
6182 vector signed short);
6183 vector signed int vec_any_lt (vector unsigned short,
6184 vector unsigned short);
6185 vector signed int vec_any_lt (vector signed short, vector signed short);
6187 vector signed int vec_any_lt (vector signed int, vector unsigned int);
6188 vector signed int vec_any_lt (vector unsigned int, vector signed int);
6189 vector signed int vec_any_lt (vector unsigned int, vector unsigned int);
6191 vector signed int vec_any_lt (vector signed int, vector signed int);
6192 vector signed int vec_any_lt (vector float, vector float);
6194 vector signed int vec_any_nan (vector float);
6196 vector signed int vec_any_ne (vector signed char, vector unsigned char);
6198 vector signed int vec_any_ne (vector signed char, vector signed char);
6199 vector signed int vec_any_ne (vector unsigned char, vector signed char);
6201 vector signed int vec_any_ne (vector unsigned char,
6202 vector unsigned char);
6203 vector signed int vec_any_ne (vector signed short,
6204 vector unsigned short);
6205 vector signed int vec_any_ne (vector signed short, vector signed short);
6207 vector signed int vec_any_ne (vector unsigned short,
6208 vector signed short);
6209 vector signed int vec_any_ne (vector unsigned short,
6210 vector unsigned short);
6211 vector signed int vec_any_ne (vector signed int, vector unsigned int);
6212 vector signed int vec_any_ne (vector signed int, vector signed int);
6213 vector signed int vec_any_ne (vector unsigned int, vector signed int);
6214 vector signed int vec_any_ne (vector unsigned int, vector unsigned int);
6216 vector signed int vec_any_ne (vector float, vector float);
6218 vector signed int vec_any_nge (vector float, vector float);
6220 vector signed int vec_any_ngt (vector float, vector float);
6222 vector signed int vec_any_nle (vector float, vector float);
6224 vector signed int vec_any_nlt (vector float, vector float);
6226 vector signed int vec_any_numeric (vector float);
6228 vector signed int vec_any_out (vector float, vector float);
6232 @section Pragmas Accepted by GCC
6236 GCC supports several types of pragmas, primarily in order to compile
6237 code originally written for other compilers. Note that in general
6238 we do not recommend the use of pragmas; @xref{Function Attributes},
6239 for further explanation.
6243 * RS/6000 and PowerPC Pragmas::
6250 @subsection ARM Pragmas
6252 The ARM target defines pragmas for controlling the default addition of
6253 @code{long_call} and @code{short_call} attributes to functions.
6254 @xref{Function Attributes}, for information about the effects of these
6259 @cindex pragma, long_calls
6260 Set all subsequent functions to have the @code{long_call} attribute.
6263 @cindex pragma, no_long_calls
6264 Set all subsequent functions to have the @code{short_call} attribute.
6266 @item long_calls_off
6267 @cindex pragma, long_calls_off
6268 Do not affect the @code{long_call} or @code{short_call} attributes of
6269 subsequent functions.
6272 @node RS/6000 and PowerPC Pragmas
6273 @subsection RS/6000 and PowerPC Pragmas
6275 The RS/6000 and PowerPC targets define one pragma for controlling
6276 whether or not the @code{longcall} attribute is added to function
6277 declarations by default. This pragma overrides the @option{-mlongcall}
6278 option, but not the @code{longcall} and @code{shortcall} attributes.
6279 @xref{RS/6000 and PowerPC Options}, for more information about when long
6280 calls are and are not necessary.
6284 @cindex pragma, longcall
6285 Apply the @code{longcall} attribute to all subsequent function
6289 Do not apply the @code{longcall} attribute to subsequent function
6293 @c Describe c4x pragmas here.
6294 @c Describe h8300 pragmas here.
6295 @c Describe i370 pragmas here.
6296 @c Describe i960 pragmas here.
6297 @c Describe sh pragmas here.
6298 @c Describe v850 pragmas here.
6300 @node Darwin Pragmas
6301 @subsection Darwin Pragmas
6303 The following pragmas are available for all architectures running the
6304 Darwin operating system. These are useful for compatibility with other
6308 @item mark @var{tokens}@dots{}
6309 @cindex pragma, mark
6310 This pragma is accepted, but has no effect.
6312 @item options align=@var{alignment}
6313 @cindex pragma, options align
6314 This pragma sets the alignment of fields in structures. The values of
6315 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
6316 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
6317 properly; to restore the previous setting, use @code{reset} for the
6320 @item segment @var{tokens}@dots{}
6321 @cindex pragma, segment
6322 This pragma is accepted, but has no effect.
6324 @item unused (@var{var} [, @var{var}]@dots{})
6325 @cindex pragma, unused
6326 This pragma declares variables to be possibly unused. GCC will not
6327 produce warnings for the listed variables. The effect is similar to
6328 that of the @code{unused} attribute, except that this pragma may appear
6329 anywhere within the variables' scopes.
6332 @node Solaris Pragmas
6333 @subsection Solaris Pragmas
6335 For compatibility with the SunPRO compiler, the following pragma
6339 @item redefine_extname @var{oldname} @var{newname}
6340 @cindex pragma, redefine_extname
6342 This pragma gives the C function @var{oldname} the assembler label
6343 @var{newname}. The pragma must appear before the function declaration.
6344 This pragma is equivalent to the asm labels extension (@pxref{Asm
6345 Labels}). The preprocessor defines @code{__PRAGMA_REDEFINE_EXTNAME}
6346 if the pragma is available.
6350 @subsection Tru64 Pragmas
6352 For compatibility with the Compaq C compiler, the following pragma
6356 @item extern_prefix @var{string}
6357 @cindex pragma, extern_prefix
6359 This pragma renames all subsequent function and variable declarations
6360 such that @var{string} is prepended to the name. This effect may be
6361 terminated by using another @code{extern_prefix} pragma with the
6364 This pragma is similar in intent to to the asm labels extension
6365 (@pxref{Asm Labels}) in that the system programmer wants to change
6366 the assembly-level ABI without changing the source-level API. The
6367 preprocessor defines @code{__EXTERN_PREFIX} if the pragma is available.
6370 @node Unnamed Fields
6371 @section Unnamed struct/union fields within structs/unions.
6375 For compatibility with other compilers, GCC allows you to define
6376 a structure or union that contains, as fields, structures and unions
6377 without names. For example:
6390 In this example, the user would be able to access members of the unnamed
6391 union with code like @samp{foo.b}. Note that only unnamed structs and
6392 unions are allowed, you may not have, for example, an unnamed
6395 You must never create such structures that cause ambiguous field definitions.
6396 For example, this structure:
6407 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
6408 Such constructs are not supported and must be avoided. In the future,
6409 such constructs may be detected and treated as compilation errors.
6412 @section Thread-Local Storage
6413 @cindex Thread-Local Storage
6414 @cindex @acronym{TLS}
6417 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
6418 are allocated such that there is one instance of the variable per extant
6419 thread. The run-time model GCC uses to implement this originates
6420 in the IA-64 processor-specific ABI, but has since been migrated
6421 to other processors as well. It requires significant support from
6422 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
6423 system libraries (@file{libc.so} and @file{libpthread.so}), so it
6424 is not available everywhere.
6426 At the user level, the extension is visible with a new storage
6427 class keyword: @code{__thread}. For example:
6431 extern __thread struct state s;
6432 static __thread char *p;
6435 The @code{__thread} specifier may be used alone, with the @code{extern}
6436 or @code{static} specifiers, but with no other storage class specifier.
6437 When used with @code{extern} or @code{static}, @code{__thread} must appear
6438 immediately after the other storage class specifier.
6440 The @code{__thread} specifier may be applied to any global, file-scoped
6441 static, function-scoped static, or static data member of a class. It may
6442 not be applied to block-scoped automatic or non-static data member.
6444 When the address-of operator is applied to a thread-local variable, it is
6445 evaluated at run-time and returns the address of the current thread's
6446 instance of that variable. An address so obtained may be used by any
6447 thread. When a thread terminates, any pointers to thread-local variables
6448 in that thread become invalid.
6450 No static initialization may refer to the address of a thread-local variable.
6452 In C++, if an initializer is present for a thread-local variable, it must
6453 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
6456 See @uref{http://people.redhat.com/drepper/tls.pdf,
6457 ELF Handling For Thread-Local Storage} for a detailed explanation of
6458 the four thread-local storage addressing models, and how the run-time
6459 is expected to function.
6462 * C99 Thread-Local Edits::
6463 * C++98 Thread-Local Edits::
6466 @node C99 Thread-Local Edits
6467 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
6469 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
6470 that document the exact semantics of the language extension.
6474 @cite{5.1.2 Execution environments}
6476 Add new text after paragraph 1
6479 Within either execution environment, a @dfn{thread} is a flow of
6480 control within a program. It is implementation defined whether
6481 or not there may be more than one thread associated with a program.
6482 It is implementation defined how threads beyond the first are
6483 created, the name and type of the function called at thread
6484 startup, and how threads may be terminated. However, objects
6485 with thread storage duration shall be initialized before thread
6490 @cite{6.2.4 Storage durations of objects}
6492 Add new text before paragraph 3
6495 An object whose identifier is declared with the storage-class
6496 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
6497 Its lifetime is the entire execution of the thread, and its
6498 stored value is initialized only once, prior to thread startup.
6502 @cite{6.4.1 Keywords}
6504 Add @code{__thread}.
6507 @cite{6.7.1 Storage-class specifiers}
6509 Add @code{__thread} to the list of storage class specifiers in
6512 Change paragraph 2 to
6515 With the exception of @code{__thread}, at most one storage-class
6516 specifier may be given [@dots{}]. The @code{__thread} specifier may
6517 be used alone, or immediately following @code{extern} or
6521 Add new text after paragraph 6
6524 The declaration of an identifier for a variable that has
6525 block scope that specifies @code{__thread} shall also
6526 specify either @code{extern} or @code{static}.
6528 The @code{__thread} specifier shall be used only with
6533 @node C++98 Thread-Local Edits
6534 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
6536 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
6537 that document the exact semantics of the language extension.
6540 @b{[intro.execution]}
6542 New text after paragraph 4
6545 A @dfn{thread} is a flow of control within the abstract machine.
6546 It is implementation defined whether or not there may be more than
6550 New text after paragraph 7
6553 It is unspecified whether additional action must be taken to
6554 ensure when and whether side effects are visible to other threads.
6560 Add @code{__thread}.
6563 @b{[basic.start.main]}
6565 Add after paragraph 5
6568 The thread that begins execution at the @code{main} function is called
6569 the @dfn{main thread}. It is implementation defined how functions
6570 beginning threads other than the main thread are designated or typed.
6571 A function so designated, as well as the @code{main} function, is called
6572 a @dfn{thread startup function}. It is implementation defined what
6573 happens if a thread startup function returns. It is implementation
6574 defined what happens to other threads when any thread calls @code{exit}.
6578 @b{[basic.start.init]}
6580 Add after paragraph 4
6583 The storage for an object of thread storage duration shall be
6584 staticly initialized before the first statement of the thread startup
6585 function. An object of thread storage duration shall not require
6586 dynamic initialization.
6590 @b{[basic.start.term]}
6592 Add after paragraph 3
6595 The type of an object with thread storage duration shall not have a
6596 non-trivial destructor, nor shall it be an array type whose elements
6597 (directly or indirectly) have non-trivial destructors.
6603 Add ``thread storage duration'' to the list in paragraph 1.
6608 Thread, static, and automatic storage durations are associated with
6609 objects introduced by declarations [@dots{}].
6612 Add @code{__thread} to the list of specifiers in paragraph 3.
6615 @b{[basic.stc.thread]}
6617 New section before @b{[basic.stc.static]}
6620 The keyword @code{__thread} applied to an non-local object gives the
6621 object thread storage duration.
6623 A local variable or class data member declared both @code{static}
6624 and @code{__thread} gives the variable or member thread storage
6629 @b{[basic.stc.static]}
6634 All objects which have neither thread storage duration, dynamic
6635 storage duration nor are local [@dots{}].
6641 Add @code{__thread} to the list in paragraph 1.
6646 With the exception of @code{__thread}, at most one
6647 @var{storage-class-specifier} shall appear in a given
6648 @var{decl-specifier-seq}. The @code{__thread} specifier may
6649 be used alone, or immediately following the @code{extern} or
6650 @code{static} specifiers. [@dots{}]
6653 Add after paragraph 5
6656 The @code{__thread} specifier can be applied only to the names of objects
6657 and to anonymous unions.
6663 Add after paragraph 6
6666 Non-@code{static} members shall not be @code{__thread}.
6670 @node C++ Extensions
6671 @chapter Extensions to the C++ Language
6672 @cindex extensions, C++ language
6673 @cindex C++ language extensions
6675 The GNU compiler provides these extensions to the C++ language (and you
6676 can also use most of the C language extensions in your C++ programs). If you
6677 want to write code that checks whether these features are available, you can
6678 test for the GNU compiler the same way as for C programs: check for a
6679 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
6680 test specifically for GNU C++ (@pxref{Standard Predefined,,Standard
6681 Predefined Macros,cpp.info,The C Preprocessor}).
6684 * Min and Max:: C++ Minimum and maximum operators.
6685 * Volatiles:: What constitutes an access to a volatile object.
6686 * Restricted Pointers:: C99 restricted pointers and references.
6687 * Vague Linkage:: Where G++ puts inlines, vtables and such.
6688 * C++ Interface:: You can use a single C++ header file for both
6689 declarations and definitions.
6690 * Template Instantiation:: Methods for ensuring that exactly one copy of
6691 each needed template instantiation is emitted.
6692 * Bound member functions:: You can extract a function pointer to the
6693 method denoted by a @samp{->*} or @samp{.*} expression.
6694 * C++ Attributes:: Variable, function, and type attributes for C++ only.
6695 * Java Exceptions:: Tweaking exception handling to work with Java.
6696 * Deprecated Features:: Things might disappear from g++.
6697 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
6701 @section Minimum and Maximum Operators in C++
6703 It is very convenient to have operators which return the ``minimum'' or the
6704 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
6707 @item @var{a} <? @var{b}
6709 @cindex minimum operator
6710 is the @dfn{minimum}, returning the smaller of the numeric values
6711 @var{a} and @var{b};
6713 @item @var{a} >? @var{b}
6715 @cindex maximum operator
6716 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
6720 These operations are not primitive in ordinary C++, since you can
6721 use a macro to return the minimum of two things in C++, as in the
6725 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
6729 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
6730 the minimum value of variables @var{i} and @var{j}.
6732 However, side effects in @code{X} or @code{Y} may cause unintended
6733 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
6734 the smaller counter twice. A GNU C extension allows you to write safe
6735 macros that avoid this kind of problem (@pxref{Naming Types,,Naming an
6736 Expression's Type}). However, writing @code{MIN} and @code{MAX} as
6737 macros also forces you to use function-call notation for a
6738 fundamental arithmetic operation. Using GNU C++ extensions, you can
6739 write @w{@samp{int min = i <? j;}} instead.
6741 Since @code{<?} and @code{>?} are built into the compiler, they properly
6742 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
6746 @section When is a Volatile Object Accessed?
6747 @cindex accessing volatiles
6748 @cindex volatile read
6749 @cindex volatile write
6750 @cindex volatile access
6752 Both the C and C++ standard have the concept of volatile objects. These
6753 are normally accessed by pointers and used for accessing hardware. The
6754 standards encourage compilers to refrain from optimizations
6755 concerning accesses to volatile objects that it might perform on
6756 non-volatile objects. The C standard leaves it implementation defined
6757 as to what constitutes a volatile access. The C++ standard omits to
6758 specify this, except to say that C++ should behave in a similar manner
6759 to C with respect to volatiles, where possible. The minimum either
6760 standard specifies is that at a sequence point all previous accesses to
6761 volatile objects have stabilized and no subsequent accesses have
6762 occurred. Thus an implementation is free to reorder and combine
6763 volatile accesses which occur between sequence points, but cannot do so
6764 for accesses across a sequence point. The use of volatiles does not
6765 allow you to violate the restriction on updating objects multiple times
6766 within a sequence point.
6768 In most expressions, it is intuitively obvious what is a read and what is
6769 a write. For instance
6772 volatile int *dst = @var{somevalue};
6773 volatile int *src = @var{someothervalue};
6778 will cause a read of the volatile object pointed to by @var{src} and stores the
6779 value into the volatile object pointed to by @var{dst}. There is no
6780 guarantee that these reads and writes are atomic, especially for objects
6781 larger than @code{int}.
6783 Less obvious expressions are where something which looks like an access
6784 is used in a void context. An example would be,
6787 volatile int *src = @var{somevalue};
6791 With C, such expressions are rvalues, and as rvalues cause a read of
6792 the object, GCC interprets this as a read of the volatile being pointed
6793 to. The C++ standard specifies that such expressions do not undergo
6794 lvalue to rvalue conversion, and that the type of the dereferenced
6795 object may be incomplete. The C++ standard does not specify explicitly
6796 that it is this lvalue to rvalue conversion which is responsible for
6797 causing an access. However, there is reason to believe that it is,
6798 because otherwise certain simple expressions become undefined. However,
6799 because it would surprise most programmers, G++ treats dereferencing a
6800 pointer to volatile object of complete type in a void context as a read
6801 of the object. When the object has incomplete type, G++ issues a
6806 struct T @{int m;@};
6807 volatile S *ptr1 = @var{somevalue};
6808 volatile T *ptr2 = @var{somevalue};
6813 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
6814 causes a read of the object pointed to. If you wish to force an error on
6815 the first case, you must force a conversion to rvalue with, for instance
6816 a static cast, @code{static_cast<S>(*ptr1)}.
6818 When using a reference to volatile, G++ does not treat equivalent
6819 expressions as accesses to volatiles, but instead issues a warning that
6820 no volatile is accessed. The rationale for this is that otherwise it
6821 becomes difficult to determine where volatile access occur, and not
6822 possible to ignore the return value from functions returning volatile
6823 references. Again, if you wish to force a read, cast the reference to
6826 @node Restricted Pointers
6827 @section Restricting Pointer Aliasing
6828 @cindex restricted pointers
6829 @cindex restricted references
6830 @cindex restricted this pointer
6832 As with gcc, g++ understands the C99 feature of restricted pointers,
6833 specified with the @code{__restrict__}, or @code{__restrict} type
6834 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
6835 language flag, @code{restrict} is not a keyword in C++.
6837 In addition to allowing restricted pointers, you can specify restricted
6838 references, which indicate that the reference is not aliased in the local
6842 void fn (int *__restrict__ rptr, int &__restrict__ rref)
6849 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
6850 @var{rref} refers to a (different) unaliased integer.
6852 You may also specify whether a member function's @var{this} pointer is
6853 unaliased by using @code{__restrict__} as a member function qualifier.
6856 void T::fn () __restrict__
6863 Within the body of @code{T::fn}, @var{this} will have the effective
6864 definition @code{T *__restrict__ const this}. Notice that the
6865 interpretation of a @code{__restrict__} member function qualifier is
6866 different to that of @code{const} or @code{volatile} qualifier, in that it
6867 is applied to the pointer rather than the object. This is consistent with
6868 other compilers which implement restricted pointers.
6870 As with all outermost parameter qualifiers, @code{__restrict__} is
6871 ignored in function definition matching. This means you only need to
6872 specify @code{__restrict__} in a function definition, rather than
6873 in a function prototype as well.
6876 @section Vague Linkage
6877 @cindex vague linkage
6879 There are several constructs in C++ which require space in the object
6880 file but are not clearly tied to a single translation unit. We say that
6881 these constructs have ``vague linkage''. Typically such constructs are
6882 emitted wherever they are needed, though sometimes we can be more
6886 @item Inline Functions
6887 Inline functions are typically defined in a header file which can be
6888 included in many different compilations. Hopefully they can usually be
6889 inlined, but sometimes an out-of-line copy is necessary, if the address
6890 of the function is taken or if inlining fails. In general, we emit an
6891 out-of-line copy in all translation units where one is needed. As an
6892 exception, we only emit inline virtual functions with the vtable, since
6893 it will always require a copy.
6895 Local static variables and string constants used in an inline function
6896 are also considered to have vague linkage, since they must be shared
6897 between all inlined and out-of-line instances of the function.
6901 C++ virtual functions are implemented in most compilers using a lookup
6902 table, known as a vtable. The vtable contains pointers to the virtual
6903 functions provided by a class, and each object of the class contains a
6904 pointer to its vtable (or vtables, in some multiple-inheritance
6905 situations). If the class declares any non-inline, non-pure virtual
6906 functions, the first one is chosen as the ``key method'' for the class,
6907 and the vtable is only emitted in the translation unit where the key
6910 @emph{Note:} If the chosen key method is later defined as inline, the
6911 vtable will still be emitted in every translation unit which defines it.
6912 Make sure that any inline virtuals are declared inline in the class
6913 body, even if they are not defined there.
6915 @item type_info objects
6918 C++ requires information about types to be written out in order to
6919 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
6920 For polymorphic classes (classes with virtual functions), the type_info
6921 object is written out along with the vtable so that @samp{dynamic_cast}
6922 can determine the dynamic type of a class object at runtime. For all
6923 other types, we write out the type_info object when it is used: when
6924 applying @samp{typeid} to an expression, throwing an object, or
6925 referring to a type in a catch clause or exception specification.
6927 @item Template Instantiations
6928 Most everything in this section also applies to template instantiations,
6929 but there are other options as well.
6930 @xref{Template Instantiation,,Where's the Template?}.
6934 When used with GNU ld version 2.8 or later on an ELF system such as
6935 Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of
6936 these constructs will be discarded at link time. This is known as
6939 On targets that don't support COMDAT, but do support weak symbols, GCC
6940 will use them. This way one copy will override all the others, but
6941 the unused copies will still take up space in the executable.
6943 For targets which do not support either COMDAT or weak symbols,
6944 most entities with vague linkage will be emitted as local symbols to
6945 avoid duplicate definition errors from the linker. This will not happen
6946 for local statics in inlines, however, as having multiple copies will
6947 almost certainly break things.
6949 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
6950 another way to control placement of these constructs.
6953 @section Declarations and Definitions in One Header
6955 @cindex interface and implementation headers, C++
6956 @cindex C++ interface and implementation headers
6957 C++ object definitions can be quite complex. In principle, your source
6958 code will need two kinds of things for each object that you use across
6959 more than one source file. First, you need an @dfn{interface}
6960 specification, describing its structure with type declarations and
6961 function prototypes. Second, you need the @dfn{implementation} itself.
6962 It can be tedious to maintain a separate interface description in a
6963 header file, in parallel to the actual implementation. It is also
6964 dangerous, since separate interface and implementation definitions may
6965 not remain parallel.
6967 @cindex pragmas, interface and implementation
6968 With GNU C++, you can use a single header file for both purposes.
6971 @emph{Warning:} The mechanism to specify this is in transition. For the
6972 nonce, you must use one of two @code{#pragma} commands; in a future
6973 release of GNU C++, an alternative mechanism will make these
6974 @code{#pragma} commands unnecessary.
6977 The header file contains the full definitions, but is marked with
6978 @samp{#pragma interface} in the source code. This allows the compiler
6979 to use the header file only as an interface specification when ordinary
6980 source files incorporate it with @code{#include}. In the single source
6981 file where the full implementation belongs, you can use either a naming
6982 convention or @samp{#pragma implementation} to indicate this alternate
6983 use of the header file.
6986 @item #pragma interface
6987 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
6988 @kindex #pragma interface
6989 Use this directive in @emph{header files} that define object classes, to save
6990 space in most of the object files that use those classes. Normally,
6991 local copies of certain information (backup copies of inline member
6992 functions, debugging information, and the internal tables that implement
6993 virtual functions) must be kept in each object file that includes class
6994 definitions. You can use this pragma to avoid such duplication. When a
6995 header file containing @samp{#pragma interface} is included in a
6996 compilation, this auxiliary information will not be generated (unless
6997 the main input source file itself uses @samp{#pragma implementation}).
6998 Instead, the object files will contain references to be resolved at link
7001 The second form of this directive is useful for the case where you have
7002 multiple headers with the same name in different directories. If you
7003 use this form, you must specify the same string to @samp{#pragma
7006 @item #pragma implementation
7007 @itemx #pragma implementation "@var{objects}.h"
7008 @kindex #pragma implementation
7009 Use this pragma in a @emph{main input file}, when you want full output from
7010 included header files to be generated (and made globally visible). The
7011 included header file, in turn, should use @samp{#pragma interface}.
7012 Backup copies of inline member functions, debugging information, and the
7013 internal tables used to implement virtual functions are all generated in
7014 implementation files.
7016 @cindex implied @code{#pragma implementation}
7017 @cindex @code{#pragma implementation}, implied
7018 @cindex naming convention, implementation headers
7019 If you use @samp{#pragma implementation} with no argument, it applies to
7020 an include file with the same basename@footnote{A file's @dfn{basename}
7021 was the name stripped of all leading path information and of trailing
7022 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
7023 file. For example, in @file{allclass.cc}, giving just
7024 @samp{#pragma implementation}
7025 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
7027 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
7028 an implementation file whenever you would include it from
7029 @file{allclass.cc} even if you never specified @samp{#pragma
7030 implementation}. This was deemed to be more trouble than it was worth,
7031 however, and disabled.
7033 If you use an explicit @samp{#pragma implementation}, it must appear in
7034 your source file @emph{before} you include the affected header files.
7036 Use the string argument if you want a single implementation file to
7037 include code from multiple header files. (You must also use
7038 @samp{#include} to include the header file; @samp{#pragma
7039 implementation} only specifies how to use the file---it doesn't actually
7042 There is no way to split up the contents of a single header file into
7043 multiple implementation files.
7046 @cindex inlining and C++ pragmas
7047 @cindex C++ pragmas, effect on inlining
7048 @cindex pragmas in C++, effect on inlining
7049 @samp{#pragma implementation} and @samp{#pragma interface} also have an
7050 effect on function inlining.
7052 If you define a class in a header file marked with @samp{#pragma
7053 interface}, the effect on a function defined in that class is similar to
7054 an explicit @code{extern} declaration---the compiler emits no code at
7055 all to define an independent version of the function. Its definition
7056 is used only for inlining with its callers.
7058 @opindex fno-implement-inlines
7059 Conversely, when you include the same header file in a main source file
7060 that declares it as @samp{#pragma implementation}, the compiler emits
7061 code for the function itself; this defines a version of the function
7062 that can be found via pointers (or by callers compiled without
7063 inlining). If all calls to the function can be inlined, you can avoid
7064 emitting the function by compiling with @option{-fno-implement-inlines}.
7065 If any calls were not inlined, you will get linker errors.
7067 @node Template Instantiation
7068 @section Where's the Template?
7070 @cindex template instantiation
7072 C++ templates are the first language feature to require more
7073 intelligence from the environment than one usually finds on a UNIX
7074 system. Somehow the compiler and linker have to make sure that each
7075 template instance occurs exactly once in the executable if it is needed,
7076 and not at all otherwise. There are two basic approaches to this
7077 problem, which I will refer to as the Borland model and the Cfront model.
7081 Borland C++ solved the template instantiation problem by adding the code
7082 equivalent of common blocks to their linker; the compiler emits template
7083 instances in each translation unit that uses them, and the linker
7084 collapses them together. The advantage of this model is that the linker
7085 only has to consider the object files themselves; there is no external
7086 complexity to worry about. This disadvantage is that compilation time
7087 is increased because the template code is being compiled repeatedly.
7088 Code written for this model tends to include definitions of all
7089 templates in the header file, since they must be seen to be
7093 The AT&T C++ translator, Cfront, solved the template instantiation
7094 problem by creating the notion of a template repository, an
7095 automatically maintained place where template instances are stored. A
7096 more modern version of the repository works as follows: As individual
7097 object files are built, the compiler places any template definitions and
7098 instantiations encountered in the repository. At link time, the link
7099 wrapper adds in the objects in the repository and compiles any needed
7100 instances that were not previously emitted. The advantages of this
7101 model are more optimal compilation speed and the ability to use the
7102 system linker; to implement the Borland model a compiler vendor also
7103 needs to replace the linker. The disadvantages are vastly increased
7104 complexity, and thus potential for error; for some code this can be
7105 just as transparent, but in practice it can been very difficult to build
7106 multiple programs in one directory and one program in multiple
7107 directories. Code written for this model tends to separate definitions
7108 of non-inline member templates into a separate file, which should be
7109 compiled separately.
7112 When used with GNU ld version 2.8 or later on an ELF system such as
7113 Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the
7114 Borland model. On other systems, g++ implements neither automatic
7117 A future version of g++ will support a hybrid model whereby the compiler
7118 will emit any instantiations for which the template definition is
7119 included in the compile, and store template definitions and
7120 instantiation context information into the object file for the rest.
7121 The link wrapper will extract that information as necessary and invoke
7122 the compiler to produce the remaining instantiations. The linker will
7123 then combine duplicate instantiations.
7125 In the mean time, you have the following options for dealing with
7126 template instantiations:
7131 Compile your template-using code with @option{-frepo}. The compiler will
7132 generate files with the extension @samp{.rpo} listing all of the
7133 template instantiations used in the corresponding object files which
7134 could be instantiated there; the link wrapper, @samp{collect2}, will
7135 then update the @samp{.rpo} files to tell the compiler where to place
7136 those instantiations and rebuild any affected object files. The
7137 link-time overhead is negligible after the first pass, as the compiler
7138 will continue to place the instantiations in the same files.
7140 This is your best option for application code written for the Borland
7141 model, as it will just work. Code written for the Cfront model will
7142 need to be modified so that the template definitions are available at
7143 one or more points of instantiation; usually this is as simple as adding
7144 @code{#include <tmethods.cc>} to the end of each template header.
7146 For library code, if you want the library to provide all of the template
7147 instantiations it needs, just try to link all of its object files
7148 together; the link will fail, but cause the instantiations to be
7149 generated as a side effect. Be warned, however, that this may cause
7150 conflicts if multiple libraries try to provide the same instantiations.
7151 For greater control, use explicit instantiation as described in the next
7155 @opindex fno-implicit-templates
7156 Compile your code with @option{-fno-implicit-templates} to disable the
7157 implicit generation of template instances, and explicitly instantiate
7158 all the ones you use. This approach requires more knowledge of exactly
7159 which instances you need than do the others, but it's less
7160 mysterious and allows greater control. You can scatter the explicit
7161 instantiations throughout your program, perhaps putting them in the
7162 translation units where the instances are used or the translation units
7163 that define the templates themselves; you can put all of the explicit
7164 instantiations you need into one big file; or you can create small files
7171 template class Foo<int>;
7172 template ostream& operator <<
7173 (ostream&, const Foo<int>&);
7176 for each of the instances you need, and create a template instantiation
7179 If you are using Cfront-model code, you can probably get away with not
7180 using @option{-fno-implicit-templates} when compiling files that don't
7181 @samp{#include} the member template definitions.
7183 If you use one big file to do the instantiations, you may want to
7184 compile it without @option{-fno-implicit-templates} so you get all of the
7185 instances required by your explicit instantiations (but not by any
7186 other files) without having to specify them as well.
7188 g++ has extended the template instantiation syntax outlined in the
7189 Working Paper to allow forward declaration of explicit instantiations
7190 (with @code{extern}), instantiation of the compiler support data for a
7191 template class (i.e.@: the vtable) without instantiating any of its
7192 members (with @code{inline}), and instantiation of only the static data
7193 members of a template class, without the support data or member
7194 functions (with (@code{static}):
7197 extern template int max (int, int);
7198 inline template class Foo<int>;
7199 static template class Foo<int>;
7203 Do nothing. Pretend g++ does implement automatic instantiation
7204 management. Code written for the Borland model will work fine, but
7205 each translation unit will contain instances of each of the templates it
7206 uses. In a large program, this can lead to an unacceptable amount of code
7209 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
7210 more discussion of these pragmas.
7213 @node Bound member functions
7214 @section Extracting the function pointer from a bound pointer to member function
7217 @cindex pointer to member function
7218 @cindex bound pointer to member function
7220 In C++, pointer to member functions (PMFs) are implemented using a wide
7221 pointer of sorts to handle all the possible call mechanisms; the PMF
7222 needs to store information about how to adjust the @samp{this} pointer,
7223 and if the function pointed to is virtual, where to find the vtable, and
7224 where in the vtable to look for the member function. If you are using
7225 PMFs in an inner loop, you should really reconsider that decision. If
7226 that is not an option, you can extract the pointer to the function that
7227 would be called for a given object/PMF pair and call it directly inside
7228 the inner loop, to save a bit of time.
7230 Note that you will still be paying the penalty for the call through a
7231 function pointer; on most modern architectures, such a call defeats the
7232 branch prediction features of the CPU@. This is also true of normal
7233 virtual function calls.
7235 The syntax for this extension is
7239 extern int (A::*fp)();
7240 typedef int (*fptr)(A *);
7242 fptr p = (fptr)(a.*fp);
7245 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
7246 no object is needed to obtain the address of the function. They can be
7247 converted to function pointers directly:
7250 fptr p1 = (fptr)(&A::foo);
7253 @opindex Wno-pmf-conversions
7254 You must specify @option{-Wno-pmf-conversions} to use this extension.
7256 @node C++ Attributes
7257 @section C++-Specific Variable, Function, and Type Attributes
7259 Some attributes only make sense for C++ programs.
7262 @item init_priority (@var{priority})
7263 @cindex init_priority attribute
7266 In Standard C++, objects defined at namespace scope are guaranteed to be
7267 initialized in an order in strict accordance with that of their definitions
7268 @emph{in a given translation unit}. No guarantee is made for initializations
7269 across translation units. However, GNU C++ allows users to control the
7270 order of initialization of objects defined at namespace scope with the
7271 @code{init_priority} attribute by specifying a relative @var{priority},
7272 a constant integral expression currently bounded between 101 and 65535
7273 inclusive. Lower numbers indicate a higher priority.
7275 In the following example, @code{A} would normally be created before
7276 @code{B}, but the @code{init_priority} attribute has reversed that order:
7279 Some_Class A __attribute__ ((init_priority (2000)));
7280 Some_Class B __attribute__ ((init_priority (543)));
7284 Note that the particular values of @var{priority} do not matter; only their
7287 @item java_interface
7288 @cindex java_interface attribute
7290 This type attribute informs C++ that the class is a Java interface. It may
7291 only be applied to classes declared within an @code{extern "Java"} block.
7292 Calls to methods declared in this interface will be dispatched using GCJ's
7293 interface table mechanism, instead of regular virtual table dispatch.
7297 @node Java Exceptions
7298 @section Java Exceptions
7300 The Java language uses a slightly different exception handling model
7301 from C++. Normally, GNU C++ will automatically detect when you are
7302 writing C++ code that uses Java exceptions, and handle them
7303 appropriately. However, if C++ code only needs to execute destructors
7304 when Java exceptions are thrown through it, GCC will guess incorrectly.
7305 Sample problematic code is:
7308 struct S @{ ~S(); @};
7309 extern void bar(); // is written in Java, and may throw exceptions
7318 The usual effect of an incorrect guess is a link failure, complaining of
7319 a missing routine called @samp{__gxx_personality_v0}.
7321 You can inform the compiler that Java exceptions are to be used in a
7322 translation unit, irrespective of what it might think, by writing
7323 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
7324 @samp{#pragma} must appear before any functions that throw or catch
7325 exceptions, or run destructors when exceptions are thrown through them.
7327 You cannot mix Java and C++ exceptions in the same translation unit. It
7328 is believed to be safe to throw a C++ exception from one file through
7329 another file compiled for the Java exception model, or vice versa, but
7330 there may be bugs in this area.
7332 @node Deprecated Features
7333 @section Deprecated Features
7335 In the past, the GNU C++ compiler was extended to experiment with new
7336 features, at a time when the C++ language was still evolving. Now that
7337 the C++ standard is complete, some of those features are superseded by
7338 superior alternatives. Using the old features might cause a warning in
7339 some cases that the feature will be dropped in the future. In other
7340 cases, the feature might be gone already.
7342 While the list below is not exhaustive, it documents some of the options
7343 that are now deprecated:
7346 @item -fexternal-templates
7347 @itemx -falt-external-templates
7348 These are two of the many ways for g++ to implement template
7349 instantiation. @xref{Template Instantiation}. The C++ standard clearly
7350 defines how template definitions have to be organized across
7351 implementation units. g++ has an implicit instantiation mechanism that
7352 should work just fine for standard-conforming code.
7354 @item -fstrict-prototype
7355 @itemx -fno-strict-prototype
7356 Previously it was possible to use an empty prototype parameter list to
7357 indicate an unspecified number of parameters (like C), rather than no
7358 parameters, as C++ demands. This feature has been removed, except where
7359 it is required for backwards compatibility @xref{Backwards Compatibility}.
7362 The named return value extension has been deprecated, and is now
7365 The use of initializer lists with new expressions has been deprecated,
7366 and is now removed from g++.
7368 Floating and complex non-type template parameters have been deprecated,
7369 and are now removed from g++.
7371 The implicit typename extension has been deprecated and will be removed
7372 from g++ at some point. In some cases g++ determines that a dependant
7373 type such as @code{TPL<T>::X} is a type without needing a
7374 @code{typename} keyword, contrary to the standard.
7376 @node Backwards Compatibility
7377 @section Backwards Compatibility
7378 @cindex Backwards Compatibility
7379 @cindex ARM [Annotated C++ Reference Manual]
7381 Now that there is a definitive ISO standard C++, G++ has a specification
7382 to adhere to. The C++ language evolved over time, and features that
7383 used to be acceptable in previous drafts of the standard, such as the ARM
7384 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
7385 compilation of C++ written to such drafts, G++ contains some backwards
7386 compatibilities. @emph{All such backwards compatibility features are
7387 liable to disappear in future versions of G++.} They should be considered
7388 deprecated @xref{Deprecated Features}.
7392 If a variable is declared at for scope, it used to remain in scope until
7393 the end of the scope which contained the for statement (rather than just
7394 within the for scope). G++ retains this, but issues a warning, if such a
7395 variable is accessed outside the for scope.
7397 @item Implicit C language
7398 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
7399 scope to set the language. On such systems, all header files are
7400 implicitly scoped inside a C language scope. Also, an empty prototype
7401 @code{()} will be treated as an unspecified number of arguments, rather
7402 than no arguments, as C++ demands.