extend.texi revision 96263
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. 4 5@node C Implementation 6@chapter C Implementation-defined behavior 7@cindex implementation-defined behavior, C language 8 9A conforming implementation of ISO C is required to document its 10choice of behavior in each of the areas that are designated 11``implementation defined.'' The following lists all such areas, 12along with the section number from the ISO/IEC 9899:1999 standard. 13 14@menu 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:: 29@end menu 30 31@node Translation implementation 32@section Translation 33 34@itemize @bullet 35@item 36@cite{How a diagnostic is identified (3.10, 5.1.1.3).} 37 38@item 39@cite{Whether each nonempty sequence of white-space characters other than 40new-line is retained or replaced by one space character in translation 41phase 3 (5.1.1.2).} 42@end itemize 43 44@node Environment implementation 45@section Environment 46 47The behavior of these points are dependent on the implementation 48of the C library, and are not defined by GCC itself. 49 50@node Identifiers implementation 51@section Identifiers 52 53@itemize @bullet 54@item 55@cite{Which additional multibyte characters may appear in identifiers 56and their correspondence to universal character names (6.4.2).} 57 58@item 59@cite{The number of significant initial characters in an identifier 60(5.2.4.1, 6.4.2).} 61@end itemize 62 63@node Characters implementation 64@section Characters 65 66@itemize @bullet 67@item 68@cite{The number of bits in a byte (3.6).} 69 70@item 71@cite{The values of the members of the execution character set (5.2.1).} 72 73@item 74@cite{The unique value of the member of the execution character set produced 75for each of the standard alphabetic escape sequences (5.2.2).} 76 77@item 78@cite{The value of a @code{char} object into which has been stored any 79character other than a member of the basic execution character set (6.2.5).} 80 81@item 82@cite{Which of @code{signed char} or @code{unsigned char} has the same range, 83representation, and behavior as ``plain'' @code{char} (6.2.5, 6.3.1.1).} 84 85@item 86@cite{The mapping of members of the source character set (in character 87constants and string literals) to members of the execution character 88set (6.4.4.4, 5.1.1.2).} 89 90@item 91@cite{The value of an integer character constant containing more than one 92character or containing a character or escape sequence that does not map 93to a single-byte execution character (6.4.4.4).} 94 95@item 96@cite{The value of a wide character constant containing more than one 97multibyte character, or containing a multibyte character or escape 98sequence not represented in the extended execution character set (6.4.4.4).} 99 100@item 101@cite{The current locale used to convert a wide character constant consisting 102of a single multibyte character that maps to a member of the extended 103execution character set into a corresponding wide character code (6.4.4.4).} 104 105@item 106@cite{The current locale used to convert a wide string literal into 107corresponding wide character codes (6.4.5).} 108 109@item 110@cite{The value of a string literal containing a multibyte character or escape 111sequence not represented in the execution character set (6.4.5).} 112@end itemize 113 114@node Integers implementation 115@section Integers 116 117@itemize @bullet 118@item 119@cite{Any extended integer types that exist in the implementation (6.2.5).} 120 121@item 122@cite{Whether signed integer types are represented using sign and magnitude, 123two's complement, or one's complement, and whether the extraordinary value 124is a trap representation or an ordinary value (6.2.6.2).} 125 126@item 127@cite{The rank of any extended integer type relative to another extended 128integer type with the same precision (6.3.1.1).} 129 130@item 131@cite{The result of, or the signal raised by, converting an integer to a 132signed integer type when the value cannot be represented in an object of 133that type (6.3.1.3).} 134 135@item 136@cite{The results of some bitwise operations on signed integers (6.5).} 137@end itemize 138 139@node Floating point implementation 140@section Floating point 141 142@itemize @bullet 143@item 144@cite{The accuracy of the floating-point operations and of the library 145functions in @code{<math.h>} and @code{<complex.h>} that return floating-point 146results (5.2.4.2.2).} 147 148@item 149@cite{The rounding behaviors characterized by non-standard values 150of @code{FLT_ROUNDS} @gol 151(5.2.4.2.2).} 152 153@item 154@cite{The evaluation methods characterized by non-standard negative 155values of @code{FLT_EVAL_METHOD} (5.2.4.2.2).} 156 157@item 158@cite{The direction of rounding when an integer is converted to a 159floating-point number that cannot exactly represent the original 160value (6.3.1.4).} 161 162@item 163@cite{The direction of rounding when a floating-point number is 164converted to a narrower floating-point number (6.3.1.5).} 165 166@item 167@cite{How the nearest representable value or the larger or smaller 168representable value immediately adjacent to the nearest representable 169value is chosen for certain floating constants (6.4.4.2).} 170 171@item 172@cite{Whether and how floating expressions are contracted when not 173disallowed by the @code{FP_CONTRACT} pragma (6.5).} 174 175@item 176@cite{The default state for the @code{FENV_ACCESS} pragma (7.6.1).} 177 178@item 179@cite{Additional floating-point exceptions, rounding modes, environments, 180and classifications, and their macro names (7.6, 7.12).} 181 182@item 183@cite{The default state for the @code{FP_CONTRACT} pragma (7.12.2).} 184 185@item 186@cite{Whether the ``inexact'' floating-point exception can be raised 187when the rounded result actually does equal the mathematical result 188in an IEC 60559 conformant implementation (F.9).} 189 190@item 191@cite{Whether the ``underflow'' (and ``inexact'') floating-point 192exception can be raised when a result is tiny but not inexact in an 193IEC 60559 conformant implementation (F.9).} 194 195@end itemize 196 197@node Arrays and pointers implementation 198@section Arrays and pointers 199 200@itemize @bullet 201@item 202@cite{The result of converting a pointer to an integer or 203vice versa (6.3.2.3).} 204 205A cast from pointer to integer discards most-significant bits if the 206pointer representation is larger than the integer type, 207sign-extends@footnote{Future versions of GCC may zero-extend, or use 208a target-defined @code{ptr_extend} pattern. Do not rely on sign extension.} 209if the pointer representation is smaller than the integer type, otherwise 210the bits are unchanged. 211@c ??? We've always claimed that pointers were unsigned entities. 212@c Shouldn't we therefore be doing zero-extension? If so, the bug 213@c is in convert_to_integer, where we call type_for_size and request 214@c a signed integral type. On the other hand, it might be most useful 215@c for the target if we extend according to POINTERS_EXTEND_UNSIGNED. 216 217A cast from integer to pointer discards most-significant bits if the 218pointer representation is smaller than the integer type, extends according 219to the signedness of the integer type if the pointer representation 220is larger than the integer type, otherwise the bits are unchanged. 221 222When casting from pointer to integer and back again, the resulting 223pointer must reference the same object as the original pointer, otherwise 224the behavior is undefined. That is, one may not use integer arithmetic to 225avoid the undefined behavior of pointer arithmetic as proscribed in 6.5.6/8. 226 227@item 228@cite{The size of the result of subtracting two pointers to elements 229of the same array (6.5.6).} 230 231@end itemize 232 233@node Hints implementation 234@section Hints 235 236@itemize @bullet 237@item 238@cite{The extent to which suggestions made by using the @code{register} 239storage-class specifier are effective (6.7.1).} 240 241@item 242@cite{The extent to which suggestions made by using the inline function 243specifier are effective (6.7.4).} 244 245@end itemize 246 247@node Structures unions enumerations and bit-fields implementation 248@section Structures, unions, enumerations, and bit-fields 249 250@itemize @bullet 251@item 252@cite{Whether a ``plain'' int bit-field is treated as a @code{signed int} 253bit-field or as an @code{unsigned int} bit-field (6.7.2, 6.7.2.1).} 254 255@item 256@cite{Allowable bit-field types other than @code{_Bool}, @code{signed int}, 257and @code{unsigned int} (6.7.2.1).} 258 259@item 260@cite{Whether a bit-field can straddle a storage-unit boundary (6.7.2.1).} 261 262@item 263@cite{The order of allocation of bit-fields within a unit (6.7.2.1).} 264 265@item 266@cite{The alignment of non-bit-field members of structures (6.7.2.1).} 267 268@item 269@cite{The integer type compatible with each enumerated type (6.7.2.2).} 270 271@end itemize 272 273@node Qualifiers implementation 274@section Qualifiers 275 276@itemize @bullet 277@item 278@cite{What constitutes an access to an object that has volatile-qualified 279type (6.7.3).} 280 281@end itemize 282 283@node Preprocessing directives implementation 284@section Preprocessing directives 285 286@itemize @bullet 287@item 288@cite{How sequences in both forms of header names are mapped to headers 289or external source file names (6.4.7).} 290 291@item 292@cite{Whether the value of a character constant in a constant expression 293that controls conditional inclusion matches the value of the same character 294constant in the execution character set (6.10.1).} 295 296@item 297@cite{Whether the value of a single-character character constant in a 298constant expression that controls conditional inclusion may have a 299negative value (6.10.1).} 300 301@item 302@cite{The places that are searched for an included @samp{<>} delimited 303header, and how the places are specified or the header is 304identified (6.10.2).} 305 306@item 307@cite{How the named source file is searched for in an included @samp{""} 308delimited header (6.10.2).} 309 310@item 311@cite{The method by which preprocessing tokens (possibly resulting from 312macro expansion) in a @code{#include} directive are combined into a header 313name (6.10.2).} 314 315@item 316@cite{The nesting limit for @code{#include} processing (6.10.2).} 317 318@item 319@cite{Whether the @samp{#} operator inserts a @samp{\} character before 320the @samp{\} character that begins a universal character name in a 321character constant or string literal (6.10.3.2).} 322 323@item 324@cite{The behavior on each recognized non-@code{STDC #pragma} 325directive (6.10.6).} 326 327@item 328@cite{The definitions for @code{__DATE__} and @code{__TIME__} when 329respectively, the date and time of translation are not available (6.10.8).} 330 331@end itemize 332 333@node Library functions implementation 334@section Library functions 335 336The behavior of these points are dependent on the implementation 337of the C library, and are not defined by GCC itself. 338 339@node Architecture implementation 340@section Architecture 341 342@itemize @bullet 343@item 344@cite{The values or expressions assigned to the macros specified in the 345headers @code{<float.h>}, @code{<limits.h>}, and @code{<stdint.h>} 346(5.2.4.2, 7.18.2, 7.18.3).} 347 348@item 349@cite{The number, order, and encoding of bytes in any object 350(when not explicitly specified in this International Standard) (6.2.6.1).} 351 352@item 353@cite{The value of the result of the sizeof operator (6.5.3.4).} 354 355@end itemize 356 357@node Locale-specific behavior implementation 358@section Locale-specific behavior 359 360The behavior of these points are dependent on the implementation 361of the C library, and are not defined by GCC itself. 362 363@node C Extensions 364@chapter Extensions to the C Language Family 365@cindex extensions, C language 366@cindex C language extensions 367 368@opindex pedantic 369GNU C provides several language features not found in ISO standard C@. 370(The @option{-pedantic} option directs GCC to print a warning message if 371any of these features is used.) To test for the availability of these 372features in conditional compilation, check for a predefined macro 373@code{__GNUC__}, which is always defined under GCC@. 374 375These extensions are available in C and Objective-C@. Most of them are 376also available in C++. @xref{C++ Extensions,,Extensions to the 377C++ Language}, for extensions that apply @emph{only} to C++. 378 379Some features that are in ISO C99 but not C89 or C++ are also, as 380extensions, accepted by GCC in C89 mode and in C++. 381 382@menu 383* Statement Exprs:: Putting statements and declarations inside expressions. 384* Local Labels:: Labels local to a statement-expression. 385* Labels as Values:: Getting pointers to labels, and computed gotos. 386* Nested Functions:: As in Algol and Pascal, lexical scoping of functions. 387* Constructing Calls:: Dispatching a call to another function. 388* Naming Types:: Giving a name to the type of some expression. 389* Typeof:: @code{typeof}: referring to the type of an expression. 390* Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues. 391* Conditionals:: Omitting the middle operand of a @samp{?:} expression. 392* Long Long:: Double-word integers---@code{long long int}. 393* Complex:: Data types for complex numbers. 394* Hex Floats:: Hexadecimal floating-point constants. 395* Zero Length:: Zero-length arrays. 396* Variable Length:: Arrays whose length is computed at run time. 397* Variadic Macros:: Macros with a variable number of arguments. 398* Escaped Newlines:: Slightly looser rules for escaped newlines. 399* Multi-line Strings:: String literals with embedded newlines. 400* Subscripting:: Any array can be subscripted, even if not an lvalue. 401* Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. 402* Initializers:: Non-constant initializers. 403* Compound Literals:: Compound literals give structures, unions 404 or arrays as values. 405* Designated Inits:: Labeling elements of initializers. 406* Cast to Union:: Casting to union type from any member of the union. 407* Case Ranges:: `case 1 ... 9' and such. 408* Mixed Declarations:: Mixing declarations and code. 409* Function Attributes:: Declaring that functions have no side effects, 410 or that they can never return. 411* Attribute Syntax:: Formal syntax for attributes. 412* Function Prototypes:: Prototype declarations and old-style definitions. 413* C++ Comments:: C++ comments are recognized. 414* Dollar Signs:: Dollar sign is allowed in identifiers. 415* Character Escapes:: @samp{\e} stands for the character @key{ESC}. 416* Variable Attributes:: Specifying attributes of variables. 417* Type Attributes:: Specifying attributes of types. 418* Alignment:: Inquiring about the alignment of a type or variable. 419* Inline:: Defining inline functions (as fast as macros). 420* Extended Asm:: Assembler instructions with C expressions as operands. 421 (With them you can define ``built-in'' functions.) 422* Constraints:: Constraints for asm operands 423* Asm Labels:: Specifying the assembler name to use for a C symbol. 424* Explicit Reg Vars:: Defining variables residing in specified registers. 425* Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. 426* Incomplete Enums:: @code{enum foo;}, with details to follow. 427* Function Names:: Printable strings which are the name of the current 428 function. 429* Return Address:: Getting the return or frame address of a function. 430* Vector Extensions:: Using vector instructions through built-in functions. 431* Other Builtins:: Other built-in functions. 432* Target Builtins:: Built-in functions specific to particular targets. 433* Pragmas:: Pragmas accepted by GCC. 434* Unnamed Fields:: Unnamed struct/union fields within structs/unions. 435@end menu 436 437@node Statement Exprs 438@section Statements and Declarations in Expressions 439@cindex statements inside expressions 440@cindex declarations inside expressions 441@cindex expressions containing statements 442@cindex macros, statements in expressions 443 444@c the above section title wrapped and causes an underfull hbox.. i 445@c changed it from "within" to "in". --mew 4feb93 446 447A compound statement enclosed in parentheses may appear as an expression 448in GNU C@. This allows you to use loops, switches, and local variables 449within an expression. 450 451Recall that a compound statement is a sequence of statements surrounded 452by braces; in this construct, parentheses go around the braces. For 453example: 454 455@example 456(@{ int y = foo (); int z; 457 if (y > 0) z = y; 458 else z = - y; 459 z; @}) 460@end example 461 462@noindent 463is a valid (though slightly more complex than necessary) expression 464for the absolute value of @code{foo ()}. 465 466The last thing in the compound statement should be an expression 467followed by a semicolon; the value of this subexpression serves as the 468value of the entire construct. (If you use some other kind of statement 469last within the braces, the construct has type @code{void}, and thus 470effectively no value.) 471 472This feature is especially useful in making macro definitions ``safe'' (so 473that they evaluate each operand exactly once). For example, the 474``maximum'' function is commonly defined as a macro in standard C as 475follows: 476 477@example 478#define max(a,b) ((a) > (b) ? (a) : (b)) 479@end example 480 481@noindent 482@cindex side effects, macro argument 483But this definition computes either @var{a} or @var{b} twice, with bad 484results if the operand has side effects. In GNU C, if you know the 485type of the operands (here let's assume @code{int}), you can define 486the macro safely as follows: 487 488@example 489#define maxint(a,b) \ 490 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @}) 491@end example 492 493Embedded statements are not allowed in constant expressions, such as 494the value of an enumeration constant, the width of a bit-field, or 495the initial value of a static variable. 496 497If you don't know the type of the operand, you can still do this, but you 498must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming 499Types}). 500 501Statement expressions are not supported fully in G++, and their fate 502there is unclear. (It is possible that they will become fully supported 503at some point, or that they will be deprecated, or that the bugs that 504are present will continue to exist indefinitely.) Presently, statement 505expressions do not work well as default arguments. 506 507In addition, there are semantic issues with statement-expressions in 508C++. If you try to use statement-expressions instead of inline 509functions in C++, you may be surprised at the way object destruction is 510handled. For example: 511 512@example 513#define foo(a) (@{int b = (a); b + 3; @}) 514@end example 515 516@noindent 517does not work the same way as: 518 519@example 520inline int foo(int a) @{ int b = a; return b + 3; @} 521@end example 522 523@noindent 524In particular, if the expression passed into @code{foo} involves the 525creation of temporaries, the destructors for those temporaries will be 526run earlier in the case of the macro than in the case of the function. 527 528These considerations mean that it is probably a bad idea to use 529statement-expressions of this form in header files that are designed to 530work with C++. (Note that some versions of the GNU C Library contained 531header files using statement-expression that lead to precisely this 532bug.) 533 534@node Local Labels 535@section Locally Declared Labels 536@cindex local labels 537@cindex macros, local labels 538 539Each statement expression is a scope in which @dfn{local labels} can be 540declared. A local label is simply an identifier; you can jump to it 541with an ordinary @code{goto} statement, but only from within the 542statement expression it belongs to. 543 544A local label declaration looks like this: 545 546@example 547__label__ @var{label}; 548@end example 549 550@noindent 551or 552 553@example 554__label__ @var{label1}, @var{label2}, @dots{}; 555@end example 556 557Local label declarations must come at the beginning of the statement 558expression, right after the @samp{(@{}, before any ordinary 559declarations. 560 561The label declaration defines the label @emph{name}, but does not define 562the label itself. You must do this in the usual way, with 563@code{@var{label}:}, within the statements of the statement expression. 564 565The local label feature is useful because statement expressions are 566often used in macros. If the macro contains nested loops, a @code{goto} 567can be useful for breaking out of them. However, an ordinary label 568whose scope is the whole function cannot be used: if the macro can be 569expanded several times in one function, the label will be multiply 570defined in that function. A local label avoids this problem. For 571example: 572 573@example 574#define SEARCH(array, target) \ 575(@{ \ 576 __label__ found; \ 577 typeof (target) _SEARCH_target = (target); \ 578 typeof (*(array)) *_SEARCH_array = (array); \ 579 int i, j; \ 580 int value; \ 581 for (i = 0; i < max; i++) \ 582 for (j = 0; j < max; j++) \ 583 if (_SEARCH_array[i][j] == _SEARCH_target) \ 584 @{ value = i; goto found; @} \ 585 value = -1; \ 586 found: \ 587 value; \ 588@}) 589@end example 590 591@node Labels as Values 592@section Labels as Values 593@cindex labels as values 594@cindex computed gotos 595@cindex goto with computed label 596@cindex address of a label 597 598You can get the address of a label defined in the current function 599(or a containing function) with the unary operator @samp{&&}. The 600value has type @code{void *}. This value is a constant and can be used 601wherever a constant of that type is valid. For example: 602 603@example 604void *ptr; 605@dots{} 606ptr = &&foo; 607@end example 608 609To use these values, you need to be able to jump to one. This is done 610with the computed goto statement@footnote{The analogous feature in 611Fortran is called an assigned goto, but that name seems inappropriate in 612C, where one can do more than simply store label addresses in label 613variables.}, @code{goto *@var{exp};}. For example, 614 615@example 616goto *ptr; 617@end example 618 619@noindent 620Any expression of type @code{void *} is allowed. 621 622One way of using these constants is in initializing a static array that 623will serve as a jump table: 624 625@example 626static void *array[] = @{ &&foo, &&bar, &&hack @}; 627@end example 628 629Then you can select a label with indexing, like this: 630 631@example 632goto *array[i]; 633@end example 634 635@noindent 636Note that this does not check whether the subscript is in bounds---array 637indexing in C never does that. 638 639Such an array of label values serves a purpose much like that of the 640@code{switch} statement. The @code{switch} statement is cleaner, so 641use that rather than an array unless the problem does not fit a 642@code{switch} statement very well. 643 644Another use of label values is in an interpreter for threaded code. 645The labels within the interpreter function can be stored in the 646threaded code for super-fast dispatching. 647 648You may not use this mechanism to jump to code in a different function. 649If you do that, totally unpredictable things will happen. The best way to 650avoid this is to store the label address only in automatic variables and 651never pass it as an argument. 652 653An alternate way to write the above example is 654 655@example 656static const int array[] = @{ &&foo - &&foo, &&bar - &&foo, 657 &&hack - &&foo @}; 658goto *(&&foo + array[i]); 659@end example 660 661@noindent 662This is more friendly to code living in shared libraries, as it reduces 663the number of dynamic relocations that are needed, and by consequence, 664allows the data to be read-only. 665 666@node Nested Functions 667@section Nested Functions 668@cindex nested functions 669@cindex downward funargs 670@cindex thunks 671 672A @dfn{nested function} is a function defined inside another function. 673(Nested functions are not supported for GNU C++.) The nested function's 674name is local to the block where it is defined. For example, here we 675define a nested function named @code{square}, and call it twice: 676 677@example 678@group 679foo (double a, double b) 680@{ 681 double square (double z) @{ return z * z; @} 682 683 return square (a) + square (b); 684@} 685@end group 686@end example 687 688The nested function can access all the variables of the containing 689function that are visible at the point of its definition. This is 690called @dfn{lexical scoping}. For example, here we show a nested 691function which uses an inherited variable named @code{offset}: 692 693@example 694@group 695bar (int *array, int offset, int size) 696@{ 697 int access (int *array, int index) 698 @{ return array[index + offset]; @} 699 int i; 700 @dots{} 701 for (i = 0; i < size; i++) 702 @dots{} access (array, i) @dots{} 703@} 704@end group 705@end example 706 707Nested function definitions are permitted within functions in the places 708where variable definitions are allowed; that is, in any block, before 709the first statement in the block. 710 711It is possible to call the nested function from outside the scope of its 712name by storing its address or passing the address to another function: 713 714@example 715hack (int *array, int size) 716@{ 717 void store (int index, int value) 718 @{ array[index] = value; @} 719 720 intermediate (store, size); 721@} 722@end example 723 724Here, the function @code{intermediate} receives the address of 725@code{store} as an argument. If @code{intermediate} calls @code{store}, 726the arguments given to @code{store} are used to store into @code{array}. 727But this technique works only so long as the containing function 728(@code{hack}, in this example) does not exit. 729 730If you try to call the nested function through its address after the 731containing function has exited, all hell will break loose. If you try 732to call it after a containing scope level has exited, and if it refers 733to some of the variables that are no longer in scope, you may be lucky, 734but it's not wise to take the risk. If, however, the nested function 735does not refer to anything that has gone out of scope, you should be 736safe. 737 738GCC implements taking the address of a nested function using a technique 739called @dfn{trampolines}. A paper describing them is available as 740 741@noindent 742@uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}. 743 744A nested function can jump to a label inherited from a containing 745function, provided the label was explicitly declared in the containing 746function (@pxref{Local Labels}). Such a jump returns instantly to the 747containing function, exiting the nested function which did the 748@code{goto} and any intermediate functions as well. Here is an example: 749 750@example 751@group 752bar (int *array, int offset, int size) 753@{ 754 __label__ failure; 755 int access (int *array, int index) 756 @{ 757 if (index > size) 758 goto failure; 759 return array[index + offset]; 760 @} 761 int i; 762 @dots{} 763 for (i = 0; i < size; i++) 764 @dots{} access (array, i) @dots{} 765 @dots{} 766 return 0; 767 768 /* @r{Control comes here from @code{access} 769 if it detects an error.} */ 770 failure: 771 return -1; 772@} 773@end group 774@end example 775 776A nested function always has internal linkage. Declaring one with 777@code{extern} is erroneous. If you need to declare the nested function 778before its definition, use @code{auto} (which is otherwise meaningless 779for function declarations). 780 781@example 782bar (int *array, int offset, int size) 783@{ 784 __label__ failure; 785 auto int access (int *, int); 786 @dots{} 787 int access (int *array, int index) 788 @{ 789 if (index > size) 790 goto failure; 791 return array[index + offset]; 792 @} 793 @dots{} 794@} 795@end example 796 797@node Constructing Calls 798@section Constructing Function Calls 799@cindex constructing calls 800@cindex forwarding calls 801 802Using the built-in functions described below, you can record 803the arguments a function received, and call another function 804with the same arguments, without knowing the number or types 805of the arguments. 806 807You can also record the return value of that function call, 808and later return that value, without knowing what data type 809the function tried to return (as long as your caller expects 810that data type). 811 812@deftypefn {Built-in Function} {void *} __builtin_apply_args () 813This built-in function returns a pointer to data 814describing how to perform a call with the same arguments as were passed 815to the current function. 816 817The function saves the arg pointer register, structure value address, 818and all registers that might be used to pass arguments to a function 819into a block of memory allocated on the stack. Then it returns the 820address of that block. 821@end deftypefn 822 823@deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size}) 824This built-in function invokes @var{function} 825with a copy of the parameters described by @var{arguments} 826and @var{size}. 827 828The value of @var{arguments} should be the value returned by 829@code{__builtin_apply_args}. The argument @var{size} specifies the size 830of the stack argument data, in bytes. 831 832This function returns a pointer to data describing 833how to return whatever value was returned by @var{function}. The data 834is saved in a block of memory allocated on the stack. 835 836It is not always simple to compute the proper value for @var{size}. The 837value is used by @code{__builtin_apply} to compute the amount of data 838that should be pushed on the stack and copied from the incoming argument 839area. 840@end deftypefn 841 842@deftypefn {Built-in Function} {void} __builtin_return (void *@var{result}) 843This built-in function returns the value described by @var{result} from 844the containing function. You should specify, for @var{result}, a value 845returned by @code{__builtin_apply}. 846@end deftypefn 847 848@node Naming Types 849@section Naming an Expression's Type 850@cindex naming types 851 852You can give a name to the type of an expression using a @code{typedef} 853declaration with an initializer. Here is how to define @var{name} as a 854type name for the type of @var{exp}: 855 856@example 857typedef @var{name} = @var{exp}; 858@end example 859 860This is useful in conjunction with the statements-within-expressions 861feature. Here is how the two together can be used to define a safe 862``maximum'' macro that operates on any arithmetic type: 863 864@example 865#define max(a,b) \ 866 (@{typedef _ta = (a), _tb = (b); \ 867 _ta _a = (a); _tb _b = (b); \ 868 _a > _b ? _a : _b; @}) 869@end example 870 871@cindex underscores in variables in macros 872@cindex @samp{_} in variables in macros 873@cindex local variables in macros 874@cindex variables, local, in macros 875@cindex macros, local variables in 876 877The reason for using names that start with underscores for the local 878variables is to avoid conflicts with variable names that occur within the 879expressions that are substituted for @code{a} and @code{b}. Eventually we 880hope to design a new form of declaration syntax that allows you to declare 881variables whose scopes start only after their initializers; this will be a 882more reliable way to prevent such conflicts. 883 884@node Typeof 885@section Referring to a Type with @code{typeof} 886@findex typeof 887@findex sizeof 888@cindex macros, types of arguments 889 890Another way to refer to the type of an expression is with @code{typeof}. 891The syntax of using of this keyword looks like @code{sizeof}, but the 892construct acts semantically like a type name defined with @code{typedef}. 893 894There are two ways of writing the argument to @code{typeof}: with an 895expression or with a type. Here is an example with an expression: 896 897@example 898typeof (x[0](1)) 899@end example 900 901@noindent 902This assumes that @code{x} is an array of pointers to functions; 903the type described is that of the values of the functions. 904 905Here is an example with a typename as the argument: 906 907@example 908typeof (int *) 909@end example 910 911@noindent 912Here the type described is that of pointers to @code{int}. 913 914If you are writing a header file that must work when included in ISO C 915programs, write @code{__typeof__} instead of @code{typeof}. 916@xref{Alternate Keywords}. 917 918A @code{typeof}-construct can be used anywhere a typedef name could be 919used. For example, you can use it in a declaration, in a cast, or inside 920of @code{sizeof} or @code{typeof}. 921 922@itemize @bullet 923@item 924This declares @code{y} with the type of what @code{x} points to. 925 926@example 927typeof (*x) y; 928@end example 929 930@item 931This declares @code{y} as an array of such values. 932 933@example 934typeof (*x) y[4]; 935@end example 936 937@item 938This declares @code{y} as an array of pointers to characters: 939 940@example 941typeof (typeof (char *)[4]) y; 942@end example 943 944@noindent 945It is equivalent to the following traditional C declaration: 946 947@example 948char *y[4]; 949@end example 950 951To see the meaning of the declaration using @code{typeof}, and why it 952might be a useful way to write, let's rewrite it with these macros: 953 954@example 955#define pointer(T) typeof(T *) 956#define array(T, N) typeof(T [N]) 957@end example 958 959@noindent 960Now the declaration can be rewritten this way: 961 962@example 963array (pointer (char), 4) y; 964@end example 965 966@noindent 967Thus, @code{array (pointer (char), 4)} is the type of arrays of 4 968pointers to @code{char}. 969@end itemize 970 971@node Lvalues 972@section Generalized Lvalues 973@cindex compound expressions as lvalues 974@cindex expressions, compound, as lvalues 975@cindex conditional expressions as lvalues 976@cindex expressions, conditional, as lvalues 977@cindex casts as lvalues 978@cindex generalized lvalues 979@cindex lvalues, generalized 980@cindex extensions, @code{?:} 981@cindex @code{?:} extensions 982Compound expressions, conditional expressions and casts are allowed as 983lvalues provided their operands are lvalues. This means that you can take 984their addresses or store values into them. 985 986Standard C++ allows compound expressions and conditional expressions as 987lvalues, and permits casts to reference type, so use of this extension 988is deprecated for C++ code. 989 990For example, a compound expression can be assigned, provided the last 991expression in the sequence is an lvalue. These two expressions are 992equivalent: 993 994@example 995(a, b) += 5 996a, (b += 5) 997@end example 998 999Similarly, the address of the compound expression can be taken. These two 1000expressions are equivalent: 1001 1002@example 1003&(a, b) 1004a, &b 1005@end example 1006 1007A conditional expression is a valid lvalue if its type is not void and the 1008true and false branches are both valid lvalues. For example, these two 1009expressions are equivalent: 1010 1011@example 1012(a ? b : c) = 5 1013(a ? b = 5 : (c = 5)) 1014@end example 1015 1016A cast is a valid lvalue if its operand is an lvalue. A simple 1017assignment whose left-hand side is a cast works by converting the 1018right-hand side first to the specified type, then to the type of the 1019inner left-hand side expression. After this is stored, the value is 1020converted back to the specified type to become the value of the 1021assignment. Thus, if @code{a} has type @code{char *}, the following two 1022expressions are equivalent: 1023 1024@example 1025(int)a = 5 1026(int)(a = (char *)(int)5) 1027@end example 1028 1029An assignment-with-arithmetic operation such as @samp{+=} applied to a cast 1030performs the arithmetic using the type resulting from the cast, and then 1031continues as in the previous case. Therefore, these two expressions are 1032equivalent: 1033 1034@example 1035(int)a += 5 1036(int)(a = (char *)(int) ((int)a + 5)) 1037@end example 1038 1039You cannot take the address of an lvalue cast, because the use of its 1040address would not work out coherently. Suppose that @code{&(int)f} were 1041permitted, where @code{f} has type @code{float}. Then the following 1042statement would try to store an integer bit-pattern where a floating 1043point number belongs: 1044 1045@example 1046*&(int)f = 1; 1047@end example 1048 1049This is quite different from what @code{(int)f = 1} would do---that 1050would convert 1 to floating point and store it. Rather than cause this 1051inconsistency, we think it is better to prohibit use of @samp{&} on a cast. 1052 1053If you really do want an @code{int *} pointer with the address of 1054@code{f}, you can simply write @code{(int *)&f}. 1055 1056@node Conditionals 1057@section Conditionals with Omitted Operands 1058@cindex conditional expressions, extensions 1059@cindex omitted middle-operands 1060@cindex middle-operands, omitted 1061@cindex extensions, @code{?:} 1062@cindex @code{?:} extensions 1063 1064The middle operand in a conditional expression may be omitted. Then 1065if the first operand is nonzero, its value is the value of the conditional 1066expression. 1067 1068Therefore, the expression 1069 1070@example 1071x ? : y 1072@end example 1073 1074@noindent 1075has the value of @code{x} if that is nonzero; otherwise, the value of 1076@code{y}. 1077 1078This example is perfectly equivalent to 1079 1080@example 1081x ? x : y 1082@end example 1083 1084@cindex side effect in ?: 1085@cindex ?: side effect 1086@noindent 1087In this simple case, the ability to omit the middle operand is not 1088especially useful. When it becomes useful is when the first operand does, 1089or may (if it is a macro argument), contain a side effect. Then repeating 1090the operand in the middle would perform the side effect twice. Omitting 1091the middle operand uses the value already computed without the undesirable 1092effects of recomputing it. 1093 1094@node Long Long 1095@section Double-Word Integers 1096@cindex @code{long long} data types 1097@cindex double-word arithmetic 1098@cindex multiprecision arithmetic 1099@cindex @code{LL} integer suffix 1100@cindex @code{ULL} integer suffix 1101 1102ISO C99 supports data types for integers that are at least 64 bits wide, 1103and as an extension GCC supports them in C89 mode and in C++. 1104Simply write @code{long long int} for a signed integer, or 1105@code{unsigned long long int} for an unsigned integer. To make an 1106integer constant of type @code{long long int}, add the suffix @samp{LL} 1107to the integer. To make an integer constant of type @code{unsigned long 1108long int}, add the suffix @samp{ULL} to the integer. 1109 1110You can use these types in arithmetic like any other integer types. 1111Addition, subtraction, and bitwise boolean operations on these types 1112are open-coded on all types of machines. Multiplication is open-coded 1113if the machine supports fullword-to-doubleword a widening multiply 1114instruction. Division and shifts are open-coded only on machines that 1115provide special support. The operations that are not open-coded use 1116special library routines that come with GCC@. 1117 1118There may be pitfalls when you use @code{long long} types for function 1119arguments, unless you declare function prototypes. If a function 1120expects type @code{int} for its argument, and you pass a value of type 1121@code{long long int}, confusion will result because the caller and the 1122subroutine will disagree about the number of bytes for the argument. 1123Likewise, if the function expects @code{long long int} and you pass 1124@code{int}. The best way to avoid such problems is to use prototypes. 1125 1126@node Complex 1127@section Complex Numbers 1128@cindex complex numbers 1129@cindex @code{_Complex} keyword 1130@cindex @code{__complex__} keyword 1131 1132ISO C99 supports complex floating data types, and as an extension GCC 1133supports them in C89 mode and in C++, and supports complex integer data 1134types which are not part of ISO C99. You can declare complex types 1135using the keyword @code{_Complex}. As an extension, the older GNU 1136keyword @code{__complex__} is also supported. 1137 1138For example, @samp{_Complex double x;} declares @code{x} as a 1139variable whose real part and imaginary part are both of type 1140@code{double}. @samp{_Complex short int y;} declares @code{y} to 1141have real and imaginary parts of type @code{short int}; this is not 1142likely to be useful, but it shows that the set of complex types is 1143complete. 1144 1145To write a constant with a complex data type, use the suffix @samp{i} or 1146@samp{j} (either one; they are equivalent). For example, @code{2.5fi} 1147has type @code{_Complex float} and @code{3i} has type 1148@code{_Complex int}. Such a constant always has a pure imaginary 1149value, but you can form any complex value you like by adding one to a 1150real constant. This is a GNU extension; if you have an ISO C99 1151conforming C library (such as GNU libc), and want to construct complex 1152constants of floating type, you should include @code{<complex.h>} and 1153use the macros @code{I} or @code{_Complex_I} instead. 1154 1155@cindex @code{__real__} keyword 1156@cindex @code{__imag__} keyword 1157To extract the real part of a complex-valued expression @var{exp}, write 1158@code{__real__ @var{exp}}. Likewise, use @code{__imag__} to 1159extract the imaginary part. This is a GNU extension; for values of 1160floating type, you should use the ISO C99 functions @code{crealf}, 1161@code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and 1162@code{cimagl}, declared in @code{<complex.h>} and also provided as 1163built-in functions by GCC@. 1164 1165@cindex complex conjugation 1166The operator @samp{~} performs complex conjugation when used on a value 1167with a complex type. This is a GNU extension; for values of 1168floating type, you should use the ISO C99 functions @code{conjf}, 1169@code{conj} and @code{conjl}, declared in @code{<complex.h>} and also 1170provided as built-in functions by GCC@. 1171 1172GCC can allocate complex automatic variables in a noncontiguous 1173fashion; it's even possible for the real part to be in a register while 1174the imaginary part is on the stack (or vice-versa). None of the 1175supported debugging info formats has a way to represent noncontiguous 1176allocation like this, so GCC describes a noncontiguous complex 1177variable as if it were two separate variables of noncomplex type. 1178If the variable's actual name is @code{foo}, the two fictitious 1179variables are named @code{foo$real} and @code{foo$imag}. You can 1180examine and set these two fictitious variables with your debugger. 1181 1182A future version of GDB will know how to recognize such pairs and treat 1183them as a single variable with a complex type. 1184 1185@node Hex Floats 1186@section Hex Floats 1187@cindex hex floats 1188 1189ISO C99 supports floating-point numbers written not only in the usual 1190decimal notation, such as @code{1.55e1}, but also numbers such as 1191@code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC 1192supports this in C89 mode (except in some cases when strictly 1193conforming) and in C++. In that format the 1194@samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are 1195mandatory. The exponent is a decimal number that indicates the power of 11962 by which the significant part will be multiplied. Thus @samp{0x1.f} is 1197@tex 1198$1 {15\over16}$, 1199@end tex 1200@ifnottex 12011 15/16, 1202@end ifnottex 1203@samp{p3} multiplies it by 8, and the value of @code{0x1.fp3} 1204is the same as @code{1.55e1}. 1205 1206Unlike for floating-point numbers in the decimal notation the exponent 1207is always required in the hexadecimal notation. Otherwise the compiler 1208would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This 1209could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the 1210extension for floating-point constants of type @code{float}. 1211 1212@node Zero Length 1213@section Arrays of Length Zero 1214@cindex arrays of length zero 1215@cindex zero-length arrays 1216@cindex length-zero arrays 1217@cindex flexible array members 1218 1219Zero-length arrays are allowed in GNU C@. They are very useful as the 1220last element of a structure which is really a header for a variable-length 1221object: 1222 1223@example 1224struct line @{ 1225 int length; 1226 char contents[0]; 1227@}; 1228 1229struct line *thisline = (struct line *) 1230 malloc (sizeof (struct line) + this_length); 1231thisline->length = this_length; 1232@end example 1233 1234In ISO C89, you would have to give @code{contents} a length of 1, which 1235means either you waste space or complicate the argument to @code{malloc}. 1236 1237In ISO C99, you would use a @dfn{flexible array member}, which is 1238slightly different in syntax and semantics: 1239 1240@itemize @bullet 1241@item 1242Flexible array members are written as @code{contents[]} without 1243the @code{0}. 1244 1245@item 1246Flexible array members have incomplete type, and so the @code{sizeof} 1247operator may not be applied. As a quirk of the original implementation 1248of zero-length arrays, @code{sizeof} evaluates to zero. 1249 1250@item 1251Flexible array members may only appear as the last member of a 1252@code{struct} that is otherwise non-empty. 1253@end itemize 1254 1255GCC versions before 3.0 allowed zero-length arrays to be statically 1256initialized, as if they were flexible arrays. In addition to those 1257cases that were useful, it also allowed initializations in situations 1258that would corrupt later data. Non-empty initialization of zero-length 1259arrays is now treated like any case where there are more initializer 1260elements than the array holds, in that a suitable warning about "excess 1261elements in array" is given, and the excess elements (all of them, in 1262this case) are ignored. 1263 1264Instead GCC allows static initialization of flexible array members. 1265This is equivalent to defining a new structure containing the original 1266structure followed by an array of sufficient size to contain the data. 1267I.e.@: in the following, @code{f1} is constructed as if it were declared 1268like @code{f2}. 1269 1270@example 1271struct f1 @{ 1272 int x; int y[]; 1273@} f1 = @{ 1, @{ 2, 3, 4 @} @}; 1274 1275struct f2 @{ 1276 struct f1 f1; int data[3]; 1277@} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @}; 1278@end example 1279 1280@noindent 1281The convenience of this extension is that @code{f1} has the desired 1282type, eliminating the need to consistently refer to @code{f2.f1}. 1283 1284This has symmetry with normal static arrays, in that an array of 1285unknown size is also written with @code{[]}. 1286 1287Of course, this extension only makes sense if the extra data comes at 1288the end of a top-level object, as otherwise we would be overwriting 1289data at subsequent offsets. To avoid undue complication and confusion 1290with initialization of deeply nested arrays, we simply disallow any 1291non-empty initialization except when the structure is the top-level 1292object. For example: 1293 1294@example 1295struct foo @{ int x; int y[]; @}; 1296struct bar @{ struct foo z; @}; 1297 1298struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.} 1299struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 1300struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.} 1301struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 1302@end example 1303 1304@node Variable Length 1305@section Arrays of Variable Length 1306@cindex variable-length arrays 1307@cindex arrays of variable length 1308@cindex VLAs 1309 1310Variable-length automatic arrays are allowed in ISO C99, and as an 1311extension GCC accepts them in C89 mode and in C++. (However, GCC's 1312implementation of variable-length arrays does not yet conform in detail 1313to the ISO C99 standard.) These arrays are 1314declared like any other automatic arrays, but with a length that is not 1315a constant expression. The storage is allocated at the point of 1316declaration and deallocated when the brace-level is exited. For 1317example: 1318 1319@example 1320FILE * 1321concat_fopen (char *s1, char *s2, char *mode) 1322@{ 1323 char str[strlen (s1) + strlen (s2) + 1]; 1324 strcpy (str, s1); 1325 strcat (str, s2); 1326 return fopen (str, mode); 1327@} 1328@end example 1329 1330@cindex scope of a variable length array 1331@cindex variable-length array scope 1332@cindex deallocating variable length arrays 1333Jumping or breaking out of the scope of the array name deallocates the 1334storage. Jumping into the scope is not allowed; you get an error 1335message for it. 1336 1337@cindex @code{alloca} vs variable-length arrays 1338You can use the function @code{alloca} to get an effect much like 1339variable-length arrays. The function @code{alloca} is available in 1340many other C implementations (but not in all). On the other hand, 1341variable-length arrays are more elegant. 1342 1343There are other differences between these two methods. Space allocated 1344with @code{alloca} exists until the containing @emph{function} returns. 1345The space for a variable-length array is deallocated as soon as the array 1346name's scope ends. (If you use both variable-length arrays and 1347@code{alloca} in the same function, deallocation of a variable-length array 1348will also deallocate anything more recently allocated with @code{alloca}.) 1349 1350You can also use variable-length arrays as arguments to functions: 1351 1352@example 1353struct entry 1354tester (int len, char data[len][len]) 1355@{ 1356 @dots{} 1357@} 1358@end example 1359 1360The length of an array is computed once when the storage is allocated 1361and is remembered for the scope of the array in case you access it with 1362@code{sizeof}. 1363 1364If you want to pass the array first and the length afterward, you can 1365use a forward declaration in the parameter list---another GNU extension. 1366 1367@example 1368struct entry 1369tester (int len; char data[len][len], int len) 1370@{ 1371 @dots{} 1372@} 1373@end example 1374 1375@cindex parameter forward declaration 1376The @samp{int len} before the semicolon is a @dfn{parameter forward 1377declaration}, and it serves the purpose of making the name @code{len} 1378known when the declaration of @code{data} is parsed. 1379 1380You can write any number of such parameter forward declarations in the 1381parameter list. They can be separated by commas or semicolons, but the 1382last one must end with a semicolon, which is followed by the ``real'' 1383parameter declarations. Each forward declaration must match a ``real'' 1384declaration in parameter name and data type. ISO C99 does not support 1385parameter forward declarations. 1386 1387@node Variadic Macros 1388@section Macros with a Variable Number of Arguments. 1389@cindex variable number of arguments 1390@cindex macro with variable arguments 1391@cindex rest argument (in macro) 1392@cindex variadic macros 1393 1394In the ISO C standard of 1999, a macro can be declared to accept a 1395variable number of arguments much as a function can. The syntax for 1396defining the macro is similar to that of a function. Here is an 1397example: 1398 1399@example 1400#define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) 1401@end example 1402 1403Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of 1404such a macro, it represents the zero or more tokens until the closing 1405parenthesis that ends the invocation, including any commas. This set of 1406tokens replaces the identifier @code{__VA_ARGS__} in the macro body 1407wherever it appears. See the CPP manual for more information. 1408 1409GCC has long supported variadic macros, and used a different syntax that 1410allowed you to give a name to the variable arguments just like any other 1411argument. Here is an example: 1412 1413@example 1414#define debug(format, args...) fprintf (stderr, format, args) 1415@end example 1416 1417This is in all ways equivalent to the ISO C example above, but arguably 1418more readable and descriptive. 1419 1420GNU CPP has two further variadic macro extensions, and permits them to 1421be used with either of the above forms of macro definition. 1422 1423In standard C, you are not allowed to leave the variable argument out 1424entirely; but you are allowed to pass an empty argument. For example, 1425this invocation is invalid in ISO C, because there is no comma after 1426the string: 1427 1428@example 1429debug ("A message") 1430@end example 1431 1432GNU CPP permits you to completely omit the variable arguments in this 1433way. In the above examples, the compiler would complain, though since 1434the expansion of the macro still has the extra comma after the format 1435string. 1436 1437To help solve this problem, CPP behaves specially for variable arguments 1438used with the token paste operator, @samp{##}. If instead you write 1439 1440@example 1441#define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) 1442@end example 1443 1444and if the variable arguments are omitted or empty, the @samp{##} 1445operator causes the preprocessor to remove the comma before it. If you 1446do provide some variable arguments in your macro invocation, GNU CPP 1447does not complain about the paste operation and instead places the 1448variable arguments after the comma. Just like any other pasted macro 1449argument, these arguments are not macro expanded. 1450 1451@node Escaped Newlines 1452@section Slightly Looser Rules for Escaped Newlines 1453@cindex escaped newlines 1454@cindex newlines (escaped) 1455 1456Recently, the non-traditional preprocessor has relaxed its treatment of 1457escaped newlines. Previously, the newline had to immediately follow a 1458backslash. The current implementation allows whitespace in the form of 1459spaces, horizontal and vertical tabs, and form feeds between the 1460backslash and the subsequent newline. The preprocessor issues a 1461warning, but treats it as a valid escaped newline and combines the two 1462lines to form a single logical line. This works within comments and 1463tokens, including multi-line strings, as well as between tokens. 1464Comments are @emph{not} treated as whitespace for the purposes of this 1465relaxation, since they have not yet been replaced with spaces. 1466 1467@node Multi-line Strings 1468@section String Literals with Embedded Newlines 1469@cindex multi-line string literals 1470 1471As an extension, GNU CPP permits string literals to cross multiple lines 1472without escaping the embedded newlines. Each embedded newline is 1473replaced with a single @samp{\n} character in the resulting string 1474literal, regardless of what form the newline took originally. 1475 1476CPP currently allows such strings in directives as well (other than the 1477@samp{#include} family). This is deprecated and will eventually be 1478removed. 1479 1480@node Subscripting 1481@section Non-Lvalue Arrays May Have Subscripts 1482@cindex subscripting 1483@cindex arrays, non-lvalue 1484 1485@cindex subscripting and function values 1486In ISO C99, arrays that are not lvalues still decay to pointers, and 1487may be subscripted, although they may not be modified or used after 1488the next sequence point and the unary @samp{&} operator may not be 1489applied to them. As an extension, GCC allows such arrays to be 1490subscripted in C89 mode, though otherwise they do not decay to 1491pointers outside C99 mode. For example, 1492this is valid in GNU C though not valid in C89: 1493 1494@example 1495@group 1496struct foo @{int a[4];@}; 1497 1498struct foo f(); 1499 1500bar (int index) 1501@{ 1502 return f().a[index]; 1503@} 1504@end group 1505@end example 1506 1507@node Pointer Arith 1508@section Arithmetic on @code{void}- and Function-Pointers 1509@cindex void pointers, arithmetic 1510@cindex void, size of pointer to 1511@cindex function pointers, arithmetic 1512@cindex function, size of pointer to 1513 1514In GNU C, addition and subtraction operations are supported on pointers to 1515@code{void} and on pointers to functions. This is done by treating the 1516size of a @code{void} or of a function as 1. 1517 1518A consequence of this is that @code{sizeof} is also allowed on @code{void} 1519and on function types, and returns 1. 1520 1521@opindex Wpointer-arith 1522The option @option{-Wpointer-arith} requests a warning if these extensions 1523are used. 1524 1525@node Initializers 1526@section Non-Constant Initializers 1527@cindex initializers, non-constant 1528@cindex non-constant initializers 1529 1530As in standard C++ and ISO C99, the elements of an aggregate initializer for an 1531automatic variable are not required to be constant expressions in GNU C@. 1532Here is an example of an initializer with run-time varying elements: 1533 1534@example 1535foo (float f, float g) 1536@{ 1537 float beat_freqs[2] = @{ f-g, f+g @}; 1538 @dots{} 1539@} 1540@end example 1541 1542@node Compound Literals 1543@section Compound Literals 1544@cindex constructor expressions 1545@cindex initializations in expressions 1546@cindex structures, constructor expression 1547@cindex expressions, constructor 1548@cindex compound literals 1549@c The GNU C name for what C99 calls compound literals was "constructor expressions". 1550 1551ISO C99 supports compound literals. A compound literal looks like 1552a cast containing an initializer. Its value is an object of the 1553type specified in the cast, containing the elements specified in 1554the initializer; it is an lvalue. As an extension, GCC supports 1555compound literals in C89 mode and in C++. 1556 1557Usually, the specified type is a structure. Assume that 1558@code{struct foo} and @code{structure} are declared as shown: 1559 1560@example 1561struct foo @{int a; char b[2];@} structure; 1562@end example 1563 1564@noindent 1565Here is an example of constructing a @code{struct foo} with a compound literal: 1566 1567@example 1568structure = ((struct foo) @{x + y, 'a', 0@}); 1569@end example 1570 1571@noindent 1572This is equivalent to writing the following: 1573 1574@example 1575@{ 1576 struct foo temp = @{x + y, 'a', 0@}; 1577 structure = temp; 1578@} 1579@end example 1580 1581You can also construct an array. If all the elements of the compound literal 1582are (made up of) simple constant expressions, suitable for use in 1583initializers of objects of static storage duration, then the compound 1584literal can be coerced to a pointer to its first element and used in 1585such an initializer, as shown here: 1586 1587@example 1588char **foo = (char *[]) @{ "x", "y", "z" @}; 1589@end example 1590 1591Compound literals for scalar types and union types are is 1592also allowed, but then the compound literal is equivalent 1593to a cast. 1594 1595As a GNU extension, GCC allows initialization of objects with static storage 1596duration by compound literals (which is not possible in ISO C99, because 1597the initializer is not a constant). 1598It is handled as if the object was initialized only with the bracket 1599enclosed list if compound literal's and object types match. 1600The initializer list of the compound literal must be constant. 1601If the object being initialized has array type of unknown size, the size is 1602determined by compound literal size. 1603 1604@example 1605static struct foo x = (struct foo) @{1, 'a', 'b'@}; 1606static int y[] = (int []) @{1, 2, 3@}; 1607static int z[] = (int [3]) @{1@}; 1608@end example 1609 1610@noindent 1611The above lines are equivalent to the following: 1612@example 1613static struct foo x = @{1, 'a', 'b'@}; 1614static int y[] = @{1, 2, 3@}; 1615static int z[] = @{1, 0, 0@}; 1616@end example 1617 1618@node Designated Inits 1619@section Designated Initializers 1620@cindex initializers with labeled elements 1621@cindex labeled elements in initializers 1622@cindex case labels in initializers 1623@cindex designated initializers 1624 1625Standard C89 requires the elements of an initializer to appear in a fixed 1626order, the same as the order of the elements in the array or structure 1627being initialized. 1628 1629In ISO C99 you can give the elements in any order, specifying the array 1630indices or structure field names they apply to, and GNU C allows this as 1631an extension in C89 mode as well. This extension is not 1632implemented in GNU C++. 1633 1634To specify an array index, write 1635@samp{[@var{index}] =} before the element value. For example, 1636 1637@example 1638int a[6] = @{ [4] = 29, [2] = 15 @}; 1639@end example 1640 1641@noindent 1642is equivalent to 1643 1644@example 1645int a[6] = @{ 0, 0, 15, 0, 29, 0 @}; 1646@end example 1647 1648@noindent 1649The index values must be constant expressions, even if the array being 1650initialized is automatic. 1651 1652An alternative syntax for this which has been obsolete since GCC 2.5 but 1653GCC still accepts is to write @samp{[@var{index}]} before the element 1654value, with no @samp{=}. 1655 1656To initialize a range of elements to the same value, write 1657@samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU 1658extension. For example, 1659 1660@example 1661int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @}; 1662@end example 1663 1664@noindent 1665If the value in it has side-effects, the side-effects will happen only once, 1666not for each initialized field by the range initializer. 1667 1668@noindent 1669Note that the length of the array is the highest value specified 1670plus one. 1671 1672In a structure initializer, specify the name of a field to initialize 1673with @samp{.@var{fieldname} =} before the element value. For example, 1674given the following structure, 1675 1676@example 1677struct point @{ int x, y; @}; 1678@end example 1679 1680@noindent 1681the following initialization 1682 1683@example 1684struct point p = @{ .y = yvalue, .x = xvalue @}; 1685@end example 1686 1687@noindent 1688is equivalent to 1689 1690@example 1691struct point p = @{ xvalue, yvalue @}; 1692@end example 1693 1694Another syntax which has the same meaning, obsolete since GCC 2.5, is 1695@samp{@var{fieldname}:}, as shown here: 1696 1697@example 1698struct point p = @{ y: yvalue, x: xvalue @}; 1699@end example 1700 1701@cindex designators 1702The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a 1703@dfn{designator}. You can also use a designator (or the obsolete colon 1704syntax) when initializing a union, to specify which element of the union 1705should be used. For example, 1706 1707@example 1708union foo @{ int i; double d; @}; 1709 1710union foo f = @{ .d = 4 @}; 1711@end example 1712 1713@noindent 1714will convert 4 to a @code{double} to store it in the union using 1715the second element. By contrast, casting 4 to type @code{union foo} 1716would store it into the union as the integer @code{i}, since it is 1717an integer. (@xref{Cast to Union}.) 1718 1719You can combine this technique of naming elements with ordinary C 1720initialization of successive elements. Each initializer element that 1721does not have a designator applies to the next consecutive element of the 1722array or structure. For example, 1723 1724@example 1725int a[6] = @{ [1] = v1, v2, [4] = v4 @}; 1726@end example 1727 1728@noindent 1729is equivalent to 1730 1731@example 1732int a[6] = @{ 0, v1, v2, 0, v4, 0 @}; 1733@end example 1734 1735Labeling the elements of an array initializer is especially useful 1736when the indices are characters or belong to an @code{enum} type. 1737For example: 1738 1739@example 1740int whitespace[256] 1741 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1, 1742 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @}; 1743@end example 1744 1745@cindex designator lists 1746You can also write a series of @samp{.@var{fieldname}} and 1747@samp{[@var{index}]} designators before an @samp{=} to specify a 1748nested subobject to initialize; the list is taken relative to the 1749subobject corresponding to the closest surrounding brace pair. For 1750example, with the @samp{struct point} declaration above: 1751 1752@example 1753struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @}; 1754@end example 1755 1756@noindent 1757If the same field is initialized multiple times, it will have value from 1758the last initialization. If any such overridden initialization has 1759side-effect, it is unspecified whether the side-effect happens or not. 1760Currently, gcc will discard them and issue a warning. 1761 1762@node Case Ranges 1763@section Case Ranges 1764@cindex case ranges 1765@cindex ranges in case statements 1766 1767You can specify a range of consecutive values in a single @code{case} label, 1768like this: 1769 1770@example 1771case @var{low} ... @var{high}: 1772@end example 1773 1774@noindent 1775This has the same effect as the proper number of individual @code{case} 1776labels, one for each integer value from @var{low} to @var{high}, inclusive. 1777 1778This feature is especially useful for ranges of ASCII character codes: 1779 1780@example 1781case 'A' ... 'Z': 1782@end example 1783 1784@strong{Be careful:} Write spaces around the @code{...}, for otherwise 1785it may be parsed wrong when you use it with integer values. For example, 1786write this: 1787 1788@example 1789case 1 ... 5: 1790@end example 1791 1792@noindent 1793rather than this: 1794 1795@example 1796case 1...5: 1797@end example 1798 1799@node Cast to Union 1800@section Cast to a Union Type 1801@cindex cast to a union 1802@cindex union, casting to a 1803 1804A cast to union type is similar to other casts, except that the type 1805specified is a union type. You can specify the type either with 1806@code{union @var{tag}} or with a typedef name. A cast to union is actually 1807a constructor though, not a cast, and hence does not yield an lvalue like 1808normal casts. (@xref{Compound Literals}.) 1809 1810The types that may be cast to the union type are those of the members 1811of the union. Thus, given the following union and variables: 1812 1813@example 1814union foo @{ int i; double d; @}; 1815int x; 1816double y; 1817@end example 1818 1819@noindent 1820both @code{x} and @code{y} can be cast to type @code{union foo}. 1821 1822Using the cast as the right-hand side of an assignment to a variable of 1823union type is equivalent to storing in a member of the union: 1824 1825@example 1826union foo u; 1827@dots{} 1828u = (union foo) x @equiv{} u.i = x 1829u = (union foo) y @equiv{} u.d = y 1830@end example 1831 1832You can also use the union cast as a function argument: 1833 1834@example 1835void hack (union foo); 1836@dots{} 1837hack ((union foo) x); 1838@end example 1839 1840@node Mixed Declarations 1841@section Mixed Declarations and Code 1842@cindex mixed declarations and code 1843@cindex declarations, mixed with code 1844@cindex code, mixed with declarations 1845 1846ISO C99 and ISO C++ allow declarations and code to be freely mixed 1847within compound statements. As an extension, GCC also allows this in 1848C89 mode. For example, you could do: 1849 1850@example 1851int i; 1852@dots{} 1853i++; 1854int j = i + 2; 1855@end example 1856 1857Each identifier is visible from where it is declared until the end of 1858the enclosing block. 1859 1860@node Function Attributes 1861@section Declaring Attributes of Functions 1862@cindex function attributes 1863@cindex declaring attributes of functions 1864@cindex functions that never return 1865@cindex functions that have no side effects 1866@cindex functions in arbitrary sections 1867@cindex functions that behave like malloc 1868@cindex @code{volatile} applied to function 1869@cindex @code{const} applied to function 1870@cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments 1871@cindex functions that are passed arguments in registers on the 386 1872@cindex functions that pop the argument stack on the 386 1873@cindex functions that do not pop the argument stack on the 386 1874 1875In GNU C, you declare certain things about functions called in your program 1876which help the compiler optimize function calls and check your code more 1877carefully. 1878 1879The keyword @code{__attribute__} allows you to specify special 1880attributes when making a declaration. This keyword is followed by an 1881attribute specification inside double parentheses. The following 1882attributes are currently defined for functions on all targets: 1883@code{noreturn}, @code{noinline}, @code{always_inline}, 1884@code{pure}, @code{const}, 1885@code{format}, @code{format_arg}, @code{no_instrument_function}, 1886@code{section}, @code{constructor}, @code{destructor}, @code{used}, 1887@code{unused}, @code{deprecated}, @code{weak}, @code{malloc}, and 1888@code{alias}. Several other attributes are defined for functions on 1889particular target systems. Other attributes, including @code{section} 1890are supported for variables declarations (@pxref{Variable Attributes}) 1891and for types (@pxref{Type Attributes}). 1892 1893You may also specify attributes with @samp{__} preceding and following 1894each keyword. This allows you to use them in header files without 1895being concerned about a possible macro of the same name. For example, 1896you may use @code{__noreturn__} instead of @code{noreturn}. 1897 1898@xref{Attribute Syntax}, for details of the exact syntax for using 1899attributes. 1900 1901@table @code 1902@cindex @code{noreturn} function attribute 1903@item noreturn 1904A few standard library functions, such as @code{abort} and @code{exit}, 1905cannot return. GCC knows this automatically. Some programs define 1906their own functions that never return. You can declare them 1907@code{noreturn} to tell the compiler this fact. For example, 1908 1909@smallexample 1910@group 1911void fatal () __attribute__ ((noreturn)); 1912 1913void 1914fatal (@dots{}) 1915@{ 1916 @dots{} /* @r{Print error message.} */ @dots{} 1917 exit (1); 1918@} 1919@end group 1920@end smallexample 1921 1922The @code{noreturn} keyword tells the compiler to assume that 1923@code{fatal} cannot return. It can then optimize without regard to what 1924would happen if @code{fatal} ever did return. This makes slightly 1925better code. More importantly, it helps avoid spurious warnings of 1926uninitialized variables. 1927 1928Do not assume that registers saved by the calling function are 1929restored before calling the @code{noreturn} function. 1930 1931It does not make sense for a @code{noreturn} function to have a return 1932type other than @code{void}. 1933 1934The attribute @code{noreturn} is not implemented in GCC versions 1935earlier than 2.5. An alternative way to declare that a function does 1936not return, which works in the current version and in some older 1937versions, is as follows: 1938 1939@smallexample 1940typedef void voidfn (); 1941 1942volatile voidfn fatal; 1943@end smallexample 1944 1945@cindex @code{noinline} function attribute 1946@item noinline 1947This function attribute prevents a function from being considered for 1948inlining. 1949 1950@cindex @code{always_inline} function attribute 1951@item always_inline 1952Generally, functions are not inlined unless optimization is specified. 1953For functions declared inline, this attribute inlines the function even 1954if no optimization level was specified. 1955 1956@cindex @code{pure} function attribute 1957@item pure 1958Many functions have no effects except the return value and their 1959return value depends only on the parameters and/or global variables. 1960Such a function can be subject 1961to common subexpression elimination and loop optimization just as an 1962arithmetic operator would be. These functions should be declared 1963with the attribute @code{pure}. For example, 1964 1965@smallexample 1966int square (int) __attribute__ ((pure)); 1967@end smallexample 1968 1969@noindent 1970says that the hypothetical function @code{square} is safe to call 1971fewer times than the program says. 1972 1973Some of common examples of pure functions are @code{strlen} or @code{memcmp}. 1974Interesting non-pure functions are functions with infinite loops or those 1975depending on volatile memory or other system resource, that may change between 1976two consecutive calls (such as @code{feof} in a multithreading environment). 1977 1978The attribute @code{pure} is not implemented in GCC versions earlier 1979than 2.96. 1980@cindex @code{const} function attribute 1981@item const 1982Many functions do not examine any values except their arguments, and 1983have no effects except the return value. Basically this is just slightly 1984more strict class than the @code{pure} attribute above, since function is not 1985allowed to read global memory. 1986 1987@cindex pointer arguments 1988Note that a function that has pointer arguments and examines the data 1989pointed to must @emph{not} be declared @code{const}. Likewise, a 1990function that calls a non-@code{const} function usually must not be 1991@code{const}. It does not make sense for a @code{const} function to 1992return @code{void}. 1993 1994The attribute @code{const} is not implemented in GCC versions earlier 1995than 2.5. An alternative way to declare that a function has no side 1996effects, which works in the current version and in some older versions, 1997is as follows: 1998 1999@smallexample 2000typedef int intfn (); 2001 2002extern const intfn square; 2003@end smallexample 2004 2005This approach does not work in GNU C++ from 2.6.0 on, since the language 2006specifies that the @samp{const} must be attached to the return value. 2007 2008 2009@item format (@var{archetype}, @var{string-index}, @var{first-to-check}) 2010@cindex @code{format} function attribute 2011@opindex Wformat 2012The @code{format} attribute specifies that a function takes @code{printf}, 2013@code{scanf}, @code{strftime} or @code{strfmon} style arguments which 2014should be type-checked against a format string. For example, the 2015declaration: 2016 2017@smallexample 2018extern int 2019my_printf (void *my_object, const char *my_format, ...) 2020 __attribute__ ((format (printf, 2, 3))); 2021@end smallexample 2022 2023@noindent 2024causes the compiler to check the arguments in calls to @code{my_printf} 2025for consistency with the @code{printf} style format string argument 2026@code{my_format}. 2027 2028The parameter @var{archetype} determines how the format string is 2029interpreted, and should be @code{printf}, @code{scanf}, @code{strftime} 2030or @code{strfmon}. (You can also use @code{__printf__}, 2031@code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The 2032parameter @var{string-index} specifies which argument is the format 2033string argument (starting from 1), while @var{first-to-check} is the 2034number of the first argument to check against the format string. For 2035functions where the arguments are not available to be checked (such as 2036@code{vprintf}), specify the third parameter as zero. In this case the 2037compiler only checks the format string for consistency. For 2038@code{strftime} formats, the third parameter is required to be zero. 2039 2040In the example above, the format string (@code{my_format}) is the second 2041argument of the function @code{my_print}, and the arguments to check 2042start with the third argument, so the correct parameters for the format 2043attribute are 2 and 3. 2044 2045@opindex ffreestanding 2046The @code{format} attribute allows you to identify your own functions 2047which take format strings as arguments, so that GCC can check the 2048calls to these functions for errors. The compiler always (unless 2049@option{-ffreestanding} is used) checks formats 2050for the standard library functions @code{printf}, @code{fprintf}, 2051@code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime}, 2052@code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such 2053warnings are requested (using @option{-Wformat}), so there is no need to 2054modify the header file @file{stdio.h}. In C99 mode, the functions 2055@code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and 2056@code{vsscanf} are also checked. Except in strictly conforming C 2057standard modes, the X/Open function @code{strfmon} is also checked as 2058are @code{printf_unlocked} and @code{fprintf_unlocked}. 2059@xref{C Dialect Options,,Options Controlling C Dialect}. 2060 2061@item format_arg (@var{string-index}) 2062@cindex @code{format_arg} function attribute 2063@opindex Wformat-nonliteral 2064The @code{format_arg} attribute specifies that a function takes a format 2065string for a @code{printf}, @code{scanf}, @code{strftime} or 2066@code{strfmon} style function and modifies it (for example, to translate 2067it into another language), so the result can be passed to a 2068@code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style 2069function (with the remaining arguments to the format function the same 2070as they would have been for the unmodified string). For example, the 2071declaration: 2072 2073@smallexample 2074extern char * 2075my_dgettext (char *my_domain, const char *my_format) 2076 __attribute__ ((format_arg (2))); 2077@end smallexample 2078 2079@noindent 2080causes the compiler to check the arguments in calls to a @code{printf}, 2081@code{scanf}, @code{strftime} or @code{strfmon} type function, whose 2082format string argument is a call to the @code{my_dgettext} function, for 2083consistency with the format string argument @code{my_format}. If the 2084@code{format_arg} attribute had not been specified, all the compiler 2085could tell in such calls to format functions would be that the format 2086string argument is not constant; this would generate a warning when 2087@option{-Wformat-nonliteral} is used, but the calls could not be checked 2088without the attribute. 2089 2090The parameter @var{string-index} specifies which argument is the format 2091string argument (starting from 1). 2092 2093The @code{format-arg} attribute allows you to identify your own 2094functions which modify format strings, so that GCC can check the 2095calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} 2096type function whose operands are a call to one of your own function. 2097The compiler always treats @code{gettext}, @code{dgettext}, and 2098@code{dcgettext} in this manner except when strict ISO C support is 2099requested by @option{-ansi} or an appropriate @option{-std} option, or 2100@option{-ffreestanding} is used. @xref{C Dialect Options,,Options 2101Controlling C Dialect}. 2102 2103@item no_instrument_function 2104@cindex @code{no_instrument_function} function attribute 2105@opindex finstrument-functions 2106If @option{-finstrument-functions} is given, profiling function calls will 2107be generated at entry and exit of most user-compiled functions. 2108Functions with this attribute will not be so instrumented. 2109 2110@item section ("@var{section-name}") 2111@cindex @code{section} function attribute 2112Normally, the compiler places the code it generates in the @code{text} section. 2113Sometimes, however, you need additional sections, or you need certain 2114particular functions to appear in special sections. The @code{section} 2115attribute specifies that a function lives in a particular section. 2116For example, the declaration: 2117 2118@smallexample 2119extern void foobar (void) __attribute__ ((section ("bar"))); 2120@end smallexample 2121 2122@noindent 2123puts the function @code{foobar} in the @code{bar} section. 2124 2125Some file formats do not support arbitrary sections so the @code{section} 2126attribute is not available on all platforms. 2127If you need to map the entire contents of a module to a particular 2128section, consider using the facilities of the linker instead. 2129 2130@item constructor 2131@itemx destructor 2132@cindex @code{constructor} function attribute 2133@cindex @code{destructor} function attribute 2134The @code{constructor} attribute causes the function to be called 2135automatically before execution enters @code{main ()}. Similarly, the 2136@code{destructor} attribute causes the function to be called 2137automatically after @code{main ()} has completed or @code{exit ()} has 2138been called. Functions with these attributes are useful for 2139initializing data that will be used implicitly during the execution of 2140the program. 2141 2142These attributes are not currently implemented for Objective-C@. 2143 2144@cindex @code{unused} attribute. 2145@item unused 2146This attribute, attached to a function, means that the function is meant 2147to be possibly unused. GCC will not produce a warning for this 2148function. GNU C++ does not currently support this attribute as 2149definitions without parameters are valid in C++. 2150 2151@cindex @code{used} attribute. 2152@item used 2153This attribute, attached to a function, means that code must be emitted 2154for the function even if it appears that the function is not referenced. 2155This is useful, for example, when the function is referenced only in 2156inline assembly. 2157 2158@cindex @code{deprecated} attribute. 2159@item deprecated 2160The @code{deprecated} attribute results in a warning if the function 2161is used anywhere in the source file. This is useful when identifying 2162functions that are expected to be removed in a future version of a 2163program. The warning also includes the location of the declaration 2164of the deprecated function, to enable users to easily find further 2165information about why the function is deprecated, or what they should 2166do instead. Note that the warnings only occurs for uses: 2167 2168@smallexample 2169int old_fn () __attribute__ ((deprecated)); 2170int old_fn (); 2171int (*fn_ptr)() = old_fn; 2172@end smallexample 2173 2174results in a warning on line 3 but not line 2. 2175 2176The @code{deprecated} attribute can also be used for variables and 2177types (@pxref{Variable Attributes}, @pxref{Type Attributes}.) 2178 2179@item weak 2180@cindex @code{weak} attribute 2181The @code{weak} attribute causes the declaration to be emitted as a weak 2182symbol rather than a global. This is primarily useful in defining 2183library functions which can be overridden in user code, though it can 2184also be used with non-function declarations. Weak symbols are supported 2185for ELF targets, and also for a.out targets when using the GNU assembler 2186and linker. 2187 2188@item malloc 2189@cindex @code{malloc} attribute 2190The @code{malloc} attribute is used to tell the compiler that a function 2191may be treated as if it were the malloc function. The compiler assumes 2192that calls to malloc result in a pointers that cannot alias anything. 2193This will often improve optimization. 2194 2195@item alias ("@var{target}") 2196@cindex @code{alias} attribute 2197The @code{alias} attribute causes the declaration to be emitted as an 2198alias for another symbol, which must be specified. For instance, 2199 2200@smallexample 2201void __f () @{ /* do something */; @} 2202void f () __attribute__ ((weak, alias ("__f"))); 2203@end smallexample 2204 2205declares @samp{f} to be a weak alias for @samp{__f}. In C++, the 2206mangled name for the target must be used. 2207 2208Not all target machines support this attribute. 2209 2210@item regparm (@var{number}) 2211@cindex functions that are passed arguments in registers on the 386 2212On the Intel 386, the @code{regparm} attribute causes the compiler to 2213pass up to @var{number} integer arguments in registers EAX, 2214EDX, and ECX instead of on the stack. Functions that take a 2215variable number of arguments will continue to be passed all of their 2216arguments on the stack. 2217 2218@item stdcall 2219@cindex functions that pop the argument stack on the 386 2220On the Intel 386, the @code{stdcall} attribute causes the compiler to 2221assume that the called function will pop off the stack space used to 2222pass arguments, unless it takes a variable number of arguments. 2223 2224The PowerPC compiler for Windows NT currently ignores the @code{stdcall} 2225attribute. 2226 2227@item cdecl 2228@cindex functions that do pop the argument stack on the 386 2229@opindex mrtd 2230On the Intel 386, the @code{cdecl} attribute causes the compiler to 2231assume that the calling function will pop off the stack space used to 2232pass arguments. This is 2233useful to override the effects of the @option{-mrtd} switch. 2234 2235The PowerPC compiler for Windows NT currently ignores the @code{cdecl} 2236attribute. 2237 2238@item longcall 2239@cindex functions called via pointer on the RS/6000 and PowerPC 2240On the RS/6000 and PowerPC, the @code{longcall} attribute causes the 2241compiler to always call the function via a pointer, so that functions 2242which reside further than 64 megabytes (67,108,864 bytes) from the 2243current location can be called. 2244 2245@item long_call/short_call 2246@cindex indirect calls on ARM 2247This attribute allows to specify how to call a particular function on 2248ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options}) 2249command line switch and @code{#pragma long_calls} settings. The 2250@code{long_call} attribute causes the compiler to always call the 2251function by first loading its address into a register and then using the 2252contents of that register. The @code{short_call} attribute always places 2253the offset to the function from the call site into the @samp{BL} 2254instruction directly. 2255 2256@item dllimport 2257@cindex functions which are imported from a dll on PowerPC Windows NT 2258On the PowerPC running Windows NT, the @code{dllimport} attribute causes 2259the compiler to call the function via a global pointer to the function 2260pointer that is set up by the Windows NT dll library. The pointer name 2261is formed by combining @code{__imp_} and the function name. 2262 2263@item dllexport 2264@cindex functions which are exported from a dll on PowerPC Windows NT 2265On the PowerPC running Windows NT, the @code{dllexport} attribute causes 2266the compiler to provide a global pointer to the function pointer, so 2267that it can be called with the @code{dllimport} attribute. The pointer 2268name is formed by combining @code{__imp_} and the function name. 2269 2270@item exception (@var{except-func} [, @var{except-arg}]) 2271@cindex functions which specify exception handling on PowerPC Windows NT 2272On the PowerPC running Windows NT, the @code{exception} attribute causes 2273the compiler to modify the structured exception table entry it emits for 2274the declared function. The string or identifier @var{except-func} is 2275placed in the third entry of the structured exception table. It 2276represents a function, which is called by the exception handling 2277mechanism if an exception occurs. If it was specified, the string or 2278identifier @var{except-arg} is placed in the fourth entry of the 2279structured exception table. 2280 2281@item function_vector 2282@cindex calling functions through the function vector on the H8/300 processors 2283Use this attribute on the H8/300 and H8/300H to indicate that the specified 2284function should be called through the function vector. Calling a 2285function through the function vector will reduce code size, however; 2286the function vector has a limited size (maximum 128 entries on the H8/300 2287and 64 entries on the H8/300H) and shares space with the interrupt vector. 2288 2289You must use GAS and GLD from GNU binutils version 2.7 or later for 2290this attribute to work correctly. 2291 2292@item interrupt 2293@cindex interrupt handler functions 2294Use this attribute on the ARM, AVR, M32R/D and Xstormy16 ports to indicate 2295that the specified function is an interrupt handler. The compiler will 2296generate function entry and exit sequences suitable for use in an 2297interrupt handler when this attribute is present. 2298 2299Note, interrupt handlers for the H8/300, H8/300H and SH processors can 2300be specified via the @code{interrupt_handler} attribute. 2301 2302Note, on the AVR interrupts will be enabled inside the function. 2303 2304Note, for the ARM you can specify the kind of interrupt to be handled by 2305adding an optional parameter to the interrupt attribute like this: 2306 2307@smallexample 2308void f () __attribute__ ((interrupt ("IRQ"))); 2309@end smallexample 2310 2311Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@. 2312 2313@item interrupt_handler 2314@cindex interrupt handler functions on the H8/300 and SH processors 2315Use this attribute on the H8/300, H8/300H and SH to indicate that the 2316specified function is an interrupt handler. The compiler will generate 2317function entry and exit sequences suitable for use in an interrupt 2318handler when this attribute is present. 2319 2320@item sp_switch 2321Use this attribute on the SH to indicate an @code{interrupt_handler} 2322function should switch to an alternate stack. It expects a string 2323argument that names a global variable holding the address of the 2324alternate stack. 2325 2326@smallexample 2327void *alt_stack; 2328void f () __attribute__ ((interrupt_handler, 2329 sp_switch ("alt_stack"))); 2330@end smallexample 2331 2332@item trap_exit 2333Use this attribute on the SH for an @code{interrupt_handle} to return using 2334@code{trapa} instead of @code{rte}. This attribute expects an integer 2335argument specifying the trap number to be used. 2336 2337@item eightbit_data 2338@cindex eight bit data on the H8/300 and H8/300H 2339Use this attribute on the H8/300 and H8/300H to indicate that the specified 2340variable should be placed into the eight bit data section. 2341The compiler will generate more efficient code for certain operations 2342on data in the eight bit data area. Note the eight bit data area is limited to 2343256 bytes of data. 2344 2345You must use GAS and GLD from GNU binutils version 2.7 or later for 2346this attribute to work correctly. 2347 2348@item tiny_data 2349@cindex tiny data section on the H8/300H 2350Use this attribute on the H8/300H to indicate that the specified 2351variable should be placed into the tiny data section. 2352The compiler will generate more efficient code for loads and stores 2353on data in the tiny data section. Note the tiny data area is limited to 2354slightly under 32kbytes of data. 2355 2356@item signal 2357@cindex signal handler functions on the AVR processors 2358Use this attribute on the AVR to indicate that the specified 2359function is an signal handler. The compiler will generate function 2360entry and exit sequences suitable for use in an signal handler when this 2361attribute is present. Interrupts will be disabled inside function. 2362 2363@item naked 2364@cindex function without a prologue/epilogue code 2365Use this attribute on the ARM or AVR ports to indicate that the specified 2366function do not need prologue/epilogue sequences generated by the 2367compiler. It is up to the programmer to provide these sequences. 2368 2369@item model (@var{model-name}) 2370@cindex function addressability on the M32R/D 2371Use this attribute on the M32R/D to set the addressability of an object, 2372and the code generated for a function. 2373The identifier @var{model-name} is one of @code{small}, @code{medium}, 2374or @code{large}, representing each of the code models. 2375 2376Small model objects live in the lower 16MB of memory (so that their 2377addresses can be loaded with the @code{ld24} instruction), and are 2378callable with the @code{bl} instruction. 2379 2380Medium model objects may live anywhere in the 32-bit address space (the 2381compiler will generate @code{seth/add3} instructions to load their addresses), 2382and are callable with the @code{bl} instruction. 2383 2384Large model objects may live anywhere in the 32-bit address space (the 2385compiler will generate @code{seth/add3} instructions to load their addresses), 2386and may not be reachable with the @code{bl} instruction (the compiler will 2387generate the much slower @code{seth/add3/jl} instruction sequence). 2388 2389@end table 2390 2391You can specify multiple attributes in a declaration by separating them 2392by commas within the double parentheses or by immediately following an 2393attribute declaration with another attribute declaration. 2394 2395@cindex @code{#pragma}, reason for not using 2396@cindex pragma, reason for not using 2397Some people object to the @code{__attribute__} feature, suggesting that 2398ISO C's @code{#pragma} should be used instead. At the time 2399@code{__attribute__} was designed, there were two reasons for not doing 2400this. 2401 2402@enumerate 2403@item 2404It is impossible to generate @code{#pragma} commands from a macro. 2405 2406@item 2407There is no telling what the same @code{#pragma} might mean in another 2408compiler. 2409@end enumerate 2410 2411These two reasons applied to almost any application that might have been 2412proposed for @code{#pragma}. It was basically a mistake to use 2413@code{#pragma} for @emph{anything}. 2414 2415The ISO C99 standard includes @code{_Pragma}, which now allows pragmas 2416to be generated from macros. In addition, a @code{#pragma GCC} 2417namespace is now in use for GCC-specific pragmas. However, it has been 2418found convenient to use @code{__attribute__} to achieve a natural 2419attachment of attributes to their corresponding declarations, whereas 2420@code{#pragma GCC} is of use for constructs that do not naturally form 2421part of the grammar. @xref{Other Directives,,Miscellaneous 2422Preprocessing Directives, cpp, The C Preprocessor}. 2423 2424@node Attribute Syntax 2425@section Attribute Syntax 2426@cindex attribute syntax 2427 2428This section describes the syntax with which @code{__attribute__} may be 2429used, and the constructs to which attribute specifiers bind, for the C 2430language. Some details may vary for C++ and Objective-C@. Because of 2431infelicities in the grammar for attributes, some forms described here 2432may not be successfully parsed in all cases. 2433 2434There are some problems with the semantics of attributes in C++. For 2435example, there are no manglings for attributes, although they may affect 2436code generation, so problems may arise when attributed types are used in 2437conjunction with templates or overloading. Similarly, @code{typeid} 2438does not distinguish between types with different attributes. Support 2439for attributes in C++ may be restricted in future to attributes on 2440declarations only, but not on nested declarators. 2441 2442@xref{Function Attributes}, for details of the semantics of attributes 2443applying to functions. @xref{Variable Attributes}, for details of the 2444semantics of attributes applying to variables. @xref{Type Attributes}, 2445for details of the semantics of attributes applying to structure, union 2446and enumerated types. 2447 2448An @dfn{attribute specifier} is of the form 2449@code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list} 2450is a possibly empty comma-separated sequence of @dfn{attributes}, where 2451each attribute is one of the following: 2452 2453@itemize @bullet 2454@item 2455Empty. Empty attributes are ignored. 2456 2457@item 2458A word (which may be an identifier such as @code{unused}, or a reserved 2459word such as @code{const}). 2460 2461@item 2462A word, followed by, in parentheses, parameters for the attribute. 2463These parameters take one of the following forms: 2464 2465@itemize @bullet 2466@item 2467An identifier. For example, @code{mode} attributes use this form. 2468 2469@item 2470An identifier followed by a comma and a non-empty comma-separated list 2471of expressions. For example, @code{format} attributes use this form. 2472 2473@item 2474A possibly empty comma-separated list of expressions. For example, 2475@code{format_arg} attributes use this form with the list being a single 2476integer constant expression, and @code{alias} attributes use this form 2477with the list being a single string constant. 2478@end itemize 2479@end itemize 2480 2481An @dfn{attribute specifier list} is a sequence of one or more attribute 2482specifiers, not separated by any other tokens. 2483 2484An attribute specifier list may appear after the colon following a 2485label, other than a @code{case} or @code{default} label. The only 2486attribute it makes sense to use after a label is @code{unused}. This 2487feature is intended for code generated by programs which contains labels 2488that may be unused but which is compiled with @option{-Wall}. It would 2489not normally be appropriate to use in it human-written code, though it 2490could be useful in cases where the code that jumps to the label is 2491contained within an @code{#ifdef} conditional. 2492 2493An attribute specifier list may appear as part of a @code{struct}, 2494@code{union} or @code{enum} specifier. It may go either immediately 2495after the @code{struct}, @code{union} or @code{enum} keyword, or after 2496the closing brace. It is ignored if the content of the structure, union 2497or enumerated type is not defined in the specifier in which the 2498attribute specifier list is used---that is, in usages such as 2499@code{struct __attribute__((foo)) bar} with no following opening brace. 2500Where attribute specifiers follow the closing brace, they are considered 2501to relate to the structure, union or enumerated type defined, not to any 2502enclosing declaration the type specifier appears in, and the type 2503defined is not complete until after the attribute specifiers. 2504@c Otherwise, there would be the following problems: a shift/reduce 2505@c conflict between attributes binding the struct/union/enum and 2506@c binding to the list of specifiers/qualifiers; and "aligned" 2507@c attributes could use sizeof for the structure, but the size could be 2508@c changed later by "packed" attributes. 2509 2510Otherwise, an attribute specifier appears as part of a declaration, 2511counting declarations of unnamed parameters and type names, and relates 2512to that declaration (which may be nested in another declaration, for 2513example in the case of a parameter declaration), or to a particular declarator 2514within a declaration. Where an 2515attribute specifier is applied to a parameter declared as a function or 2516an array, it should apply to the function or array rather than the 2517pointer to which the parameter is implicitly converted, but this is not 2518yet correctly implemented. 2519 2520Any list of specifiers and qualifiers at the start of a declaration may 2521contain attribute specifiers, whether or not such a list may in that 2522context contain storage class specifiers. (Some attributes, however, 2523are essentially in the nature of storage class specifiers, and only make 2524sense where storage class specifiers may be used; for example, 2525@code{section}.) There is one necessary limitation to this syntax: the 2526first old-style parameter declaration in a function definition cannot 2527begin with an attribute specifier, because such an attribute applies to 2528the function instead by syntax described below (which, however, is not 2529yet implemented in this case). In some other cases, attribute 2530specifiers are permitted by this grammar but not yet supported by the 2531compiler. All attribute specifiers in this place relate to the 2532declaration as a whole. In the obsolescent usage where a type of 2533@code{int} is implied by the absence of type specifiers, such a list of 2534specifiers and qualifiers may be an attribute specifier list with no 2535other specifiers or qualifiers. 2536 2537An attribute specifier list may appear immediately before a declarator 2538(other than the first) in a comma-separated list of declarators in a 2539declaration of more than one identifier using a single list of 2540specifiers and qualifiers. Such attribute specifiers apply 2541only to the identifier before whose declarator they appear. For 2542example, in 2543 2544@smallexample 2545__attribute__((noreturn)) void d0 (void), 2546 __attribute__((format(printf, 1, 2))) d1 (const char *, ...), 2547 d2 (void) 2548@end smallexample 2549 2550@noindent 2551the @code{noreturn} attribute applies to all the functions 2552declared; the @code{format} attribute only applies to @code{d1}. 2553 2554An attribute specifier list may appear immediately before the comma, 2555@code{=} or semicolon terminating the declaration of an identifier other 2556than a function definition. At present, such attribute specifiers apply 2557to the declared object or function, but in future they may attach to the 2558outermost adjacent declarator. In simple cases there is no difference, 2559but, for example, in 2560 2561@smallexample 2562void (****f)(void) __attribute__((noreturn)); 2563@end smallexample 2564 2565@noindent 2566at present the @code{noreturn} attribute applies to @code{f}, which 2567causes a warning since @code{f} is not a function, but in future it may 2568apply to the function @code{****f}. The precise semantics of what 2569attributes in such cases will apply to are not yet specified. Where an 2570assembler name for an object or function is specified (@pxref{Asm 2571Labels}), at present the attribute must follow the @code{asm} 2572specification; in future, attributes before the @code{asm} specification 2573may apply to the adjacent declarator, and those after it to the declared 2574object or function. 2575 2576An attribute specifier list may, in future, be permitted to appear after 2577the declarator in a function definition (before any old-style parameter 2578declarations or the function body). 2579 2580Attribute specifiers may be mixed with type qualifiers appearing inside 2581the @code{[]} of a parameter array declarator, in the C99 construct by 2582which such qualifiers are applied to the pointer to which the array is 2583implicitly converted. Such attribute specifiers apply to the pointer, 2584not to the array, but at present this is not implemented and they are 2585ignored. 2586 2587An attribute specifier list may appear at the start of a nested 2588declarator. At present, there are some limitations in this usage: the 2589attributes correctly apply to the declarator, but for most individual 2590attributes the semantics this implies are not implemented. 2591When attribute specifiers follow the @code{*} of a pointer 2592declarator, they may be mixed with any type qualifiers present. 2593The following describes the formal semantics of this syntax. It will make the 2594most sense if you are familiar with the formal specification of 2595declarators in the ISO C standard. 2596 2597Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T 2598D1}, where @code{T} contains declaration specifiers that specify a type 2599@var{Type} (such as @code{int}) and @code{D1} is a declarator that 2600contains an identifier @var{ident}. The type specified for @var{ident} 2601for derived declarators whose type does not include an attribute 2602specifier is as in the ISO C standard. 2603 2604If @code{D1} has the form @code{( @var{attribute-specifier-list} D )}, 2605and the declaration @code{T D} specifies the type 2606``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 2607@code{T D1} specifies the type ``@var{derived-declarator-type-list} 2608@var{attribute-specifier-list} @var{Type}'' for @var{ident}. 2609 2610If @code{D1} has the form @code{* 2611@var{type-qualifier-and-attribute-specifier-list} D}, and the 2612declaration @code{T D} specifies the type 2613``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 2614@code{T D1} specifies the type ``@var{derived-declarator-type-list} 2615@var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for 2616@var{ident}. 2617 2618For example, 2619 2620@smallexample 2621void (__attribute__((noreturn)) ****f) (void); 2622@end smallexample 2623 2624@noindent 2625specifies the type ``pointer to pointer to pointer to pointer to 2626non-returning function returning @code{void}''. As another example, 2627 2628@smallexample 2629char *__attribute__((aligned(8))) *f; 2630@end smallexample 2631 2632@noindent 2633specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''. 2634Note again that this does not work with most attributes; for example, 2635the usage of @samp{aligned} and @samp{noreturn} attributes given above 2636is not yet supported. 2637 2638For compatibility with existing code written for compiler versions that 2639did not implement attributes on nested declarators, some laxity is 2640allowed in the placing of attributes. If an attribute that only applies 2641to types is applied to a declaration, it will be treated as applying to 2642the type of that declaration. If an attribute that only applies to 2643declarations is applied to the type of a declaration, it will be treated 2644as applying to that declaration; and, for compatibility with code 2645placing the attributes immediately before the identifier declared, such 2646an attribute applied to a function return type will be treated as 2647applying to the function type, and such an attribute applied to an array 2648element type will be treated as applying to the array type. If an 2649attribute that only applies to function types is applied to a 2650pointer-to-function type, it will be treated as applying to the pointer 2651target type; if such an attribute is applied to a function return type 2652that is not a pointer-to-function type, it will be treated as applying 2653to the function type. 2654 2655@node Function Prototypes 2656@section Prototypes and Old-Style Function Definitions 2657@cindex function prototype declarations 2658@cindex old-style function definitions 2659@cindex promotion of formal parameters 2660 2661GNU C extends ISO C to allow a function prototype to override a later 2662old-style non-prototype definition. Consider the following example: 2663 2664@example 2665/* @r{Use prototypes unless the compiler is old-fashioned.} */ 2666#ifdef __STDC__ 2667#define P(x) x 2668#else 2669#define P(x) () 2670#endif 2671 2672/* @r{Prototype function declaration.} */ 2673int isroot P((uid_t)); 2674 2675/* @r{Old-style function definition.} */ 2676int 2677isroot (x) /* ??? lossage here ??? */ 2678 uid_t x; 2679@{ 2680 return x == 0; 2681@} 2682@end example 2683 2684Suppose the type @code{uid_t} happens to be @code{short}. ISO C does 2685not allow this example, because subword arguments in old-style 2686non-prototype definitions are promoted. Therefore in this example the 2687function definition's argument is really an @code{int}, which does not 2688match the prototype argument type of @code{short}. 2689 2690This restriction of ISO C makes it hard to write code that is portable 2691to traditional C compilers, because the programmer does not know 2692whether the @code{uid_t} type is @code{short}, @code{int}, or 2693@code{long}. Therefore, in cases like these GNU C allows a prototype 2694to override a later old-style definition. More precisely, in GNU C, a 2695function prototype argument type overrides the argument type specified 2696by a later old-style definition if the former type is the same as the 2697latter type before promotion. Thus in GNU C the above example is 2698equivalent to the following: 2699 2700@example 2701int isroot (uid_t); 2702 2703int 2704isroot (uid_t x) 2705@{ 2706 return x == 0; 2707@} 2708@end example 2709 2710@noindent 2711GNU C++ does not support old-style function definitions, so this 2712extension is irrelevant. 2713 2714@node C++ Comments 2715@section C++ Style Comments 2716@cindex // 2717@cindex C++ comments 2718@cindex comments, C++ style 2719 2720In GNU C, you may use C++ style comments, which start with @samp{//} and 2721continue until the end of the line. Many other C implementations allow 2722such comments, and they are likely to be in a future C standard. 2723However, C++ style comments are not recognized if you specify 2724@w{@option{-ansi}}, a @option{-std} option specifying a version of ISO C 2725before C99, or @w{@option{-traditional}}, since they are incompatible 2726with traditional constructs like @code{dividend//*comment*/divisor}. 2727 2728@node Dollar Signs 2729@section Dollar Signs in Identifier Names 2730@cindex $ 2731@cindex dollar signs in identifier names 2732@cindex identifier names, dollar signs in 2733 2734In GNU C, you may normally use dollar signs in identifier names. 2735This is because many traditional C implementations allow such identifiers. 2736However, dollar signs in identifiers are not supported on a few target 2737machines, typically because the target assembler does not allow them. 2738 2739@node Character Escapes 2740@section The Character @key{ESC} in Constants 2741 2742You can use the sequence @samp{\e} in a string or character constant to 2743stand for the ASCII character @key{ESC}. 2744 2745@node Alignment 2746@section Inquiring on Alignment of Types or Variables 2747@cindex alignment 2748@cindex type alignment 2749@cindex variable alignment 2750 2751The keyword @code{__alignof__} allows you to inquire about how an object 2752is aligned, or the minimum alignment usually required by a type. Its 2753syntax is just like @code{sizeof}. 2754 2755For example, if the target machine requires a @code{double} value to be 2756aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8. 2757This is true on many RISC machines. On more traditional machine 2758designs, @code{__alignof__ (double)} is 4 or even 2. 2759 2760Some machines never actually require alignment; they allow reference to any 2761data type even at an odd addresses. For these machines, @code{__alignof__} 2762reports the @emph{recommended} alignment of a type. 2763 2764If the operand of @code{__alignof__} is an lvalue rather than a type, 2765its value is the required alignment for its type, taking into account 2766any minimum alignment specified with GCC's @code{__attribute__} 2767extension (@pxref{Variable Attributes}). For example, after this 2768declaration: 2769 2770@example 2771struct foo @{ int x; char y; @} foo1; 2772@end example 2773 2774@noindent 2775the value of @code{__alignof__ (foo1.y)} is 1, even though its actual 2776alignment is probably 2 or 4, the same as @code{__alignof__ (int)}. 2777 2778It is an error to ask for the alignment of an incomplete type. 2779 2780@node Variable Attributes 2781@section Specifying Attributes of Variables 2782@cindex attribute of variables 2783@cindex variable attributes 2784 2785The keyword @code{__attribute__} allows you to specify special 2786attributes of variables or structure fields. This keyword is followed 2787by an attribute specification inside double parentheses. Ten 2788attributes are currently defined for variables: @code{aligned}, 2789@code{mode}, @code{nocommon}, @code{packed}, @code{section}, 2790@code{transparent_union}, @code{unused}, @code{deprecated}, 2791@code{vector_size}, and @code{weak}. Some other attributes are defined 2792for variables on particular target systems. Other attributes are 2793available for functions (@pxref{Function Attributes}) and for types 2794(@pxref{Type Attributes}). Other front ends might define more 2795attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}). 2796 2797You may also specify attributes with @samp{__} preceding and following 2798each keyword. This allows you to use them in header files without 2799being concerned about a possible macro of the same name. For example, 2800you may use @code{__aligned__} instead of @code{aligned}. 2801 2802@xref{Attribute Syntax}, for details of the exact syntax for using 2803attributes. 2804 2805@table @code 2806@cindex @code{aligned} attribute 2807@item aligned (@var{alignment}) 2808This attribute specifies a minimum alignment for the variable or 2809structure field, measured in bytes. For example, the declaration: 2810 2811@smallexample 2812int x __attribute__ ((aligned (16))) = 0; 2813@end smallexample 2814 2815@noindent 2816causes the compiler to allocate the global variable @code{x} on a 281716-byte boundary. On a 68040, this could be used in conjunction with 2818an @code{asm} expression to access the @code{move16} instruction which 2819requires 16-byte aligned operands. 2820 2821You can also specify the alignment of structure fields. For example, to 2822create a double-word aligned @code{int} pair, you could write: 2823 2824@smallexample 2825struct foo @{ int x[2] __attribute__ ((aligned (8))); @}; 2826@end smallexample 2827 2828@noindent 2829This is an alternative to creating a union with a @code{double} member 2830that forces the union to be double-word aligned. 2831 2832As in the preceding examples, you can explicitly specify the alignment 2833(in bytes) that you wish the compiler to use for a given variable or 2834structure field. Alternatively, you can leave out the alignment factor 2835and just ask the compiler to align a variable or field to the maximum 2836useful alignment for the target machine you are compiling for. For 2837example, you could write: 2838 2839@smallexample 2840short array[3] __attribute__ ((aligned)); 2841@end smallexample 2842 2843Whenever you leave out the alignment factor in an @code{aligned} attribute 2844specification, the compiler automatically sets the alignment for the declared 2845variable or field to the largest alignment which is ever used for any data 2846type on the target machine you are compiling for. Doing this can often make 2847copy operations more efficient, because the compiler can use whatever 2848instructions copy the biggest chunks of memory when performing copies to 2849or from the variables or fields that you have aligned this way. 2850 2851The @code{aligned} attribute can only increase the alignment; but you 2852can decrease it by specifying @code{packed} as well. See below. 2853 2854Note that the effectiveness of @code{aligned} attributes may be limited 2855by inherent limitations in your linker. On many systems, the linker is 2856only able to arrange for variables to be aligned up to a certain maximum 2857alignment. (For some linkers, the maximum supported alignment may 2858be very very small.) If your linker is only able to align variables 2859up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} 2860in an @code{__attribute__} will still only provide you with 8 byte 2861alignment. See your linker documentation for further information. 2862 2863@item mode (@var{mode}) 2864@cindex @code{mode} attribute 2865This attribute specifies the data type for the declaration---whichever 2866type corresponds to the mode @var{mode}. This in effect lets you 2867request an integer or floating point type according to its width. 2868 2869You may also specify a mode of @samp{byte} or @samp{__byte__} to 2870indicate the mode corresponding to a one-byte integer, @samp{word} or 2871@samp{__word__} for the mode of a one-word integer, and @samp{pointer} 2872or @samp{__pointer__} for the mode used to represent pointers. 2873 2874@item nocommon 2875@cindex @code{nocommon} attribute 2876@opindex fno-common 2877This attribute specifies requests GCC not to place a variable 2878``common'' but instead to allocate space for it directly. If you 2879specify the @option{-fno-common} flag, GCC will do this for all 2880variables. 2881 2882Specifying the @code{nocommon} attribute for a variable provides an 2883initialization of zeros. A variable may only be initialized in one 2884source file. 2885 2886@item packed 2887@cindex @code{packed} attribute 2888The @code{packed} attribute specifies that a variable or structure field 2889should have the smallest possible alignment---one byte for a variable, 2890and one bit for a field, unless you specify a larger value with the 2891@code{aligned} attribute. 2892 2893Here is a structure in which the field @code{x} is packed, so that it 2894immediately follows @code{a}: 2895 2896@example 2897struct foo 2898@{ 2899 char a; 2900 int x[2] __attribute__ ((packed)); 2901@}; 2902@end example 2903 2904@item section ("@var{section-name}") 2905@cindex @code{section} variable attribute 2906Normally, the compiler places the objects it generates in sections like 2907@code{data} and @code{bss}. Sometimes, however, you need additional sections, 2908or you need certain particular variables to appear in special sections, 2909for example to map to special hardware. The @code{section} 2910attribute specifies that a variable (or function) lives in a particular 2911section. For example, this small program uses several specific section names: 2912 2913@smallexample 2914struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @}; 2915struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @}; 2916char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @}; 2917int init_data __attribute__ ((section ("INITDATA"))) = 0; 2918 2919main() 2920@{ 2921 /* Initialize stack pointer */ 2922 init_sp (stack + sizeof (stack)); 2923 2924 /* Initialize initialized data */ 2925 memcpy (&init_data, &data, &edata - &data); 2926 2927 /* Turn on the serial ports */ 2928 init_duart (&a); 2929 init_duart (&b); 2930@} 2931@end smallexample 2932 2933@noindent 2934Use the @code{section} attribute with an @emph{initialized} definition 2935of a @emph{global} variable, as shown in the example. GCC issues 2936a warning and otherwise ignores the @code{section} attribute in 2937uninitialized variable declarations. 2938 2939You may only use the @code{section} attribute with a fully initialized 2940global definition because of the way linkers work. The linker requires 2941each object be defined once, with the exception that uninitialized 2942variables tentatively go in the @code{common} (or @code{bss}) section 2943and can be multiply ``defined''. You can force a variable to be 2944initialized with the @option{-fno-common} flag or the @code{nocommon} 2945attribute. 2946 2947Some file formats do not support arbitrary sections so the @code{section} 2948attribute is not available on all platforms. 2949If you need to map the entire contents of a module to a particular 2950section, consider using the facilities of the linker instead. 2951 2952@item shared 2953@cindex @code{shared} variable attribute 2954On Windows NT, in addition to putting variable definitions in a named 2955section, the section can also be shared among all running copies of an 2956executable or DLL@. For example, this small program defines shared data 2957by putting it in a named section @code{shared} and marking the section 2958shareable: 2959 2960@smallexample 2961int foo __attribute__((section ("shared"), shared)) = 0; 2962 2963int 2964main() 2965@{ 2966 /* Read and write foo. All running 2967 copies see the same value. */ 2968 return 0; 2969@} 2970@end smallexample 2971 2972@noindent 2973You may only use the @code{shared} attribute along with @code{section} 2974attribute with a fully initialized global definition because of the way 2975linkers work. See @code{section} attribute for more information. 2976 2977The @code{shared} attribute is only available on Windows NT@. 2978 2979@item transparent_union 2980This attribute, attached to a function parameter which is a union, means 2981that the corresponding argument may have the type of any union member, 2982but the argument is passed as if its type were that of the first union 2983member. For more details see @xref{Type Attributes}. You can also use 2984this attribute on a @code{typedef} for a union data type; then it 2985applies to all function parameters with that type. 2986 2987@item unused 2988This attribute, attached to a variable, means that the variable is meant 2989to be possibly unused. GCC will not produce a warning for this 2990variable. 2991 2992@item deprecated 2993The @code{deprecated} attribute results in a warning if the variable 2994is used anywhere in the source file. This is useful when identifying 2995variables that are expected to be removed in a future version of a 2996program. The warning also includes the location of the declaration 2997of the deprecated variable, to enable users to easily find further 2998information about why the variable is deprecated, or what they should 2999do instead. Note that the warnings only occurs for uses: 3000 3001@smallexample 3002extern int old_var __attribute__ ((deprecated)); 3003extern int old_var; 3004int new_fn () @{ return old_var; @} 3005@end smallexample 3006 3007results in a warning on line 3 but not line 2. 3008 3009The @code{deprecated} attribute can also be used for functions and 3010types (@pxref{Function Attributes}, @pxref{Type Attributes}.) 3011 3012@item vector_size (@var{bytes}) 3013This attribute specifies the vector size for the variable, measured in 3014bytes. For example, the declaration: 3015 3016@smallexample 3017int foo __attribute__ ((vector_size (16))); 3018@end smallexample 3019 3020@noindent 3021causes the compiler to set the mode for @code{foo}, to be 16 bytes, 3022divided into @code{int} sized units. Assuming a 32-bit int (a vector of 30234 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@. 3024 3025This attribute is only applicable to integral and float scalars, 3026although arrays, pointers, and function return values are allowed in 3027conjunction with this construct. 3028 3029Aggregates with this attribute are invalid, even if they are of the same 3030size as a corresponding scalar. For example, the declaration: 3031 3032@smallexample 3033struct S @{ int a; @}; 3034struct S __attribute__ ((vector_size (16))) foo; 3035@end smallexample 3036 3037@noindent 3038is invalid even if the size of the structure is the same as the size of 3039the @code{int}. 3040 3041@item weak 3042The @code{weak} attribute is described in @xref{Function Attributes}. 3043 3044@item model (@var{model-name}) 3045@cindex variable addressability on the M32R/D 3046Use this attribute on the M32R/D to set the addressability of an object. 3047The identifier @var{model-name} is one of @code{small}, @code{medium}, 3048or @code{large}, representing each of the code models. 3049 3050Small model objects live in the lower 16MB of memory (so that their 3051addresses can be loaded with the @code{ld24} instruction). 3052 3053Medium and large model objects may live anywhere in the 32-bit address space 3054(the compiler will generate @code{seth/add3} instructions to load their 3055addresses). 3056 3057@end table 3058 3059To specify multiple attributes, separate them by commas within the 3060double parentheses: for example, @samp{__attribute__ ((aligned (16), 3061packed))}. 3062 3063@node Type Attributes 3064@section Specifying Attributes of Types 3065@cindex attribute of types 3066@cindex type attributes 3067 3068The keyword @code{__attribute__} allows you to specify special 3069attributes of @code{struct} and @code{union} types when you define such 3070types. This keyword is followed by an attribute specification inside 3071double parentheses. Five attributes are currently defined for types: 3072@code{aligned}, @code{packed}, @code{transparent_union}, @code{unused}, 3073and @code{deprecated}. Other attributes are defined for functions 3074(@pxref{Function Attributes}) and for variables (@pxref{Variable Attributes}). 3075 3076You may also specify any one of these attributes with @samp{__} 3077preceding and following its keyword. This allows you to use these 3078attributes in header files without being concerned about a possible 3079macro of the same name. For example, you may use @code{__aligned__} 3080instead of @code{aligned}. 3081 3082You may specify the @code{aligned} and @code{transparent_union} 3083attributes either in a @code{typedef} declaration or just past the 3084closing curly brace of a complete enum, struct or union type 3085@emph{definition} and the @code{packed} attribute only past the closing 3086brace of a definition. 3087 3088You may also specify attributes between the enum, struct or union 3089tag and the name of the type rather than after the closing brace. 3090 3091@xref{Attribute Syntax}, for details of the exact syntax for using 3092attributes. 3093 3094@table @code 3095@cindex @code{aligned} attribute 3096@item aligned (@var{alignment}) 3097This attribute specifies a minimum alignment (in bytes) for variables 3098of the specified type. For example, the declarations: 3099 3100@smallexample 3101struct S @{ short f[3]; @} __attribute__ ((aligned (8))); 3102typedef int more_aligned_int __attribute__ ((aligned (8))); 3103@end smallexample 3104 3105@noindent 3106force the compiler to insure (as far as it can) that each variable whose 3107type is @code{struct S} or @code{more_aligned_int} will be allocated and 3108aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all 3109variables of type @code{struct S} aligned to 8-byte boundaries allows 3110the compiler to use the @code{ldd} and @code{std} (doubleword load and 3111store) instructions when copying one variable of type @code{struct S} to 3112another, thus improving run-time efficiency. 3113 3114Note that the alignment of any given @code{struct} or @code{union} type 3115is required by the ISO C standard to be at least a perfect multiple of 3116the lowest common multiple of the alignments of all of the members of 3117the @code{struct} or @code{union} in question. This means that you @emph{can} 3118effectively adjust the alignment of a @code{struct} or @code{union} 3119type by attaching an @code{aligned} attribute to any one of the members 3120of such a type, but the notation illustrated in the example above is a 3121more obvious, intuitive, and readable way to request the compiler to 3122adjust the alignment of an entire @code{struct} or @code{union} type. 3123 3124As in the preceding example, you can explicitly specify the alignment 3125(in bytes) that you wish the compiler to use for a given @code{struct} 3126or @code{union} type. Alternatively, you can leave out the alignment factor 3127and just ask the compiler to align a type to the maximum 3128useful alignment for the target machine you are compiling for. For 3129example, you could write: 3130 3131@smallexample 3132struct S @{ short f[3]; @} __attribute__ ((aligned)); 3133@end smallexample 3134 3135Whenever you leave out the alignment factor in an @code{aligned} 3136attribute specification, the compiler automatically sets the alignment 3137for the type to the largest alignment which is ever used for any data 3138type on the target machine you are compiling for. Doing this can often 3139make copy operations more efficient, because the compiler can use 3140whatever instructions copy the biggest chunks of memory when performing 3141copies to or from the variables which have types that you have aligned 3142this way. 3143 3144In the example above, if the size of each @code{short} is 2 bytes, then 3145the size of the entire @code{struct S} type is 6 bytes. The smallest 3146power of two which is greater than or equal to that is 8, so the 3147compiler sets the alignment for the entire @code{struct S} type to 8 3148bytes. 3149 3150Note that although you can ask the compiler to select a time-efficient 3151alignment for a given type and then declare only individual stand-alone 3152objects of that type, the compiler's ability to select a time-efficient 3153alignment is primarily useful only when you plan to create arrays of 3154variables having the relevant (efficiently aligned) type. If you 3155declare or use arrays of variables of an efficiently-aligned type, then 3156it is likely that your program will also be doing pointer arithmetic (or 3157subscripting, which amounts to the same thing) on pointers to the 3158relevant type, and the code that the compiler generates for these 3159pointer arithmetic operations will often be more efficient for 3160efficiently-aligned types than for other types. 3161 3162The @code{aligned} attribute can only increase the alignment; but you 3163can decrease it by specifying @code{packed} as well. See below. 3164 3165Note that the effectiveness of @code{aligned} attributes may be limited 3166by inherent limitations in your linker. On many systems, the linker is 3167only able to arrange for variables to be aligned up to a certain maximum 3168alignment. (For some linkers, the maximum supported alignment may 3169be very very small.) If your linker is only able to align variables 3170up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} 3171in an @code{__attribute__} will still only provide you with 8 byte 3172alignment. See your linker documentation for further information. 3173 3174@item packed 3175This attribute, attached to an @code{enum}, @code{struct}, or 3176@code{union} type definition, specified that the minimum required memory 3177be used to represent the type. 3178 3179@opindex fshort-enums 3180Specifying this attribute for @code{struct} and @code{union} types is 3181equivalent to specifying the @code{packed} attribute on each of the 3182structure or union members. Specifying the @option{-fshort-enums} 3183flag on the line is equivalent to specifying the @code{packed} 3184attribute on all @code{enum} definitions. 3185 3186You may only specify this attribute after a closing curly brace on an 3187@code{enum} definition, not in a @code{typedef} declaration, unless that 3188declaration also contains the definition of the @code{enum}. 3189 3190@item transparent_union 3191This attribute, attached to a @code{union} type definition, indicates 3192that any function parameter having that union type causes calls to that 3193function to be treated in a special way. 3194 3195First, the argument corresponding to a transparent union type can be of 3196any type in the union; no cast is required. Also, if the union contains 3197a pointer type, the corresponding argument can be a null pointer 3198constant or a void pointer expression; and if the union contains a void 3199pointer type, the corresponding argument can be any pointer expression. 3200If the union member type is a pointer, qualifiers like @code{const} on 3201the referenced type must be respected, just as with normal pointer 3202conversions. 3203 3204Second, the argument is passed to the function using the calling 3205conventions of first member of the transparent union, not the calling 3206conventions of the union itself. All members of the union must have the 3207same machine representation; this is necessary for this argument passing 3208to work properly. 3209 3210Transparent unions are designed for library functions that have multiple 3211interfaces for compatibility reasons. For example, suppose the 3212@code{wait} function must accept either a value of type @code{int *} to 3213comply with Posix, or a value of type @code{union wait *} to comply with 3214the 4.1BSD interface. If @code{wait}'s parameter were @code{void *}, 3215@code{wait} would accept both kinds of arguments, but it would also 3216accept any other pointer type and this would make argument type checking 3217less useful. Instead, @code{<sys/wait.h>} might define the interface 3218as follows: 3219 3220@smallexample 3221typedef union 3222 @{ 3223 int *__ip; 3224 union wait *__up; 3225 @} wait_status_ptr_t __attribute__ ((__transparent_union__)); 3226 3227pid_t wait (wait_status_ptr_t); 3228@end smallexample 3229 3230This interface allows either @code{int *} or @code{union wait *} 3231arguments to be passed, using the @code{int *} calling convention. 3232The program can call @code{wait} with arguments of either type: 3233 3234@example 3235int w1 () @{ int w; return wait (&w); @} 3236int w2 () @{ union wait w; return wait (&w); @} 3237@end example 3238 3239With this interface, @code{wait}'s implementation might look like this: 3240 3241@example 3242pid_t wait (wait_status_ptr_t p) 3243@{ 3244 return waitpid (-1, p.__ip, 0); 3245@} 3246@end example 3247 3248@item unused 3249When attached to a type (including a @code{union} or a @code{struct}), 3250this attribute means that variables of that type are meant to appear 3251possibly unused. GCC will not produce a warning for any variables of 3252that type, even if the variable appears to do nothing. This is often 3253the case with lock or thread classes, which are usually defined and then 3254not referenced, but contain constructors and destructors that have 3255nontrivial bookkeeping functions. 3256 3257@item deprecated 3258The @code{deprecated} attribute results in a warning if the type 3259is used anywhere in the source file. This is useful when identifying 3260types that are expected to be removed in a future version of a program. 3261If possible, the warning also includes the location of the declaration 3262of the deprecated type, to enable users to easily find further 3263information about why the type is deprecated, or what they should do 3264instead. Note that the warnings only occur for uses and then only 3265if the type is being applied to an identifier that itself is not being 3266declared as deprecated. 3267 3268@smallexample 3269typedef int T1 __attribute__ ((deprecated)); 3270T1 x; 3271typedef T1 T2; 3272T2 y; 3273typedef T1 T3 __attribute__ ((deprecated)); 3274T3 z __attribute__ ((deprecated)); 3275@end smallexample 3276 3277results in a warning on line 2 and 3 but not lines 4, 5, or 6. No 3278warning is issued for line 4 because T2 is not explicitly 3279deprecated. Line 5 has no warning because T3 is explicitly 3280deprecated. Similarly for line 6. 3281 3282The @code{deprecated} attribute can also be used for functions and 3283variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.) 3284 3285@end table 3286 3287To specify multiple attributes, separate them by commas within the 3288double parentheses: for example, @samp{__attribute__ ((aligned (16), 3289packed))}. 3290 3291@node Inline 3292@section An Inline Function is As Fast As a Macro 3293@cindex inline functions 3294@cindex integrating function code 3295@cindex open coding 3296@cindex macros, inline alternative 3297 3298By declaring a function @code{inline}, you can direct GCC to 3299integrate that function's code into the code for its callers. This 3300makes execution faster by eliminating the function-call overhead; in 3301addition, if any of the actual argument values are constant, their known 3302values may permit simplifications at compile time so that not all of the 3303inline function's code needs to be included. The effect on code size is 3304less predictable; object code may be larger or smaller with function 3305inlining, depending on the particular case. Inlining of functions is an 3306optimization and it really ``works'' only in optimizing compilation. If 3307you don't use @option{-O}, no function is really inline. 3308 3309Inline functions are included in the ISO C99 standard, but there are 3310currently substantial differences between what GCC implements and what 3311the ISO C99 standard requires. 3312 3313To declare a function inline, use the @code{inline} keyword in its 3314declaration, like this: 3315 3316@example 3317inline int 3318inc (int *a) 3319@{ 3320 (*a)++; 3321@} 3322@end example 3323 3324(If you are writing a header file to be included in ISO C programs, write 3325@code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.) 3326You can also make all ``simple enough'' functions inline with the option 3327@option{-finline-functions}. 3328 3329@opindex Winline 3330Note that certain usages in a function definition can make it unsuitable 3331for inline substitution. Among these usages are: use of varargs, use of 3332alloca, use of variable sized data types (@pxref{Variable Length}), 3333use of computed goto (@pxref{Labels as Values}), use of nonlocal goto, 3334and nested functions (@pxref{Nested Functions}). Using @option{-Winline} 3335will warn when a function marked @code{inline} could not be substituted, 3336and will give the reason for the failure. 3337 3338Note that in C and Objective-C, unlike C++, the @code{inline} keyword 3339does not affect the linkage of the function. 3340 3341@cindex automatic @code{inline} for C++ member fns 3342@cindex @code{inline} automatic for C++ member fns 3343@cindex member fns, automatically @code{inline} 3344@cindex C++ member fns, automatically @code{inline} 3345@opindex fno-default-inline 3346GCC automatically inlines member functions defined within the class 3347body of C++ programs even if they are not explicitly declared 3348@code{inline}. (You can override this with @option{-fno-default-inline}; 3349@pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.) 3350 3351@cindex inline functions, omission of 3352@opindex fkeep-inline-functions 3353When a function is both inline and @code{static}, if all calls to the 3354function are integrated into the caller, and the function's address is 3355never used, then the function's own assembler code is never referenced. 3356In this case, GCC does not actually output assembler code for the 3357function, unless you specify the option @option{-fkeep-inline-functions}. 3358Some calls cannot be integrated for various reasons (in particular, 3359calls that precede the function's definition cannot be integrated, and 3360neither can recursive calls within the definition). If there is a 3361nonintegrated call, then the function is compiled to assembler code as 3362usual. The function must also be compiled as usual if the program 3363refers to its address, because that can't be inlined. 3364 3365@cindex non-static inline function 3366When an inline function is not @code{static}, then the compiler must assume 3367that there may be calls from other source files; since a global symbol can 3368be defined only once in any program, the function must not be defined in 3369the other source files, so the calls therein cannot be integrated. 3370Therefore, a non-@code{static} inline function is always compiled on its 3371own in the usual fashion. 3372 3373If you specify both @code{inline} and @code{extern} in the function 3374definition, then the definition is used only for inlining. In no case 3375is the function compiled on its own, not even if you refer to its 3376address explicitly. Such an address becomes an external reference, as 3377if you had only declared the function, and had not defined it. 3378 3379This combination of @code{inline} and @code{extern} has almost the 3380effect of a macro. The way to use it is to put a function definition in 3381a header file with these keywords, and put another copy of the 3382definition (lacking @code{inline} and @code{extern}) in a library file. 3383The definition in the header file will cause most calls to the function 3384to be inlined. If any uses of the function remain, they will refer to 3385the single copy in the library. 3386 3387For future compatibility with when GCC implements ISO C99 semantics for 3388inline functions, it is best to use @code{static inline} only. (The 3389existing semantics will remain available when @option{-std=gnu89} is 3390specified, but eventually the default will be @option{-std=gnu99} and 3391that will implement the C99 semantics, though it does not do so yet.) 3392 3393GCC does not inline any functions when not optimizing unless you specify 3394the @samp{always_inline} attribute for the function, like this: 3395 3396@example 3397/* Prototype. */ 3398inline void foo (const char) __attribute__((always_inline)); 3399@end example 3400 3401@node Extended Asm 3402@section Assembler Instructions with C Expression Operands 3403@cindex extended @code{asm} 3404@cindex @code{asm} expressions 3405@cindex assembler instructions 3406@cindex registers 3407 3408In an assembler instruction using @code{asm}, you can specify the 3409operands of the instruction using C expressions. This means you need not 3410guess which registers or memory locations will contain the data you want 3411to use. 3412 3413You must specify an assembler instruction template much like what 3414appears in a machine description, plus an operand constraint string for 3415each operand. 3416 3417For example, here is how to use the 68881's @code{fsinx} instruction: 3418 3419@example 3420asm ("fsinx %1,%0" : "=f" (result) : "f" (angle)); 3421@end example 3422 3423@noindent 3424Here @code{angle} is the C expression for the input operand while 3425@code{result} is that of the output operand. Each has @samp{"f"} as its 3426operand constraint, saying that a floating point register is required. 3427The @samp{=} in @samp{=f} indicates that the operand is an output; all 3428output operands' constraints must use @samp{=}. The constraints use the 3429same language used in the machine description (@pxref{Constraints}). 3430 3431Each operand is described by an operand-constraint string followed by 3432the C expression in parentheses. A colon separates the assembler 3433template from the first output operand and another separates the last 3434output operand from the first input, if any. Commas separate the 3435operands within each group. The total number of operands is currently 3436limited to 30; this limitation may be lifted in some future version of 3437GCC. 3438 3439If there are no output operands but there are input operands, you must 3440place two consecutive colons surrounding the place where the output 3441operands would go. 3442 3443As of GCC version 3.1, it is also possible to specify input and output 3444operands using symbolic names which can be referenced within the 3445assembler code. These names are specified inside square brackets 3446preceding the constraint string, and can be referenced inside the 3447assembler code using @code{%[@var{name}]} instead of a percentage sign 3448followed by the operand number. Using named operands the above example 3449could look like: 3450 3451@example 3452asm ("fsinx %[angle],%[output]" 3453 : [output] "=f" (result) 3454 : [angle] "f" (angle)); 3455@end example 3456 3457@noindent 3458Note that the symbolic operand names have no relation whatsoever to 3459other C identifiers. You may use any name you like, even those of 3460existing C symbols, but must ensure that no two operands within the same 3461assembler construct use the same symbolic name. 3462 3463Output operand expressions must be lvalues; the compiler can check this. 3464The input operands need not be lvalues. The compiler cannot check 3465whether the operands have data types that are reasonable for the 3466instruction being executed. It does not parse the assembler instruction 3467template and does not know what it means or even whether it is valid 3468assembler input. The extended @code{asm} feature is most often used for 3469machine instructions the compiler itself does not know exist. If 3470the output expression cannot be directly addressed (for example, it is a 3471bit-field), your constraint must allow a register. In that case, GCC 3472will use the register as the output of the @code{asm}, and then store 3473that register into the output. 3474 3475The ordinary output operands must be write-only; GCC will assume that 3476the values in these operands before the instruction are dead and need 3477not be generated. Extended asm supports input-output or read-write 3478operands. Use the constraint character @samp{+} to indicate such an 3479operand and list it with the output operands. 3480 3481When the constraints for the read-write operand (or the operand in which 3482only some of the bits are to be changed) allows a register, you may, as 3483an alternative, logically split its function into two separate operands, 3484one input operand and one write-only output operand. The connection 3485between them is expressed by constraints which say they need to be in 3486the same location when the instruction executes. You can use the same C 3487expression for both operands, or different expressions. For example, 3488here we write the (fictitious) @samp{combine} instruction with 3489@code{bar} as its read-only source operand and @code{foo} as its 3490read-write destination: 3491 3492@example 3493asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar)); 3494@end example 3495 3496@noindent 3497The constraint @samp{"0"} for operand 1 says that it must occupy the 3498same location as operand 0. A number in constraint is allowed only in 3499an input operand and it must refer to an output operand. 3500 3501Only a number in the constraint can guarantee that one operand will be in 3502the same place as another. The mere fact that @code{foo} is the value 3503of both operands is not enough to guarantee that they will be in the 3504same place in the generated assembler code. The following would not 3505work reliably: 3506 3507@example 3508asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar)); 3509@end example 3510 3511Various optimizations or reloading could cause operands 0 and 1 to be in 3512different registers; GCC knows no reason not to do so. For example, the 3513compiler might find a copy of the value of @code{foo} in one register and 3514use it for operand 1, but generate the output operand 0 in a different 3515register (copying it afterward to @code{foo}'s own address). Of course, 3516since the register for operand 1 is not even mentioned in the assembler 3517code, the result will not work, but GCC can't tell that. 3518 3519As of GCC version 3.1, one may write @code{[@var{name}]} instead of 3520the operand number for a matching constraint. For example: 3521 3522@example 3523asm ("cmoveq %1,%2,%[result]" 3524 : [result] "=r"(result) 3525 : "r" (test), "r"(new), "[result]"(old)); 3526@end example 3527 3528Some instructions clobber specific hard registers. To describe this, 3529write a third colon after the input operands, followed by the names of 3530the clobbered hard registers (given as strings). Here is a realistic 3531example for the VAX: 3532 3533@example 3534asm volatile ("movc3 %0,%1,%2" 3535 : /* no outputs */ 3536 : "g" (from), "g" (to), "g" (count) 3537 : "r0", "r1", "r2", "r3", "r4", "r5"); 3538@end example 3539 3540You may not write a clobber description in a way that overlaps with an 3541input or output operand. For example, you may not have an operand 3542describing a register class with one member if you mention that register 3543in the clobber list. There is no way for you to specify that an input 3544operand is modified without also specifying it as an output 3545operand. Note that if all the output operands you specify are for this 3546purpose (and hence unused), you will then also need to specify 3547@code{volatile} for the @code{asm} construct, as described below, to 3548prevent GCC from deleting the @code{asm} statement as unused. 3549 3550If you refer to a particular hardware register from the assembler code, 3551you will probably have to list the register after the third colon to 3552tell the compiler the register's value is modified. In some assemblers, 3553the register names begin with @samp{%}; to produce one @samp{%} in the 3554assembler code, you must write @samp{%%} in the input. 3555 3556If your assembler instruction can alter the condition code register, add 3557@samp{cc} to the list of clobbered registers. GCC on some machines 3558represents the condition codes as a specific hardware register; 3559@samp{cc} serves to name this register. On other machines, the 3560condition code is handled differently, and specifying @samp{cc} has no 3561effect. But it is valid no matter what the machine. 3562 3563If your assembler instruction modifies memory in an unpredictable 3564fashion, add @samp{memory} to the list of clobbered registers. This 3565will cause GCC to not keep memory values cached in registers across 3566the assembler instruction. You will also want to add the 3567@code{volatile} keyword if the memory affected is not listed in the 3568inputs or outputs of the @code{asm}, as the @samp{memory} clobber does 3569not count as a side-effect of the @code{asm}. 3570 3571You can put multiple assembler instructions together in a single 3572@code{asm} template, separated by the characters normally used in assembly 3573code for the system. A combination that works in most places is a newline 3574to break the line, plus a tab character to move to the instruction field 3575(written as @samp{\n\t}). Sometimes semicolons can be used, if the 3576assembler allows semicolons as a line-breaking character. Note that some 3577assembler dialects use semicolons to start a comment. 3578The input operands are guaranteed not to use any of the clobbered 3579registers, and neither will the output operands' addresses, so you can 3580read and write the clobbered registers as many times as you like. Here 3581is an example of multiple instructions in a template; it assumes the 3582subroutine @code{_foo} accepts arguments in registers 9 and 10: 3583 3584@example 3585asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo" 3586 : /* no outputs */ 3587 : "g" (from), "g" (to) 3588 : "r9", "r10"); 3589@end example 3590 3591Unless an output operand has the @samp{&} constraint modifier, GCC 3592may allocate it in the same register as an unrelated input operand, on 3593the assumption the inputs are consumed before the outputs are produced. 3594This assumption may be false if the assembler code actually consists of 3595more than one instruction. In such a case, use @samp{&} for each output 3596operand that may not overlap an input. @xref{Modifiers}. 3597 3598If you want to test the condition code produced by an assembler 3599instruction, you must include a branch and a label in the @code{asm} 3600construct, as follows: 3601 3602@example 3603asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:" 3604 : "g" (result) 3605 : "g" (input)); 3606@end example 3607 3608@noindent 3609This assumes your assembler supports local labels, as the GNU assembler 3610and most Unix assemblers do. 3611 3612Speaking of labels, jumps from one @code{asm} to another are not 3613supported. The compiler's optimizers do not know about these jumps, and 3614therefore they cannot take account of them when deciding how to 3615optimize. 3616 3617@cindex macros containing @code{asm} 3618Usually the most convenient way to use these @code{asm} instructions is to 3619encapsulate them in macros that look like functions. For example, 3620 3621@example 3622#define sin(x) \ 3623(@{ double __value, __arg = (x); \ 3624 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \ 3625 __value; @}) 3626@end example 3627 3628@noindent 3629Here the variable @code{__arg} is used to make sure that the instruction 3630operates on a proper @code{double} value, and to accept only those 3631arguments @code{x} which can convert automatically to a @code{double}. 3632 3633Another way to make sure the instruction operates on the correct data 3634type is to use a cast in the @code{asm}. This is different from using a 3635variable @code{__arg} in that it converts more different types. For 3636example, if the desired type were @code{int}, casting the argument to 3637@code{int} would accept a pointer with no complaint, while assigning the 3638argument to an @code{int} variable named @code{__arg} would warn about 3639using a pointer unless the caller explicitly casts it. 3640 3641If an @code{asm} has output operands, GCC assumes for optimization 3642purposes the instruction has no side effects except to change the output 3643operands. This does not mean instructions with a side effect cannot be 3644used, but you must be careful, because the compiler may eliminate them 3645if the output operands aren't used, or move them out of loops, or 3646replace two with one if they constitute a common subexpression. Also, 3647if your instruction does have a side effect on a variable that otherwise 3648appears not to change, the old value of the variable may be reused later 3649if it happens to be found in a register. 3650 3651You can prevent an @code{asm} instruction from being deleted, moved 3652significantly, or combined, by writing the keyword @code{volatile} after 3653the @code{asm}. For example: 3654 3655@example 3656#define get_and_set_priority(new) \ 3657(@{ int __old; \ 3658 asm volatile ("get_and_set_priority %0, %1" \ 3659 : "=g" (__old) : "g" (new)); \ 3660 __old; @}) 3661@end example 3662 3663@noindent 3664If you write an @code{asm} instruction with no outputs, GCC will know 3665the instruction has side-effects and will not delete the instruction or 3666move it outside of loops. 3667 3668The @code{volatile} keyword indicates that the instruction has 3669important side-effects. GCC will not delete a volatile @code{asm} if 3670it is reachable. (The instruction can still be deleted if GCC can 3671prove that control-flow will never reach the location of the 3672instruction.) In addition, GCC will not reschedule instructions 3673across a volatile @code{asm} instruction. For example: 3674 3675@example 3676*(volatile int *)addr = foo; 3677asm volatile ("eieio" : : ); 3678@end example 3679 3680@noindent 3681Assume @code{addr} contains the address of a memory mapped device 3682register. The PowerPC @code{eieio} instruction (Enforce In-order 3683Execution of I/O) tells the CPU to make sure that the store to that 3684device register happens before it issues any other I/O@. 3685 3686Note that even a volatile @code{asm} instruction can be moved in ways 3687that appear insignificant to the compiler, such as across jump 3688instructions. You can't expect a sequence of volatile @code{asm} 3689instructions to remain perfectly consecutive. If you want consecutive 3690output, use a single @code{asm}. Also, GCC will perform some 3691optimizations across a volatile @code{asm} instruction; GCC does not 3692``forget everything'' when it encounters a volatile @code{asm} 3693instruction the way some other compilers do. 3694 3695An @code{asm} instruction without any operands or clobbers (an ``old 3696style'' @code{asm}) will be treated identically to a volatile 3697@code{asm} instruction. 3698 3699It is a natural idea to look for a way to give access to the condition 3700code left by the assembler instruction. However, when we attempted to 3701implement this, we found no way to make it work reliably. The problem 3702is that output operands might need reloading, which would result in 3703additional following ``store'' instructions. On most machines, these 3704instructions would alter the condition code before there was time to 3705test it. This problem doesn't arise for ordinary ``test'' and 3706``compare'' instructions because they don't have any output operands. 3707 3708For reasons similar to those described above, it is not possible to give 3709an assembler instruction access to the condition code left by previous 3710instructions. 3711 3712If you are writing a header file that should be includable in ISO C 3713programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate 3714Keywords}. 3715 3716@subsection i386 floating point asm operands 3717 3718There are several rules on the usage of stack-like regs in 3719asm_operands insns. These rules apply only to the operands that are 3720stack-like regs: 3721 3722@enumerate 3723@item 3724Given a set of input regs that die in an asm_operands, it is 3725necessary to know which are implicitly popped by the asm, and 3726which must be explicitly popped by gcc. 3727 3728An input reg that is implicitly popped by the asm must be 3729explicitly clobbered, unless it is constrained to match an 3730output operand. 3731 3732@item 3733For any input reg that is implicitly popped by an asm, it is 3734necessary to know how to adjust the stack to compensate for the pop. 3735If any non-popped input is closer to the top of the reg-stack than 3736the implicitly popped reg, it would not be possible to know what the 3737stack looked like---it's not clear how the rest of the stack ``slides 3738up''. 3739 3740All implicitly popped input regs must be closer to the top of 3741the reg-stack than any input that is not implicitly popped. 3742 3743It is possible that if an input dies in an insn, reload might 3744use the input reg for an output reload. Consider this example: 3745 3746@example 3747asm ("foo" : "=t" (a) : "f" (b)); 3748@end example 3749 3750This asm says that input B is not popped by the asm, and that 3751the asm pushes a result onto the reg-stack, i.e., the stack is one 3752deeper after the asm than it was before. But, it is possible that 3753reload will think that it can use the same reg for both the input and 3754the output, if input B dies in this insn. 3755 3756If any input operand uses the @code{f} constraint, all output reg 3757constraints must use the @code{&} earlyclobber. 3758 3759The asm above would be written as 3760 3761@example 3762asm ("foo" : "=&t" (a) : "f" (b)); 3763@end example 3764 3765@item 3766Some operands need to be in particular places on the stack. All 3767output operands fall in this category---there is no other way to 3768know which regs the outputs appear in unless the user indicates 3769this in the constraints. 3770 3771Output operands must specifically indicate which reg an output 3772appears in after an asm. @code{=f} is not allowed: the operand 3773constraints must select a class with a single reg. 3774 3775@item 3776Output operands may not be ``inserted'' between existing stack regs. 3777Since no 387 opcode uses a read/write operand, all output operands 3778are dead before the asm_operands, and are pushed by the asm_operands. 3779It makes no sense to push anywhere but the top of the reg-stack. 3780 3781Output operands must start at the top of the reg-stack: output 3782operands may not ``skip'' a reg. 3783 3784@item 3785Some asm statements may need extra stack space for internal 3786calculations. This can be guaranteed by clobbering stack registers 3787unrelated to the inputs and outputs. 3788 3789@end enumerate 3790 3791Here are a couple of reasonable asms to want to write. This asm 3792takes one input, which is internally popped, and produces two outputs. 3793 3794@example 3795asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); 3796@end example 3797 3798This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode, 3799and replaces them with one output. The user must code the @code{st(1)} 3800clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs. 3801 3802@example 3803asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); 3804@end example 3805 3806@include md.texi 3807 3808@node Asm Labels 3809@section Controlling Names Used in Assembler Code 3810@cindex assembler names for identifiers 3811@cindex names used in assembler code 3812@cindex identifiers, names in assembler code 3813 3814You can specify the name to be used in the assembler code for a C 3815function or variable by writing the @code{asm} (or @code{__asm__}) 3816keyword after the declarator as follows: 3817 3818@example 3819int foo asm ("myfoo") = 2; 3820@end example 3821 3822@noindent 3823This specifies that the name to be used for the variable @code{foo} in 3824the assembler code should be @samp{myfoo} rather than the usual 3825@samp{_foo}. 3826 3827On systems where an underscore is normally prepended to the name of a C 3828function or variable, this feature allows you to define names for the 3829linker that do not start with an underscore. 3830 3831It does not make sense to use this feature with a non-static local 3832variable since such variables do not have assembler names. If you are 3833trying to put the variable in a particular register, see @ref{Explicit 3834Reg Vars}. GCC presently accepts such code with a warning, but will 3835probably be changed to issue an error, rather than a warning, in the 3836future. 3837 3838You cannot use @code{asm} in this way in a function @emph{definition}; but 3839you can get the same effect by writing a declaration for the function 3840before its definition and putting @code{asm} there, like this: 3841 3842@example 3843extern func () asm ("FUNC"); 3844 3845func (x, y) 3846 int x, y; 3847@dots{} 3848@end example 3849 3850It is up to you to make sure that the assembler names you choose do not 3851conflict with any other assembler symbols. Also, you must not use a 3852register name; that would produce completely invalid assembler code. GCC 3853does not as yet have the ability to store static variables in registers. 3854Perhaps that will be added. 3855 3856@node Explicit Reg Vars 3857@section Variables in Specified Registers 3858@cindex explicit register variables 3859@cindex variables in specified registers 3860@cindex specified registers 3861@cindex registers, global allocation 3862 3863GNU C allows you to put a few global variables into specified hardware 3864registers. You can also specify the register in which an ordinary 3865register variable should be allocated. 3866 3867@itemize @bullet 3868@item 3869Global register variables reserve registers throughout the program. 3870This may be useful in programs such as programming language 3871interpreters which have a couple of global variables that are accessed 3872very often. 3873 3874@item 3875Local register variables in specific registers do not reserve the 3876registers. The compiler's data flow analysis is capable of determining 3877where the specified registers contain live values, and where they are 3878available for other uses. Stores into local register variables may be deleted 3879when they appear to be dead according to dataflow analysis. References 3880to local register variables may be deleted or moved or simplified. 3881 3882These local variables are sometimes convenient for use with the extended 3883@code{asm} feature (@pxref{Extended Asm}), if you want to write one 3884output of the assembler instruction directly into a particular register. 3885(This will work provided the register you specify fits the constraints 3886specified for that operand in the @code{asm}.) 3887@end itemize 3888 3889@menu 3890* Global Reg Vars:: 3891* Local Reg Vars:: 3892@end menu 3893 3894@node Global Reg Vars 3895@subsection Defining Global Register Variables 3896@cindex global register variables 3897@cindex registers, global variables in 3898 3899You can define a global register variable in GNU C like this: 3900 3901@example 3902register int *foo asm ("a5"); 3903@end example 3904 3905@noindent 3906Here @code{a5} is the name of the register which should be used. Choose a 3907register which is normally saved and restored by function calls on your 3908machine, so that library routines will not clobber it. 3909 3910Naturally the register name is cpu-dependent, so you would need to 3911conditionalize your program according to cpu type. The register 3912@code{a5} would be a good choice on a 68000 for a variable of pointer 3913type. On machines with register windows, be sure to choose a ``global'' 3914register that is not affected magically by the function call mechanism. 3915 3916In addition, operating systems on one type of cpu may differ in how they 3917name the registers; then you would need additional conditionals. For 3918example, some 68000 operating systems call this register @code{%a5}. 3919 3920Eventually there may be a way of asking the compiler to choose a register 3921automatically, but first we need to figure out how it should choose and 3922how to enable you to guide the choice. No solution is evident. 3923 3924Defining a global register variable in a certain register reserves that 3925register entirely for this use, at least within the current compilation. 3926The register will not be allocated for any other purpose in the functions 3927in the current compilation. The register will not be saved and restored by 3928these functions. Stores into this register are never deleted even if they 3929would appear to be dead, but references may be deleted or moved or 3930simplified. 3931 3932It is not safe to access the global register variables from signal 3933handlers, or from more than one thread of control, because the system 3934library routines may temporarily use the register for other things (unless 3935you recompile them specially for the task at hand). 3936 3937@cindex @code{qsort}, and global register variables 3938It is not safe for one function that uses a global register variable to 3939call another such function @code{foo} by way of a third function 3940@code{lose} that was compiled without knowledge of this variable (i.e.@: in a 3941different source file in which the variable wasn't declared). This is 3942because @code{lose} might save the register and put some other value there. 3943For example, you can't expect a global register variable to be available in 3944the comparison-function that you pass to @code{qsort}, since @code{qsort} 3945might have put something else in that register. (If you are prepared to 3946recompile @code{qsort} with the same global register variable, you can 3947solve this problem.) 3948 3949If you want to recompile @code{qsort} or other source files which do not 3950actually use your global register variable, so that they will not use that 3951register for any other purpose, then it suffices to specify the compiler 3952option @option{-ffixed-@var{reg}}. You need not actually add a global 3953register declaration to their source code. 3954 3955A function which can alter the value of a global register variable cannot 3956safely be called from a function compiled without this variable, because it 3957could clobber the value the caller expects to find there on return. 3958Therefore, the function which is the entry point into the part of the 3959program that uses the global register variable must explicitly save and 3960restore the value which belongs to its caller. 3961 3962@cindex register variable after @code{longjmp} 3963@cindex global register after @code{longjmp} 3964@cindex value after @code{longjmp} 3965@findex longjmp 3966@findex setjmp 3967On most machines, @code{longjmp} will restore to each global register 3968variable the value it had at the time of the @code{setjmp}. On some 3969machines, however, @code{longjmp} will not change the value of global 3970register variables. To be portable, the function that called @code{setjmp} 3971should make other arrangements to save the values of the global register 3972variables, and to restore them in a @code{longjmp}. This way, the same 3973thing will happen regardless of what @code{longjmp} does. 3974 3975All global register variable declarations must precede all function 3976definitions. If such a declaration could appear after function 3977definitions, the declaration would be too late to prevent the register from 3978being used for other purposes in the preceding functions. 3979 3980Global register variables may not have initial values, because an 3981executable file has no means to supply initial contents for a register. 3982 3983On the Sparc, there are reports that g3 @dots{} g7 are suitable 3984registers, but certain library functions, such as @code{getwd}, as well 3985as the subroutines for division and remainder, modify g3 and g4. g1 and 3986g2 are local temporaries. 3987 3988On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7. 3989Of course, it will not do to use more than a few of those. 3990 3991@node Local Reg Vars 3992@subsection Specifying Registers for Local Variables 3993@cindex local variables, specifying registers 3994@cindex specifying registers for local variables 3995@cindex registers for local variables 3996 3997You can define a local register variable with a specified register 3998like this: 3999 4000@example 4001register int *foo asm ("a5"); 4002@end example 4003 4004@noindent 4005Here @code{a5} is the name of the register which should be used. Note 4006that this is the same syntax used for defining global register 4007variables, but for a local variable it would appear within a function. 4008 4009Naturally the register name is cpu-dependent, but this is not a 4010problem, since specific registers are most often useful with explicit 4011assembler instructions (@pxref{Extended Asm}). Both of these things 4012generally require that you conditionalize your program according to 4013cpu type. 4014 4015In addition, operating systems on one type of cpu may differ in how they 4016name the registers; then you would need additional conditionals. For 4017example, some 68000 operating systems call this register @code{%a5}. 4018 4019Defining such a register variable does not reserve the register; it 4020remains available for other uses in places where flow control determines 4021the variable's value is not live. However, these registers are made 4022unavailable for use in the reload pass; excessive use of this feature 4023leaves the compiler too few available registers to compile certain 4024functions. 4025 4026This option does not guarantee that GCC will generate code that has 4027this variable in the register you specify at all times. You may not 4028code an explicit reference to this register in an @code{asm} statement 4029and assume it will always refer to this variable. 4030 4031Stores into local register variables may be deleted when they appear to be dead 4032according to dataflow analysis. References to local register variables may 4033be deleted or moved or simplified. 4034 4035@node Alternate Keywords 4036@section Alternate Keywords 4037@cindex alternate keywords 4038@cindex keywords, alternate 4039 4040The option @option{-traditional} disables certain keywords; 4041@option{-ansi} and the various @option{-std} options disable certain 4042others. This causes trouble when you want to use GNU C extensions, or 4043ISO C features, in a general-purpose header file that should be usable 4044by all programs, including ISO C programs and traditional ones. The 4045keywords @code{asm}, @code{typeof} and @code{inline} cannot be used 4046since they won't work in a program compiled with @option{-ansi} 4047(although @code{inline} can be used in a program compiled with 4048@option{-std=c99}), while the keywords @code{const}, @code{volatile}, 4049@code{signed}, @code{typeof} and @code{inline} won't work in a program 4050compiled with @option{-traditional}. The ISO C99 keyword 4051@code{restrict} is only available when @option{-std=gnu99} (which will 4052eventually be the default) or @option{-std=c99} (or the equivalent 4053@option{-std=iso9899:1999}) is used. 4054 4055The way to solve these problems is to put @samp{__} at the beginning and 4056end of each problematical keyword. For example, use @code{__asm__} 4057instead of @code{asm}, @code{__const__} instead of @code{const}, and 4058@code{__inline__} instead of @code{inline}. 4059 4060Other C compilers won't accept these alternative keywords; if you want to 4061compile with another compiler, you can define the alternate keywords as 4062macros to replace them with the customary keywords. It looks like this: 4063 4064@example 4065#ifndef __GNUC__ 4066#define __asm__ asm 4067#endif 4068@end example 4069 4070@findex __extension__ 4071@opindex pedantic 4072@option{-pedantic} and other options cause warnings for many GNU C extensions. 4073You can 4074prevent such warnings within one expression by writing 4075@code{__extension__} before the expression. @code{__extension__} has no 4076effect aside from this. 4077 4078@node Incomplete Enums 4079@section Incomplete @code{enum} Types 4080 4081You can define an @code{enum} tag without specifying its possible values. 4082This results in an incomplete type, much like what you get if you write 4083@code{struct foo} without describing the elements. A later declaration 4084which does specify the possible values completes the type. 4085 4086You can't allocate variables or storage using the type while it is 4087incomplete. However, you can work with pointers to that type. 4088 4089This extension may not be very useful, but it makes the handling of 4090@code{enum} more consistent with the way @code{struct} and @code{union} 4091are handled. 4092 4093This extension is not supported by GNU C++. 4094 4095@node Function Names 4096@section Function Names as Strings 4097@cindex @code{__FUNCTION__} identifier 4098@cindex @code{__PRETTY_FUNCTION__} identifier 4099@cindex @code{__func__} identifier 4100 4101GCC predefines two magic identifiers to hold the name of the current 4102function. The identifier @code{__FUNCTION__} holds the name of the function 4103as it appears in the source. The identifier @code{__PRETTY_FUNCTION__} 4104holds the name of the function pretty printed in a language specific 4105fashion. 4106 4107These names are always the same in a C function, but in a C++ function 4108they may be different. For example, this program: 4109 4110@smallexample 4111extern "C" @{ 4112extern int printf (char *, ...); 4113@} 4114 4115class a @{ 4116 public: 4117 sub (int i) 4118 @{ 4119 printf ("__FUNCTION__ = %s\n", __FUNCTION__); 4120 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); 4121 @} 4122@}; 4123 4124int 4125main (void) 4126@{ 4127 a ax; 4128 ax.sub (0); 4129 return 0; 4130@} 4131@end smallexample 4132 4133@noindent 4134gives this output: 4135 4136@smallexample 4137__FUNCTION__ = sub 4138__PRETTY_FUNCTION__ = int a::sub (int) 4139@end smallexample 4140 4141The compiler automagically replaces the identifiers with a string 4142literal containing the appropriate name. Thus, they are neither 4143preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor 4144variables. This means that they catenate with other string literals, and 4145that they can be used to initialize char arrays. For example 4146 4147@smallexample 4148char here[] = "Function " __FUNCTION__ " in " __FILE__; 4149@end smallexample 4150 4151On the other hand, @samp{#ifdef __FUNCTION__} does not have any special 4152meaning inside a function, since the preprocessor does not do anything 4153special with the identifier @code{__FUNCTION__}. 4154 4155Note that these semantics are deprecated, and that GCC 3.2 will handle 4156@code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} the same way as 4157@code{__func__}. @code{__func__} is defined by the ISO standard C99: 4158 4159@display 4160The identifier @code{__func__} is implicitly declared by the translator 4161as if, immediately following the opening brace of each function 4162definition, the declaration 4163 4164@smallexample 4165static const char __func__[] = "function-name"; 4166@end smallexample 4167 4168appeared, where function-name is the name of the lexically-enclosing 4169function. This name is the unadorned name of the function. 4170@end display 4171 4172By this definition, @code{__func__} is a variable, not a string literal. 4173In particular, @code{__func__} does not catenate with other string 4174literals. 4175 4176In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are 4177variables, declared in the same way as @code{__func__}. 4178 4179@node Return Address 4180@section Getting the Return or Frame Address of a Function 4181 4182These functions may be used to get information about the callers of a 4183function. 4184 4185@deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level}) 4186This function returns the return address of the current function, or of 4187one of its callers. The @var{level} argument is number of frames to 4188scan up the call stack. A value of @code{0} yields the return address 4189of the current function, a value of @code{1} yields the return address 4190of the caller of the current function, and so forth. 4191 4192The @var{level} argument must be a constant integer. 4193 4194On some machines it may be impossible to determine the return address of 4195any function other than the current one; in such cases, or when the top 4196of the stack has been reached, this function will return @code{0} or a 4197random value. In addition, @code{__builtin_frame_address} may be used 4198to determine if the top of the stack has been reached. 4199 4200This function should only be used with a nonzero argument for debugging 4201purposes. 4202@end deftypefn 4203 4204@deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level}) 4205This function is similar to @code{__builtin_return_address}, but it 4206returns the address of the function frame rather than the return address 4207of the function. Calling @code{__builtin_frame_address} with a value of 4208@code{0} yields the frame address of the current function, a value of 4209@code{1} yields the frame address of the caller of the current function, 4210and so forth. 4211 4212The frame is the area on the stack which holds local variables and saved 4213registers. The frame address is normally the address of the first word 4214pushed on to the stack by the function. However, the exact definition 4215depends upon the processor and the calling convention. If the processor 4216has a dedicated frame pointer register, and the function has a frame, 4217then @code{__builtin_frame_address} will return the value of the frame 4218pointer register. 4219 4220On some machines it may be impossible to determine the frame address of 4221any function other than the current one; in such cases, or when the top 4222of the stack has been reached, this function will return @code{0} if 4223the first frame pointer is properly initialized by the startup code. 4224 4225This function should only be used with a nonzero argument for debugging 4226purposes. 4227@end deftypefn 4228 4229@node Vector Extensions 4230@section Using vector instructions through built-in functions 4231 4232On some targets, the instruction set contains SIMD vector instructions that 4233operate on multiple values contained in one large register at the same time. 4234For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used 4235this way. 4236 4237The first step in using these extensions is to provide the necessary data 4238types. This should be done using an appropriate @code{typedef}: 4239 4240@example 4241typedef int v4si __attribute__ ((mode(V4SI))); 4242@end example 4243 4244The base type @code{int} is effectively ignored by the compiler, the 4245actual properties of the new type @code{v4si} are defined by the 4246@code{__attribute__}. It defines the machine mode to be used; for vector 4247types these have the form @code{V@var{n}@var{B}}; @var{n} should be the 4248number of elements in the vector, and @var{B} should be the base mode of the 4249individual elements. The following can be used as base modes: 4250 4251@table @code 4252@item QI 4253An integer that is as wide as the smallest addressable unit, usually 8 bits. 4254@item HI 4255An integer, twice as wide as a QI mode integer, usually 16 bits. 4256@item SI 4257An integer, four times as wide as a QI mode integer, usually 32 bits. 4258@item DI 4259An integer, eight times as wide as a QI mode integer, usually 64 bits. 4260@item SF 4261A floating point value, as wide as a SI mode integer, usually 32 bits. 4262@item DF 4263A floating point value, as wide as a DI mode integer, usually 64 bits. 4264@end table 4265 4266Not all base types or combinations are always valid; which modes can be used 4267is determined by the target machine. For example, if targetting the i386 MMX 4268extensions, only @code{V8QI}, @code{V4HI} and @code{V2SI} are allowed modes. 4269 4270There are no @code{V1xx} vector modes - they would be identical to the 4271corresponding base mode. 4272 4273There is no distinction between signed and unsigned vector modes. This 4274distinction is made by the operations that perform on the vectors, not 4275by the data type. 4276 4277The types defined in this manner are somewhat special, they cannot be 4278used with most normal C operations (i.e., a vector addition can @emph{not} 4279be represented by a normal addition of two vector type variables). You 4280can declare only variables and use them in function calls and returns, as 4281well as in assignments and some casts. It is possible to cast from one 4282vector type to another, provided they are of the same size (in fact, you 4283can also cast vectors to and from other datatypes of the same size). 4284 4285A port that supports vector operations provides a set of built-in functions 4286that can be used to operate on vectors. For example, a function to add two 4287vectors and multiply the result by a third could look like this: 4288 4289@example 4290v4si f (v4si a, v4si b, v4si c) 4291@{ 4292 v4si tmp = __builtin_addv4si (a, b); 4293 return __builtin_mulv4si (tmp, c); 4294@} 4295 4296@end example 4297 4298@node Other Builtins 4299@section Other built-in functions provided by GCC 4300@cindex built-in functions 4301@findex __builtin_isgreater 4302@findex __builtin_isgreaterequal 4303@findex __builtin_isless 4304@findex __builtin_islessequal 4305@findex __builtin_islessgreater 4306@findex __builtin_isunordered 4307@findex abort 4308@findex abs 4309@findex alloca 4310@findex bcmp 4311@findex bzero 4312@findex cimag 4313@findex cimagf 4314@findex cimagl 4315@findex conj 4316@findex conjf 4317@findex conjl 4318@findex cos 4319@findex cosf 4320@findex cosl 4321@findex creal 4322@findex crealf 4323@findex creall 4324@findex exit 4325@findex _exit 4326@findex _Exit 4327@findex fabs 4328@findex fabsf 4329@findex fabsl 4330@findex ffs 4331@findex fprintf 4332@findex fprintf_unlocked 4333@findex fputs 4334@findex fputs_unlocked 4335@findex imaxabs 4336@findex index 4337@findex labs 4338@findex llabs 4339@findex memcmp 4340@findex memcpy 4341@findex memset 4342@findex printf 4343@findex printf_unlocked 4344@findex rindex 4345@findex sin 4346@findex sinf 4347@findex sinl 4348@findex sqrt 4349@findex sqrtf 4350@findex sqrtl 4351@findex strcat 4352@findex strchr 4353@findex strcmp 4354@findex strcpy 4355@findex strcspn 4356@findex strlen 4357@findex strncat 4358@findex strncmp 4359@findex strncpy 4360@findex strpbrk 4361@findex strrchr 4362@findex strspn 4363@findex strstr 4364 4365GCC provides a large number of built-in functions other than the ones 4366mentioned above. Some of these are for internal use in the processing 4367of exceptions or variable-length argument lists and will not be 4368documented here because they may change from time to time; we do not 4369recommend general use of these functions. 4370 4371The remaining functions are provided for optimization purposes. 4372 4373@opindex fno-builtin 4374GCC includes built-in versions of many of the functions in the standard 4375C library. The versions prefixed with @code{__builtin_} will always be 4376treated as having the same meaning as the C library function even if you 4377specify the @option{-fno-builtin} option. (@pxref{C Dialect Options}) 4378Many of these functions are only optimized in certain cases; if they are 4379not optimized in a particular case, a call to the library function will 4380be emitted. 4381 4382@opindex ansi 4383@opindex std 4384The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit} 4385are recognized and presumed not to return, but otherwise are not built 4386in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi}, 4387@option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in 4388strict C89 mode (@option{-ansi} or @option{-std=c89}). 4389 4390Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp}, 4391@code{bzero}, @code{index}, @code{rindex}, @code{ffs}, @code{fputs_unlocked}, 4392@code{printf_unlocked} and @code{fprintf_unlocked} may be handled as 4393built-in functions. All these functions have corresponding versions 4394prefixed with @code{__builtin_}, which may be used even in strict C89 4395mode. 4396 4397The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl}, 4398@code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf}, 4399@code{cimagl}, @code{llabs} and @code{imaxabs} are handled as built-in 4400functions except in strict ISO C89 mode. There are also built-in 4401versions of the ISO C99 functions @code{cosf}, @code{cosl}, 4402@code{fabsf}, @code{fabsl}, @code{sinf}, @code{sinl}, @code{sqrtf}, and 4403@code{sqrtl}, that are recognized in any mode since ISO C89 reserves 4404these names for the purpose to which ISO C99 puts them. All these 4405functions have corresponding versions prefixed with @code{__builtin_}. 4406 4407The ISO C89 functions @code{abs}, @code{cos}, @code{fabs}, 4408@code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy}, 4409@code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat}, 4410@code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn}, 4411@code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy}, 4412@code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr} are all 4413recognized as built-in functions unless @option{-fno-builtin} is 4414specified (or @option{-fno-builtin-@var{function}} is specified for an 4415individual function). All of these functions have corresponding 4416versions prefixed with @code{__builtin_}. 4417 4418GCC provides built-in versions of the ISO C99 floating point comparison 4419macros that avoid raising exceptions for unordered operands. They have 4420the same names as the standard macros ( @code{isgreater}, 4421@code{isgreaterequal}, @code{isless}, @code{islessequal}, 4422@code{islessgreater}, and @code{isunordered}) , with @code{__builtin_} 4423prefixed. We intend for a library implementor to be able to simply 4424@code{#define} each standard macro to its built-in equivalent. 4425 4426@deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2}) 4427 4428You can use the built-in function @code{__builtin_types_compatible_p} to 4429determine whether two types are the same. 4430 4431This built-in function returns 1 if the unqualified versions of the 4432types @var{type1} and @var{type2} (which are types, not expressions) are 4433compatible, 0 otherwise. The result of this built-in function can be 4434used in integer constant expressions. 4435 4436This built-in function ignores top level qualifiers (e.g., @code{const}, 4437@code{volatile}). For example, @code{int} is equivalent to @code{const 4438int}. 4439 4440The type @code{int[]} and @code{int[5]} are compatible. On the other 4441hand, @code{int} and @code{char *} are not compatible, even if the size 4442of their types, on the particular architecture are the same. Also, the 4443amount of pointer indirection is taken into account when determining 4444similarity. Consequently, @code{short *} is not similar to 4445@code{short **}. Furthermore, two types that are typedefed are 4446considered compatible if their underlying types are compatible. 4447 4448An @code{enum} type is considered to be compatible with another 4449@code{enum} type. For example, @code{enum @{foo, bar@}} is similar to 4450@code{enum @{hot, dog@}}. 4451 4452You would typically use this function in code whose execution varies 4453depending on the arguments' types. For example: 4454 4455@smallexample 4456#define foo(x) \ 4457 (@{ \ 4458 typeof (x) tmp; \ 4459 if (__builtin_types_compatible_p (typeof (x), long double)) \ 4460 tmp = foo_long_double (tmp); \ 4461 else if (__builtin_types_compatible_p (typeof (x), double)) \ 4462 tmp = foo_double (tmp); \ 4463 else if (__builtin_types_compatible_p (typeof (x), float)) \ 4464 tmp = foo_float (tmp); \ 4465 else \ 4466 abort (); \ 4467 tmp; \ 4468 @}) 4469@end smallexample 4470 4471@emph{Note:} This construct is only available for C. 4472 4473@end deftypefn 4474 4475@deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2}) 4476 4477You can use the built-in function @code{__builtin_choose_expr} to 4478evaluate code depending on the value of a constant expression. This 4479built-in function returns @var{exp1} if @var{const_exp}, which is a 4480constant expression that must be able to be determined at compile time, 4481is nonzero. Otherwise it returns 0. 4482 4483This built-in function is analogous to the @samp{? :} operator in C, 4484except that the expression returned has its type unaltered by promotion 4485rules. Also, the built-in function does not evaluate the expression 4486that was not chosen. For example, if @var{const_exp} evaluates to true, 4487@var{exp2} is not evaluated even if it has side-effects. 4488 4489This built-in function can return an lvalue if the chosen argument is an 4490lvalue. 4491 4492If @var{exp1} is returned, the return type is the same as @var{exp1}'s 4493type. Similarly, if @var{exp2} is returned, its return type is the same 4494as @var{exp2}. 4495 4496Example: 4497 4498@smallexample 4499#define foo(x) \ 4500 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), double), \ 4501 foo_double (x), \ 4502 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), float), \ 4503 foo_float (x), \ 4504 /* @r{The void expression results in a compile-time error} \ 4505 @r{when assigning the result to something.} */ \ 4506 (void)0)) 4507@end smallexample 4508 4509@emph{Note:} This construct is only available for C. Furthermore, the 4510unused expression (@var{exp1} or @var{exp2} depending on the value of 4511@var{const_exp}) may still generate syntax errors. This may change in 4512future revisions. 4513 4514@end deftypefn 4515 4516@deftypefn {Built-in Function} int __builtin_constant_p (@var{exp}) 4517You can use the built-in function @code{__builtin_constant_p} to 4518determine if a value is known to be constant at compile-time and hence 4519that GCC can perform constant-folding on expressions involving that 4520value. The argument of the function is the value to test. The function 4521returns the integer 1 if the argument is known to be a compile-time 4522constant and 0 if it is not known to be a compile-time constant. A 4523return of 0 does not indicate that the value is @emph{not} a constant, 4524but merely that GCC cannot prove it is a constant with the specified 4525value of the @option{-O} option. 4526 4527You would typically use this function in an embedded application where 4528memory was a critical resource. If you have some complex calculation, 4529you may want it to be folded if it involves constants, but need to call 4530a function if it does not. For example: 4531 4532@smallexample 4533#define Scale_Value(X) \ 4534 (__builtin_constant_p (X) \ 4535 ? ((X) * SCALE + OFFSET) : Scale (X)) 4536@end smallexample 4537 4538You may use this built-in function in either a macro or an inline 4539function. However, if you use it in an inlined function and pass an 4540argument of the function as the argument to the built-in, GCC will 4541never return 1 when you call the inline function with a string constant 4542or compound literal (@pxref{Compound Literals}) and will not return 1 4543when you pass a constant numeric value to the inline function unless you 4544specify the @option{-O} option. 4545 4546You may also use @code{__builtin_constant_p} in initializers for static 4547data. For instance, you can write 4548 4549@smallexample 4550static const int table[] = @{ 4551 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1, 4552 /* ... */ 4553@}; 4554@end smallexample 4555 4556@noindent 4557This is an acceptable initializer even if @var{EXPRESSION} is not a 4558constant expression. GCC must be more conservative about evaluating the 4559built-in in this case, because it has no opportunity to perform 4560optimization. 4561 4562Previous versions of GCC did not accept this built-in in data 4563initializers. The earliest version where it is completely safe is 45643.0.1. 4565@end deftypefn 4566 4567@deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c}) 4568@opindex fprofile-arcs 4569You may use @code{__builtin_expect} to provide the compiler with 4570branch prediction information. In general, you should prefer to 4571use actual profile feedback for this (@option{-fprofile-arcs}), as 4572programmers are notoriously bad at predicting how their programs 4573actually perform. However, there are applications in which this 4574data is hard to collect. 4575 4576The return value is the value of @var{exp}, which should be an 4577integral expression. The value of @var{c} must be a compile-time 4578constant. The semantics of the built-in are that it is expected 4579that @var{exp} == @var{c}. For example: 4580 4581@smallexample 4582if (__builtin_expect (x, 0)) 4583 foo (); 4584@end smallexample 4585 4586@noindent 4587would indicate that we do not expect to call @code{foo}, since 4588we expect @code{x} to be zero. Since you are limited to integral 4589expressions for @var{exp}, you should use constructions such as 4590 4591@smallexample 4592if (__builtin_expect (ptr != NULL, 1)) 4593 error (); 4594@end smallexample 4595 4596@noindent 4597when testing pointer or floating-point values. 4598@end deftypefn 4599 4600@deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...) 4601This function is used to minimize cache-miss latency by moving data into 4602a cache before it is accessed. 4603You can insert calls to @code{__builtin_prefetch} into code for which 4604you know addresses of data in memory that is likely to be accessed soon. 4605If the target supports them, data prefetch instructions will be generated. 4606If the prefetch is done early enough before the access then the data will 4607be in the cache by the time it is accessed. 4608 4609The value of @var{addr} is the address of the memory to prefetch. 4610There are two optional arguments, @var{rw} and @var{locality}. 4611The value of @var{rw} is a compile-time constant one or zero; one 4612means that the prefetch is preparing for a write to the memory address 4613and zero, the default, means that the prefetch is preparing for a read. 4614The value @var{locality} must be a compile-time constant integer between 4615zero and three. A value of zero means that the data has no temporal 4616locality, so it need not be left in the cache after the access. A value 4617of three means that the data has a high degree of temporal locality and 4618should be left in all levels of cache possible. Values of one and two 4619mean, respectively, a low or moderate degree of temporal locality. The 4620default is three. 4621 4622@smallexample 4623for (i = 0; i < n; i++) 4624 @{ 4625 a[i] = a[i] + b[i]; 4626 __builtin_prefetch (&a[i+j], 1, 1); 4627 __builtin_prefetch (&b[i+j], 0, 1); 4628 /* ... */ 4629 @} 4630@end smallexample 4631 4632Data prefetch does not generate faults if @var{addr} is invalid, but 4633the address expression itself must be valid. For example, a prefetch 4634of @code{p->next} will not fault if @code{p->next} is not a valid 4635address, but evaluation will fault if @code{p} is not a valid address. 4636 4637If the target does not support data prefetch, the address expression 4638is evaluated if it includes side effects but no other code is generated 4639and GCC does not issue a warning. 4640@end deftypefn 4641 4642@node Target Builtins 4643@section Built-in Functions Specific to Particular Target Machines 4644 4645On some target machines, GCC supports many built-in functions specific 4646to those machines. Generally these generate calls to specific machine 4647instructions, but allow the compiler to schedule those calls. 4648 4649@menu 4650* X86 Built-in Functions:: 4651* PowerPC AltiVec Built-in Functions:: 4652@end menu 4653 4654@node X86 Built-in Functions 4655@subsection X86 Built-in Functions 4656 4657These built-in functions are available for the i386 and x86-64 family 4658of computers, depending on the command-line switches used. 4659 4660The following machine modes are available for use with MMX built-in functions 4661(@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers, 4662@code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a 4663vector of eight 8-bit integers. Some of the built-in functions operate on 4664MMX registers as a whole 64-bit entity, these use @code{DI} as their mode. 4665 4666If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector 4667of two 32-bit floating point values. 4668 4669If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit 4670floating point values. Some instructions use a vector of four 32-bit 4671integers, these use @code{V4SI}. Finally, some instructions operate on an 4672entire vector register, interpreting it as a 128-bit integer, these use mode 4673@code{TI}. 4674 4675The following built-in functions are made available by @option{-mmmx}. 4676All of them generate the machine instruction that is part of the name. 4677 4678@example 4679v8qi __builtin_ia32_paddb (v8qi, v8qi) 4680v4hi __builtin_ia32_paddw (v4hi, v4hi) 4681v2si __builtin_ia32_paddd (v2si, v2si) 4682v8qi __builtin_ia32_psubb (v8qi, v8qi) 4683v4hi __builtin_ia32_psubw (v4hi, v4hi) 4684v2si __builtin_ia32_psubd (v2si, v2si) 4685v8qi __builtin_ia32_paddsb (v8qi, v8qi) 4686v4hi __builtin_ia32_paddsw (v4hi, v4hi) 4687v8qi __builtin_ia32_psubsb (v8qi, v8qi) 4688v4hi __builtin_ia32_psubsw (v4hi, v4hi) 4689v8qi __builtin_ia32_paddusb (v8qi, v8qi) 4690v4hi __builtin_ia32_paddusw (v4hi, v4hi) 4691v8qi __builtin_ia32_psubusb (v8qi, v8qi) 4692v4hi __builtin_ia32_psubusw (v4hi, v4hi) 4693v4hi __builtin_ia32_pmullw (v4hi, v4hi) 4694v4hi __builtin_ia32_pmulhw (v4hi, v4hi) 4695di __builtin_ia32_pand (di, di) 4696di __builtin_ia32_pandn (di,di) 4697di __builtin_ia32_por (di, di) 4698di __builtin_ia32_pxor (di, di) 4699v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi) 4700v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi) 4701v2si __builtin_ia32_pcmpeqd (v2si, v2si) 4702v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi) 4703v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi) 4704v2si __builtin_ia32_pcmpgtd (v2si, v2si) 4705v8qi __builtin_ia32_punpckhbw (v8qi, v8qi) 4706v4hi __builtin_ia32_punpckhwd (v4hi, v4hi) 4707v2si __builtin_ia32_punpckhdq (v2si, v2si) 4708v8qi __builtin_ia32_punpcklbw (v8qi, v8qi) 4709v4hi __builtin_ia32_punpcklwd (v4hi, v4hi) 4710v2si __builtin_ia32_punpckldq (v2si, v2si) 4711v8qi __builtin_ia32_packsswb (v4hi, v4hi) 4712v4hi __builtin_ia32_packssdw (v2si, v2si) 4713v8qi __builtin_ia32_packuswb (v4hi, v4hi) 4714@end example 4715 4716The following built-in functions are made available either with 4717@option{-msse}, or with a combination of @option{-m3dnow} and 4718@option{-march=athlon}. All of them generate the machine 4719instruction that is part of the name. 4720 4721@example 4722v4hi __builtin_ia32_pmulhuw (v4hi, v4hi) 4723v8qi __builtin_ia32_pavgb (v8qi, v8qi) 4724v4hi __builtin_ia32_pavgw (v4hi, v4hi) 4725v4hi __builtin_ia32_psadbw (v8qi, v8qi) 4726v8qi __builtin_ia32_pmaxub (v8qi, v8qi) 4727v4hi __builtin_ia32_pmaxsw (v4hi, v4hi) 4728v8qi __builtin_ia32_pminub (v8qi, v8qi) 4729v4hi __builtin_ia32_pminsw (v4hi, v4hi) 4730int __builtin_ia32_pextrw (v4hi, int) 4731v4hi __builtin_ia32_pinsrw (v4hi, int, int) 4732int __builtin_ia32_pmovmskb (v8qi) 4733void __builtin_ia32_maskmovq (v8qi, v8qi, char *) 4734void __builtin_ia32_movntq (di *, di) 4735void __builtin_ia32_sfence (void) 4736@end example 4737 4738The following built-in functions are available when @option{-msse} is used. 4739All of them generate the machine instruction that is part of the name. 4740 4741@example 4742int __builtin_ia32_comieq (v4sf, v4sf) 4743int __builtin_ia32_comineq (v4sf, v4sf) 4744int __builtin_ia32_comilt (v4sf, v4sf) 4745int __builtin_ia32_comile (v4sf, v4sf) 4746int __builtin_ia32_comigt (v4sf, v4sf) 4747int __builtin_ia32_comige (v4sf, v4sf) 4748int __builtin_ia32_ucomieq (v4sf, v4sf) 4749int __builtin_ia32_ucomineq (v4sf, v4sf) 4750int __builtin_ia32_ucomilt (v4sf, v4sf) 4751int __builtin_ia32_ucomile (v4sf, v4sf) 4752int __builtin_ia32_ucomigt (v4sf, v4sf) 4753int __builtin_ia32_ucomige (v4sf, v4sf) 4754v4sf __builtin_ia32_addps (v4sf, v4sf) 4755v4sf __builtin_ia32_subps (v4sf, v4sf) 4756v4sf __builtin_ia32_mulps (v4sf, v4sf) 4757v4sf __builtin_ia32_divps (v4sf, v4sf) 4758v4sf __builtin_ia32_addss (v4sf, v4sf) 4759v4sf __builtin_ia32_subss (v4sf, v4sf) 4760v4sf __builtin_ia32_mulss (v4sf, v4sf) 4761v4sf __builtin_ia32_divss (v4sf, v4sf) 4762v4si __builtin_ia32_cmpeqps (v4sf, v4sf) 4763v4si __builtin_ia32_cmpltps (v4sf, v4sf) 4764v4si __builtin_ia32_cmpleps (v4sf, v4sf) 4765v4si __builtin_ia32_cmpgtps (v4sf, v4sf) 4766v4si __builtin_ia32_cmpgeps (v4sf, v4sf) 4767v4si __builtin_ia32_cmpunordps (v4sf, v4sf) 4768v4si __builtin_ia32_cmpneqps (v4sf, v4sf) 4769v4si __builtin_ia32_cmpnltps (v4sf, v4sf) 4770v4si __builtin_ia32_cmpnleps (v4sf, v4sf) 4771v4si __builtin_ia32_cmpngtps (v4sf, v4sf) 4772v4si __builtin_ia32_cmpngeps (v4sf, v4sf) 4773v4si __builtin_ia32_cmpordps (v4sf, v4sf) 4774v4si __builtin_ia32_cmpeqss (v4sf, v4sf) 4775v4si __builtin_ia32_cmpltss (v4sf, v4sf) 4776v4si __builtin_ia32_cmpless (v4sf, v4sf) 4777v4si __builtin_ia32_cmpgtss (v4sf, v4sf) 4778v4si __builtin_ia32_cmpgess (v4sf, v4sf) 4779v4si __builtin_ia32_cmpunordss (v4sf, v4sf) 4780v4si __builtin_ia32_cmpneqss (v4sf, v4sf) 4781v4si __builtin_ia32_cmpnlts (v4sf, v4sf) 4782v4si __builtin_ia32_cmpnless (v4sf, v4sf) 4783v4si __builtin_ia32_cmpngtss (v4sf, v4sf) 4784v4si __builtin_ia32_cmpngess (v4sf, v4sf) 4785v4si __builtin_ia32_cmpordss (v4sf, v4sf) 4786v4sf __builtin_ia32_maxps (v4sf, v4sf) 4787v4sf __builtin_ia32_maxss (v4sf, v4sf) 4788v4sf __builtin_ia32_minps (v4sf, v4sf) 4789v4sf __builtin_ia32_minss (v4sf, v4sf) 4790v4sf __builtin_ia32_andps (v4sf, v4sf) 4791v4sf __builtin_ia32_andnps (v4sf, v4sf) 4792v4sf __builtin_ia32_orps (v4sf, v4sf) 4793v4sf __builtin_ia32_xorps (v4sf, v4sf) 4794v4sf __builtin_ia32_movss (v4sf, v4sf) 4795v4sf __builtin_ia32_movhlps (v4sf, v4sf) 4796v4sf __builtin_ia32_movlhps (v4sf, v4sf) 4797v4sf __builtin_ia32_unpckhps (v4sf, v4sf) 4798v4sf __builtin_ia32_unpcklps (v4sf, v4sf) 4799v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si) 4800v4sf __builtin_ia32_cvtsi2ss (v4sf, int) 4801v2si __builtin_ia32_cvtps2pi (v4sf) 4802int __builtin_ia32_cvtss2si (v4sf) 4803v2si __builtin_ia32_cvttps2pi (v4sf) 4804int __builtin_ia32_cvttss2si (v4sf) 4805v4sf __builtin_ia32_rcpps (v4sf) 4806v4sf __builtin_ia32_rsqrtps (v4sf) 4807v4sf __builtin_ia32_sqrtps (v4sf) 4808v4sf __builtin_ia32_rcpss (v4sf) 4809v4sf __builtin_ia32_rsqrtss (v4sf) 4810v4sf __builtin_ia32_sqrtss (v4sf) 4811v4sf __builtin_ia32_shufps (v4sf, v4sf, int) 4812void __builtin_ia32_movntps (float *, v4sf) 4813int __builtin_ia32_movmskps (v4sf) 4814@end example 4815 4816The following built-in functions are available when @option{-msse} is used. 4817 4818@table @code 4819@item v4sf __builtin_ia32_loadaps (float *) 4820Generates the @code{movaps} machine instruction as a load from memory. 4821@item void __builtin_ia32_storeaps (float *, v4sf) 4822Generates the @code{movaps} machine instruction as a store to memory. 4823@item v4sf __builtin_ia32_loadups (float *) 4824Generates the @code{movups} machine instruction as a load from memory. 4825@item void __builtin_ia32_storeups (float *, v4sf) 4826Generates the @code{movups} machine instruction as a store to memory. 4827@item v4sf __builtin_ia32_loadsss (float *) 4828Generates the @code{movss} machine instruction as a load from memory. 4829@item void __builtin_ia32_storess (float *, v4sf) 4830Generates the @code{movss} machine instruction as a store to memory. 4831@item v4sf __builtin_ia32_loadhps (v4sf, v2si *) 4832Generates the @code{movhps} machine instruction as a load from memory. 4833@item v4sf __builtin_ia32_loadlps (v4sf, v2si *) 4834Generates the @code{movlps} machine instruction as a load from memory 4835@item void __builtin_ia32_storehps (v4sf, v2si *) 4836Generates the @code{movhps} machine instruction as a store to memory. 4837@item void __builtin_ia32_storelps (v4sf, v2si *) 4838Generates the @code{movlps} machine instruction as a store to memory. 4839@end table 4840 4841The following built-in functions are available when @option{-m3dnow} is used. 4842All of them generate the machine instruction that is part of the name. 4843 4844@example 4845void __builtin_ia32_femms (void) 4846v8qi __builtin_ia32_pavgusb (v8qi, v8qi) 4847v2si __builtin_ia32_pf2id (v2sf) 4848v2sf __builtin_ia32_pfacc (v2sf, v2sf) 4849v2sf __builtin_ia32_pfadd (v2sf, v2sf) 4850v2si __builtin_ia32_pfcmpeq (v2sf, v2sf) 4851v2si __builtin_ia32_pfcmpge (v2sf, v2sf) 4852v2si __builtin_ia32_pfcmpgt (v2sf, v2sf) 4853v2sf __builtin_ia32_pfmax (v2sf, v2sf) 4854v2sf __builtin_ia32_pfmin (v2sf, v2sf) 4855v2sf __builtin_ia32_pfmul (v2sf, v2sf) 4856v2sf __builtin_ia32_pfrcp (v2sf) 4857v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf) 4858v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf) 4859v2sf __builtin_ia32_pfrsqrt (v2sf) 4860v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf) 4861v2sf __builtin_ia32_pfsub (v2sf, v2sf) 4862v2sf __builtin_ia32_pfsubr (v2sf, v2sf) 4863v2sf __builtin_ia32_pi2fd (v2si) 4864v4hi __builtin_ia32_pmulhrw (v4hi, v4hi) 4865@end example 4866 4867The following built-in functions are available when both @option{-m3dnow} 4868and @option{-march=athlon} are used. All of them generate the machine 4869instruction that is part of the name. 4870 4871@example 4872v2si __builtin_ia32_pf2iw (v2sf) 4873v2sf __builtin_ia32_pfnacc (v2sf, v2sf) 4874v2sf __builtin_ia32_pfpnacc (v2sf, v2sf) 4875v2sf __builtin_ia32_pi2fw (v2si) 4876v2sf __builtin_ia32_pswapdsf (v2sf) 4877v2si __builtin_ia32_pswapdsi (v2si) 4878@end example 4879 4880@node PowerPC AltiVec Built-in Functions 4881@subsection PowerPC AltiVec Built-in Functions 4882 4883These built-in functions are available for the PowerPC family 4884of computers, depending on the command-line switches used. 4885 4886The following machine modes are available for use with AltiVec built-in 4887functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four 488832-bit integers, @code{V4SF} for a vector of four 32-bit floating point 4889numbers, @code{V8HI} for a vector of eight 16-bit integers, and 4890@code{V16QI} for a vector of sixteen 8-bit integers. 4891 4892The following functions are made available by including 4893@code{<altivec.h>} and using @option{-maltivec} and 4894@option{-mabi=altivec}. The functions implement the functionality 4895described in Motorola's AltiVec Programming Interface Manual. 4896 4897@emph{Note:} Only the @code{<altivec.h>} interface is supported. 4898Internally, GCC uses built-in functions to achieve the functionality in 4899the aforementioned header file, but they are not supported and are 4900subject to change without notice. 4901 4902@smallexample 4903vector signed char vec_abs (vector signed char, vector signed char); 4904vector signed short vec_abs (vector signed short, vector signed short); 4905vector signed int vec_abs (vector signed int, vector signed int); 4906vector signed float vec_abs (vector signed float, vector signed float); 4907 4908vector signed char vec_abss (vector signed char, vector signed char); 4909vector signed short vec_abss (vector signed short, vector signed short); 4910 4911vector signed char vec_add (vector signed char, vector signed char); 4912vector unsigned char vec_add (vector signed char, vector unsigned char); 4913 4914vector unsigned char vec_add (vector unsigned char, vector signed char); 4915 4916vector unsigned char vec_add (vector unsigned char, 4917 vector unsigned char); 4918vector signed short vec_add (vector signed short, vector signed short); 4919vector unsigned short vec_add (vector signed short, 4920 vector unsigned short); 4921vector unsigned short vec_add (vector unsigned short, 4922 vector signed short); 4923vector unsigned short vec_add (vector unsigned short, 4924 vector unsigned short); 4925vector signed int vec_add (vector signed int, vector signed int); 4926vector unsigned int vec_add (vector signed int, vector unsigned int); 4927vector unsigned int vec_add (vector unsigned int, vector signed int); 4928vector unsigned int vec_add (vector unsigned int, vector unsigned int); 4929vector float vec_add (vector float, vector float); 4930 4931vector unsigned int vec_addc (vector unsigned int, vector unsigned int); 4932 4933vector unsigned char vec_adds (vector signed char, 4934 vector unsigned char); 4935vector unsigned char vec_adds (vector unsigned char, 4936 vector signed char); 4937vector unsigned char vec_adds (vector unsigned char, 4938 vector unsigned char); 4939vector signed char vec_adds (vector signed char, vector signed char); 4940vector unsigned short vec_adds (vector signed short, 4941 vector unsigned short); 4942vector unsigned short vec_adds (vector unsigned short, 4943 vector signed short); 4944vector unsigned short vec_adds (vector unsigned short, 4945 vector unsigned short); 4946vector signed short vec_adds (vector signed short, vector signed short); 4947 4948vector unsigned int vec_adds (vector signed int, vector unsigned int); 4949vector unsigned int vec_adds (vector unsigned int, vector signed int); 4950vector unsigned int vec_adds (vector unsigned int, vector unsigned int); 4951 4952vector signed int vec_adds (vector signed int, vector signed int); 4953 4954vector float vec_and (vector float, vector float); 4955vector float vec_and (vector float, vector signed int); 4956vector float vec_and (vector signed int, vector float); 4957vector signed int vec_and (vector signed int, vector signed int); 4958vector unsigned int vec_and (vector signed int, vector unsigned int); 4959vector unsigned int vec_and (vector unsigned int, vector signed int); 4960vector unsigned int vec_and (vector unsigned int, vector unsigned int); 4961vector signed short vec_and (vector signed short, vector signed short); 4962vector unsigned short vec_and (vector signed short, 4963 vector unsigned short); 4964vector unsigned short vec_and (vector unsigned short, 4965 vector signed short); 4966vector unsigned short vec_and (vector unsigned short, 4967 vector unsigned short); 4968vector signed char vec_and (vector signed char, vector signed char); 4969vector unsigned char vec_and (vector signed char, vector unsigned char); 4970 4971vector unsigned char vec_and (vector unsigned char, vector signed char); 4972 4973vector unsigned char vec_and (vector unsigned char, 4974 vector unsigned char); 4975 4976vector float vec_andc (vector float, vector float); 4977vector float vec_andc (vector float, vector signed int); 4978vector float vec_andc (vector signed int, vector float); 4979vector signed int vec_andc (vector signed int, vector signed int); 4980vector unsigned int vec_andc (vector signed int, vector unsigned int); 4981vector unsigned int vec_andc (vector unsigned int, vector signed int); 4982vector unsigned int vec_andc (vector unsigned int, vector unsigned int); 4983 4984vector signed short vec_andc (vector signed short, vector signed short); 4985 4986vector unsigned short vec_andc (vector signed short, 4987 vector unsigned short); 4988vector unsigned short vec_andc (vector unsigned short, 4989 vector signed short); 4990vector unsigned short vec_andc (vector unsigned short, 4991 vector unsigned short); 4992vector signed char vec_andc (vector signed char, vector signed char); 4993vector unsigned char vec_andc (vector signed char, 4994 vector unsigned char); 4995vector unsigned char vec_andc (vector unsigned char, 4996 vector signed char); 4997vector unsigned char vec_andc (vector unsigned char, 4998 vector unsigned char); 4999 5000vector unsigned char vec_avg (vector unsigned char, 5001 vector unsigned char); 5002vector signed char vec_avg (vector signed char, vector signed char); 5003vector unsigned short vec_avg (vector unsigned short, 5004 vector unsigned short); 5005vector signed short vec_avg (vector signed short, vector signed short); 5006vector unsigned int vec_avg (vector unsigned int, vector unsigned int); 5007vector signed int vec_avg (vector signed int, vector signed int); 5008 5009vector float vec_ceil (vector float); 5010 5011vector signed int vec_cmpb (vector float, vector float); 5012 5013vector signed char vec_cmpeq (vector signed char, vector signed char); 5014vector signed char vec_cmpeq (vector unsigned char, 5015 vector unsigned char); 5016vector signed short vec_cmpeq (vector signed short, 5017 vector signed short); 5018vector signed short vec_cmpeq (vector unsigned short, 5019 vector unsigned short); 5020vector signed int vec_cmpeq (vector signed int, vector signed int); 5021vector signed int vec_cmpeq (vector unsigned int, vector unsigned int); 5022vector signed int vec_cmpeq (vector float, vector float); 5023 5024vector signed int vec_cmpge (vector float, vector float); 5025 5026vector signed char vec_cmpgt (vector unsigned char, 5027 vector unsigned char); 5028vector signed char vec_cmpgt (vector signed char, vector signed char); 5029vector signed short vec_cmpgt (vector unsigned short, 5030 vector unsigned short); 5031vector signed short vec_cmpgt (vector signed short, 5032 vector signed short); 5033vector signed int vec_cmpgt (vector unsigned int, vector unsigned int); 5034vector signed int vec_cmpgt (vector signed int, vector signed int); 5035vector signed int vec_cmpgt (vector float, vector float); 5036 5037vector signed int vec_cmple (vector float, vector float); 5038 5039vector signed char vec_cmplt (vector unsigned char, 5040 vector unsigned char); 5041vector signed char vec_cmplt (vector signed char, vector signed char); 5042vector signed short vec_cmplt (vector unsigned short, 5043 vector unsigned short); 5044vector signed short vec_cmplt (vector signed short, 5045 vector signed short); 5046vector signed int vec_cmplt (vector unsigned int, vector unsigned int); 5047vector signed int vec_cmplt (vector signed int, vector signed int); 5048vector signed int vec_cmplt (vector float, vector float); 5049 5050vector float vec_ctf (vector unsigned int, const char); 5051vector float vec_ctf (vector signed int, const char); 5052 5053vector signed int vec_cts (vector float, const char); 5054 5055vector unsigned int vec_ctu (vector float, const char); 5056 5057void vec_dss (const char); 5058 5059void vec_dssall (void); 5060 5061void vec_dst (void *, int, const char); 5062 5063void vec_dstst (void *, int, const char); 5064 5065void vec_dststt (void *, int, const char); 5066 5067void vec_dstt (void *, int, const char); 5068 5069vector float vec_expte (vector float, vector float); 5070 5071vector float vec_floor (vector float, vector float); 5072 5073vector float vec_ld (int, vector float *); 5074vector float vec_ld (int, float *): 5075vector signed int vec_ld (int, int *); 5076vector signed int vec_ld (int, vector signed int *); 5077vector unsigned int vec_ld (int, vector unsigned int *); 5078vector unsigned int vec_ld (int, unsigned int *); 5079vector signed short vec_ld (int, short *, vector signed short *); 5080vector unsigned short vec_ld (int, unsigned short *, 5081 vector unsigned short *); 5082vector signed char vec_ld (int, signed char *); 5083vector signed char vec_ld (int, vector signed char *); 5084vector unsigned char vec_ld (int, unsigned char *); 5085vector unsigned char vec_ld (int, vector unsigned char *); 5086 5087vector signed char vec_lde (int, signed char *); 5088vector unsigned char vec_lde (int, unsigned char *); 5089vector signed short vec_lde (int, short *); 5090vector unsigned short vec_lde (int, unsigned short *); 5091vector float vec_lde (int, float *); 5092vector signed int vec_lde (int, int *); 5093vector unsigned int vec_lde (int, unsigned int *); 5094 5095void float vec_ldl (int, float *); 5096void float vec_ldl (int, vector float *); 5097void signed int vec_ldl (int, vector signed int *); 5098void signed int vec_ldl (int, int *); 5099void unsigned int vec_ldl (int, unsigned int *); 5100void unsigned int vec_ldl (int, vector unsigned int *); 5101void signed short vec_ldl (int, vector signed short *); 5102void signed short vec_ldl (int, short *); 5103void unsigned short vec_ldl (int, vector unsigned short *); 5104void unsigned short vec_ldl (int, unsigned short *); 5105void signed char vec_ldl (int, vector signed char *); 5106void signed char vec_ldl (int, signed char *); 5107void unsigned char vec_ldl (int, vector unsigned char *); 5108void unsigned char vec_ldl (int, unsigned char *); 5109 5110vector float vec_loge (vector float); 5111 5112vector unsigned char vec_lvsl (int, void *, int *); 5113 5114vector unsigned char vec_lvsr (int, void *, int *); 5115 5116vector float vec_madd (vector float, vector float, vector float); 5117 5118vector signed short vec_madds (vector signed short, vector signed short, 5119 vector signed short); 5120 5121vector unsigned char vec_max (vector signed char, vector unsigned char); 5122 5123vector unsigned char vec_max (vector unsigned char, vector signed char); 5124 5125vector unsigned char vec_max (vector unsigned char, 5126 vector unsigned char); 5127vector signed char vec_max (vector signed char, vector signed char); 5128vector unsigned short vec_max (vector signed short, 5129 vector unsigned short); 5130vector unsigned short vec_max (vector unsigned short, 5131 vector signed short); 5132vector unsigned short vec_max (vector unsigned short, 5133 vector unsigned short); 5134vector signed short vec_max (vector signed short, vector signed short); 5135vector unsigned int vec_max (vector signed int, vector unsigned int); 5136vector unsigned int vec_max (vector unsigned int, vector signed int); 5137vector unsigned int vec_max (vector unsigned int, vector unsigned int); 5138vector signed int vec_max (vector signed int, vector signed int); 5139vector float vec_max (vector float, vector float); 5140 5141vector signed char vec_mergeh (vector signed char, vector signed char); 5142vector unsigned char vec_mergeh (vector unsigned char, 5143 vector unsigned char); 5144vector signed short vec_mergeh (vector signed short, 5145 vector signed short); 5146vector unsigned short vec_mergeh (vector unsigned short, 5147 vector unsigned short); 5148vector float vec_mergeh (vector float, vector float); 5149vector signed int vec_mergeh (vector signed int, vector signed int); 5150vector unsigned int vec_mergeh (vector unsigned int, 5151 vector unsigned int); 5152 5153vector signed char vec_mergel (vector signed char, vector signed char); 5154vector unsigned char vec_mergel (vector unsigned char, 5155 vector unsigned char); 5156vector signed short vec_mergel (vector signed short, 5157 vector signed short); 5158vector unsigned short vec_mergel (vector unsigned short, 5159 vector unsigned short); 5160vector float vec_mergel (vector float, vector float); 5161vector signed int vec_mergel (vector signed int, vector signed int); 5162vector unsigned int vec_mergel (vector unsigned int, 5163 vector unsigned int); 5164 5165vector unsigned short vec_mfvscr (void); 5166 5167vector unsigned char vec_min (vector signed char, vector unsigned char); 5168 5169vector unsigned char vec_min (vector unsigned char, vector signed char); 5170 5171vector unsigned char vec_min (vector unsigned char, 5172 vector unsigned char); 5173vector signed char vec_min (vector signed char, vector signed char); 5174vector unsigned short vec_min (vector signed short, 5175 vector unsigned short); 5176vector unsigned short vec_min (vector unsigned short, 5177 vector signed short); 5178vector unsigned short vec_min (vector unsigned short, 5179 vector unsigned short); 5180vector signed short vec_min (vector signed short, vector signed short); 5181vector unsigned int vec_min (vector signed int, vector unsigned int); 5182vector unsigned int vec_min (vector unsigned int, vector signed int); 5183vector unsigned int vec_min (vector unsigned int, vector unsigned int); 5184vector signed int vec_min (vector signed int, vector signed int); 5185vector float vec_min (vector float, vector float); 5186 5187vector signed short vec_mladd (vector signed short, vector signed short, 5188 vector signed short); 5189vector signed short vec_mladd (vector signed short, 5190 vector unsigned short, 5191 vector unsigned short); 5192vector signed short vec_mladd (vector unsigned short, 5193 vector signed short, 5194 vector signed short); 5195vector unsigned short vec_mladd (vector unsigned short, 5196 vector unsigned short, 5197 vector unsigned short); 5198 5199vector signed short vec_mradds (vector signed short, 5200 vector signed short, 5201 vector signed short); 5202 5203vector unsigned int vec_msum (vector unsigned char, 5204 vector unsigned char, 5205 vector unsigned int); 5206vector signed int vec_msum (vector signed char, vector unsigned char, 5207 vector signed int); 5208vector unsigned int vec_msum (vector unsigned short, 5209 vector unsigned short, 5210 vector unsigned int); 5211vector signed int vec_msum (vector signed short, vector signed short, 5212 vector signed int); 5213 5214vector unsigned int vec_msums (vector unsigned short, 5215 vector unsigned short, 5216 vector unsigned int); 5217vector signed int vec_msums (vector signed short, vector signed short, 5218 vector signed int); 5219 5220void vec_mtvscr (vector signed int); 5221void vec_mtvscr (vector unsigned int); 5222void vec_mtvscr (vector signed short); 5223void vec_mtvscr (vector unsigned short); 5224void vec_mtvscr (vector signed char); 5225void vec_mtvscr (vector unsigned char); 5226 5227vector unsigned short vec_mule (vector unsigned char, 5228 vector unsigned char); 5229vector signed short vec_mule (vector signed char, vector signed char); 5230vector unsigned int vec_mule (vector unsigned short, 5231 vector unsigned short); 5232vector signed int vec_mule (vector signed short, vector signed short); 5233 5234vector unsigned short vec_mulo (vector unsigned char, 5235 vector unsigned char); 5236vector signed short vec_mulo (vector signed char, vector signed char); 5237vector unsigned int vec_mulo (vector unsigned short, 5238 vector unsigned short); 5239vector signed int vec_mulo (vector signed short, vector signed short); 5240 5241vector float vec_nmsub (vector float, vector float, vector float); 5242 5243vector float vec_nor (vector float, vector float); 5244vector signed int vec_nor (vector signed int, vector signed int); 5245vector unsigned int vec_nor (vector unsigned int, vector unsigned int); 5246vector signed short vec_nor (vector signed short, vector signed short); 5247vector unsigned short vec_nor (vector unsigned short, 5248 vector unsigned short); 5249vector signed char vec_nor (vector signed char, vector signed char); 5250vector unsigned char vec_nor (vector unsigned char, 5251 vector unsigned char); 5252 5253vector float vec_or (vector float, vector float); 5254vector float vec_or (vector float, vector signed int); 5255vector float vec_or (vector signed int, vector float); 5256vector signed int vec_or (vector signed int, vector signed int); 5257vector unsigned int vec_or (vector signed int, vector unsigned int); 5258vector unsigned int vec_or (vector unsigned int, vector signed int); 5259vector unsigned int vec_or (vector unsigned int, vector unsigned int); 5260vector signed short vec_or (vector signed short, vector signed short); 5261vector unsigned short vec_or (vector signed short, 5262 vector unsigned short); 5263vector unsigned short vec_or (vector unsigned short, 5264 vector signed short); 5265vector unsigned short vec_or (vector unsigned short, 5266 vector unsigned short); 5267vector signed char vec_or (vector signed char, vector signed char); 5268vector unsigned char vec_or (vector signed char, vector unsigned char); 5269vector unsigned char vec_or (vector unsigned char, vector signed char); 5270vector unsigned char vec_or (vector unsigned char, 5271 vector unsigned char); 5272 5273vector signed char vec_pack (vector signed short, vector signed short); 5274vector unsigned char vec_pack (vector unsigned short, 5275 vector unsigned short); 5276vector signed short vec_pack (vector signed int, vector signed int); 5277vector unsigned short vec_pack (vector unsigned int, 5278 vector unsigned int); 5279 5280vector signed short vec_packpx (vector unsigned int, 5281 vector unsigned int); 5282 5283vector unsigned char vec_packs (vector unsigned short, 5284 vector unsigned short); 5285vector signed char vec_packs (vector signed short, vector signed short); 5286 5287vector unsigned short vec_packs (vector unsigned int, 5288 vector unsigned int); 5289vector signed short vec_packs (vector signed int, vector signed int); 5290 5291vector unsigned char vec_packsu (vector unsigned short, 5292 vector unsigned short); 5293vector unsigned char vec_packsu (vector signed short, 5294 vector signed short); 5295vector unsigned short vec_packsu (vector unsigned int, 5296 vector unsigned int); 5297vector unsigned short vec_packsu (vector signed int, vector signed int); 5298 5299vector float vec_perm (vector float, vector float, 5300 vector unsigned char); 5301vector signed int vec_perm (vector signed int, vector signed int, 5302 vector unsigned char); 5303vector unsigned int vec_perm (vector unsigned int, vector unsigned int, 5304 vector unsigned char); 5305vector signed short vec_perm (vector signed short, vector signed short, 5306 vector unsigned char); 5307vector unsigned short vec_perm (vector unsigned short, 5308 vector unsigned short, 5309 vector unsigned char); 5310vector signed char vec_perm (vector signed char, vector signed char, 5311 vector unsigned char); 5312vector unsigned char vec_perm (vector unsigned char, 5313 vector unsigned char, 5314 vector unsigned char); 5315 5316vector float vec_re (vector float); 5317 5318vector signed char vec_rl (vector signed char, vector unsigned char); 5319vector unsigned char vec_rl (vector unsigned char, 5320 vector unsigned char); 5321vector signed short vec_rl (vector signed short, vector unsigned short); 5322 5323vector unsigned short vec_rl (vector unsigned short, 5324 vector unsigned short); 5325vector signed int vec_rl (vector signed int, vector unsigned int); 5326vector unsigned int vec_rl (vector unsigned int, vector unsigned int); 5327 5328vector float vec_round (vector float); 5329 5330vector float vec_rsqrte (vector float); 5331 5332vector float vec_sel (vector float, vector float, vector signed int); 5333vector float vec_sel (vector float, vector float, vector unsigned int); 5334vector signed int vec_sel (vector signed int, vector signed int, 5335 vector signed int); 5336vector signed int vec_sel (vector signed int, vector signed int, 5337 vector unsigned int); 5338vector unsigned int vec_sel (vector unsigned int, vector unsigned int, 5339 vector signed int); 5340vector unsigned int vec_sel (vector unsigned int, vector unsigned int, 5341 vector unsigned int); 5342vector signed short vec_sel (vector signed short, vector signed short, 5343 vector signed short); 5344vector signed short vec_sel (vector signed short, vector signed short, 5345 vector unsigned short); 5346vector unsigned short vec_sel (vector unsigned short, 5347 vector unsigned short, 5348 vector signed short); 5349vector unsigned short vec_sel (vector unsigned short, 5350 vector unsigned short, 5351 vector unsigned short); 5352vector signed char vec_sel (vector signed char, vector signed char, 5353 vector signed char); 5354vector signed char vec_sel (vector signed char, vector signed char, 5355 vector unsigned char); 5356vector unsigned char vec_sel (vector unsigned char, 5357 vector unsigned char, 5358 vector signed char); 5359vector unsigned char vec_sel (vector unsigned char, 5360 vector unsigned char, 5361 vector unsigned char); 5362 5363vector signed char vec_sl (vector signed char, vector unsigned char); 5364vector unsigned char vec_sl (vector unsigned char, 5365 vector unsigned char); 5366vector signed short vec_sl (vector signed short, vector unsigned short); 5367 5368vector unsigned short vec_sl (vector unsigned short, 5369 vector unsigned short); 5370vector signed int vec_sl (vector signed int, vector unsigned int); 5371vector unsigned int vec_sl (vector unsigned int, vector unsigned int); 5372 5373vector float vec_sld (vector float, vector float, const char); 5374vector signed int vec_sld (vector signed int, vector signed int, 5375 const char); 5376vector unsigned int vec_sld (vector unsigned int, vector unsigned int, 5377 const char); 5378vector signed short vec_sld (vector signed short, vector signed short, 5379 const char); 5380vector unsigned short vec_sld (vector unsigned short, 5381 vector unsigned short, const char); 5382vector signed char vec_sld (vector signed char, vector signed char, 5383 const char); 5384vector unsigned char vec_sld (vector unsigned char, 5385 vector unsigned char, 5386 const char); 5387 5388vector signed int vec_sll (vector signed int, vector unsigned int); 5389vector signed int vec_sll (vector signed int, vector unsigned short); 5390vector signed int vec_sll (vector signed int, vector unsigned char); 5391vector unsigned int vec_sll (vector unsigned int, vector unsigned int); 5392vector unsigned int vec_sll (vector unsigned int, 5393 vector unsigned short); 5394vector unsigned int vec_sll (vector unsigned int, vector unsigned char); 5395 5396vector signed short vec_sll (vector signed short, vector unsigned int); 5397vector signed short vec_sll (vector signed short, 5398 vector unsigned short); 5399vector signed short vec_sll (vector signed short, vector unsigned char); 5400 5401vector unsigned short vec_sll (vector unsigned short, 5402 vector unsigned int); 5403vector unsigned short vec_sll (vector unsigned short, 5404 vector unsigned short); 5405vector unsigned short vec_sll (vector unsigned short, 5406 vector unsigned char); 5407vector signed char vec_sll (vector signed char, vector unsigned int); 5408vector signed char vec_sll (vector signed char, vector unsigned short); 5409vector signed char vec_sll (vector signed char, vector unsigned char); 5410vector unsigned char vec_sll (vector unsigned char, 5411 vector unsigned int); 5412vector unsigned char vec_sll (vector unsigned char, 5413 vector unsigned short); 5414vector unsigned char vec_sll (vector unsigned char, 5415 vector unsigned char); 5416 5417vector float vec_slo (vector float, vector signed char); 5418vector float vec_slo (vector float, vector unsigned char); 5419vector signed int vec_slo (vector signed int, vector signed char); 5420vector signed int vec_slo (vector signed int, vector unsigned char); 5421vector unsigned int vec_slo (vector unsigned int, vector signed char); 5422vector unsigned int vec_slo (vector unsigned int, vector unsigned char); 5423 5424vector signed short vec_slo (vector signed short, vector signed char); 5425vector signed short vec_slo (vector signed short, vector unsigned char); 5426 5427vector unsigned short vec_slo (vector unsigned short, 5428 vector signed char); 5429vector unsigned short vec_slo (vector unsigned short, 5430 vector unsigned char); 5431vector signed char vec_slo (vector signed char, vector signed char); 5432vector signed char vec_slo (vector signed char, vector unsigned char); 5433vector unsigned char vec_slo (vector unsigned char, vector signed char); 5434 5435vector unsigned char vec_slo (vector unsigned char, 5436 vector unsigned char); 5437 5438vector signed char vec_splat (vector signed char, const char); 5439vector unsigned char vec_splat (vector unsigned char, const char); 5440vector signed short vec_splat (vector signed short, const char); 5441vector unsigned short vec_splat (vector unsigned short, const char); 5442vector float vec_splat (vector float, const char); 5443vector signed int vec_splat (vector signed int, const char); 5444vector unsigned int vec_splat (vector unsigned int, const char); 5445 5446vector signed char vec_splat_s8 (const char); 5447 5448vector signed short vec_splat_s16 (const char); 5449 5450vector signed int vec_splat_s32 (const char); 5451 5452vector unsigned char vec_splat_u8 (const char); 5453 5454vector unsigned short vec_splat_u16 (const char); 5455 5456vector unsigned int vec_splat_u32 (const char); 5457 5458vector signed char vec_sr (vector signed char, vector unsigned char); 5459vector unsigned char vec_sr (vector unsigned char, 5460 vector unsigned char); 5461vector signed short vec_sr (vector signed short, vector unsigned short); 5462 5463vector unsigned short vec_sr (vector unsigned short, 5464 vector unsigned short); 5465vector signed int vec_sr (vector signed int, vector unsigned int); 5466vector unsigned int vec_sr (vector unsigned int, vector unsigned int); 5467 5468vector signed char vec_sra (vector signed char, vector unsigned char); 5469vector unsigned char vec_sra (vector unsigned char, 5470 vector unsigned char); 5471vector signed short vec_sra (vector signed short, 5472 vector unsigned short); 5473vector unsigned short vec_sra (vector unsigned short, 5474 vector unsigned short); 5475vector signed int vec_sra (vector signed int, vector unsigned int); 5476vector unsigned int vec_sra (vector unsigned int, vector unsigned int); 5477 5478vector signed int vec_srl (vector signed int, vector unsigned int); 5479vector signed int vec_srl (vector signed int, vector unsigned short); 5480vector signed int vec_srl (vector signed int, vector unsigned char); 5481vector unsigned int vec_srl (vector unsigned int, vector unsigned int); 5482vector unsigned int vec_srl (vector unsigned int, 5483 vector unsigned short); 5484vector unsigned int vec_srl (vector unsigned int, vector unsigned char); 5485 5486vector signed short vec_srl (vector signed short, vector unsigned int); 5487vector signed short vec_srl (vector signed short, 5488 vector unsigned short); 5489vector signed short vec_srl (vector signed short, vector unsigned char); 5490 5491vector unsigned short vec_srl (vector unsigned short, 5492 vector unsigned int); 5493vector unsigned short vec_srl (vector unsigned short, 5494 vector unsigned short); 5495vector unsigned short vec_srl (vector unsigned short, 5496 vector unsigned char); 5497vector signed char vec_srl (vector signed char, vector unsigned int); 5498vector signed char vec_srl (vector signed char, vector unsigned short); 5499vector signed char vec_srl (vector signed char, vector unsigned char); 5500vector unsigned char vec_srl (vector unsigned char, 5501 vector unsigned int); 5502vector unsigned char vec_srl (vector unsigned char, 5503 vector unsigned short); 5504vector unsigned char vec_srl (vector unsigned char, 5505 vector unsigned char); 5506 5507vector float vec_sro (vector float, vector signed char); 5508vector float vec_sro (vector float, vector unsigned char); 5509vector signed int vec_sro (vector signed int, vector signed char); 5510vector signed int vec_sro (vector signed int, vector unsigned char); 5511vector unsigned int vec_sro (vector unsigned int, vector signed char); 5512vector unsigned int vec_sro (vector unsigned int, vector unsigned char); 5513 5514vector signed short vec_sro (vector signed short, vector signed char); 5515vector signed short vec_sro (vector signed short, vector unsigned char); 5516 5517vector unsigned short vec_sro (vector unsigned short, 5518 vector signed char); 5519vector unsigned short vec_sro (vector unsigned short, 5520 vector unsigned char); 5521vector signed char vec_sro (vector signed char, vector signed char); 5522vector signed char vec_sro (vector signed char, vector unsigned char); 5523vector unsigned char vec_sro (vector unsigned char, vector signed char); 5524 5525vector unsigned char vec_sro (vector unsigned char, 5526 vector unsigned char); 5527 5528void vec_st (vector float, int, float *); 5529void vec_st (vector float, int, vector float *); 5530void vec_st (vector signed int, int, int *); 5531void vec_st (vector signed int, int, unsigned int *); 5532void vec_st (vector unsigned int, int, unsigned int *); 5533void vec_st (vector unsigned int, int, vector unsigned int *); 5534void vec_st (vector signed short, int, short *); 5535void vec_st (vector signed short, int, vector unsigned short *); 5536void vec_st (vector signed short, int, vector signed short *); 5537void vec_st (vector unsigned short, int, unsigned short *); 5538void vec_st (vector unsigned short, int, vector unsigned short *); 5539void vec_st (vector signed char, int, signed char *); 5540void vec_st (vector signed char, int, unsigned char *); 5541void vec_st (vector signed char, int, vector signed char *); 5542void vec_st (vector unsigned char, int, unsigned char *); 5543void vec_st (vector unsigned char, int, vector unsigned char *); 5544 5545void vec_ste (vector signed char, int, unsigned char *); 5546void vec_ste (vector signed char, int, signed char *); 5547void vec_ste (vector unsigned char, int, unsigned char *); 5548void vec_ste (vector signed short, int, short *); 5549void vec_ste (vector signed short, int, unsigned short *); 5550void vec_ste (vector unsigned short, int, void *); 5551void vec_ste (vector signed int, int, unsigned int *); 5552void vec_ste (vector signed int, int, int *); 5553void vec_ste (vector unsigned int, int, unsigned int *); 5554void vec_ste (vector float, int, float *); 5555 5556void vec_stl (vector float, int, vector float *); 5557void vec_stl (vector float, int, float *); 5558void vec_stl (vector signed int, int, vector signed int *); 5559void vec_stl (vector signed int, int, int *); 5560void vec_stl (vector signed int, int, unsigned int *); 5561void vec_stl (vector unsigned int, int, vector unsigned int *); 5562void vec_stl (vector unsigned int, int, unsigned int *); 5563void vec_stl (vector signed short, int, short *); 5564void vec_stl (vector signed short, int, unsigned short *); 5565void vec_stl (vector signed short, int, vector signed short *); 5566void vec_stl (vector unsigned short, int, unsigned short *); 5567void vec_stl (vector unsigned short, int, vector signed short *); 5568void vec_stl (vector signed char, int, signed char *); 5569void vec_stl (vector signed char, int, unsigned char *); 5570void vec_stl (vector signed char, int, vector signed char *); 5571void vec_stl (vector unsigned char, int, unsigned char *); 5572void vec_stl (vector unsigned char, int, vector unsigned char *); 5573 5574vector signed char vec_sub (vector signed char, vector signed char); 5575vector unsigned char vec_sub (vector signed char, vector unsigned char); 5576 5577vector unsigned char vec_sub (vector unsigned char, vector signed char); 5578 5579vector unsigned char vec_sub (vector unsigned char, 5580 vector unsigned char); 5581vector signed short vec_sub (vector signed short, vector signed short); 5582vector unsigned short vec_sub (vector signed short, 5583 vector unsigned short); 5584vector unsigned short vec_sub (vector unsigned short, 5585 vector signed short); 5586vector unsigned short vec_sub (vector unsigned short, 5587 vector unsigned short); 5588vector signed int vec_sub (vector signed int, vector signed int); 5589vector unsigned int vec_sub (vector signed int, vector unsigned int); 5590vector unsigned int vec_sub (vector unsigned int, vector signed int); 5591vector unsigned int vec_sub (vector unsigned int, vector unsigned int); 5592vector float vec_sub (vector float, vector float); 5593 5594vector unsigned int vec_subc (vector unsigned int, vector unsigned int); 5595 5596vector unsigned char vec_subs (vector signed char, 5597 vector unsigned char); 5598vector unsigned char vec_subs (vector unsigned char, 5599 vector signed char); 5600vector unsigned char vec_subs (vector unsigned char, 5601 vector unsigned char); 5602vector signed char vec_subs (vector signed char, vector signed char); 5603vector unsigned short vec_subs (vector signed short, 5604 vector unsigned short); 5605vector unsigned short vec_subs (vector unsigned short, 5606 vector signed short); 5607vector unsigned short vec_subs (vector unsigned short, 5608 vector unsigned short); 5609vector signed short vec_subs (vector signed short, vector signed short); 5610 5611vector unsigned int vec_subs (vector signed int, vector unsigned int); 5612vector unsigned int vec_subs (vector unsigned int, vector signed int); 5613vector unsigned int vec_subs (vector unsigned int, vector unsigned int); 5614 5615vector signed int vec_subs (vector signed int, vector signed int); 5616 5617vector unsigned int vec_sum4s (vector unsigned char, 5618 vector unsigned int); 5619vector signed int vec_sum4s (vector signed char, vector signed int); 5620vector signed int vec_sum4s (vector signed short, vector signed int); 5621 5622vector signed int vec_sum2s (vector signed int, vector signed int); 5623 5624vector signed int vec_sums (vector signed int, vector signed int); 5625 5626vector float vec_trunc (vector float); 5627 5628vector signed short vec_unpackh (vector signed char); 5629vector unsigned int vec_unpackh (vector signed short); 5630vector signed int vec_unpackh (vector signed short); 5631 5632vector signed short vec_unpackl (vector signed char); 5633vector unsigned int vec_unpackl (vector signed short); 5634vector signed int vec_unpackl (vector signed short); 5635 5636vector float vec_xor (vector float, vector float); 5637vector float vec_xor (vector float, vector signed int); 5638vector float vec_xor (vector signed int, vector float); 5639vector signed int vec_xor (vector signed int, vector signed int); 5640vector unsigned int vec_xor (vector signed int, vector unsigned int); 5641vector unsigned int vec_xor (vector unsigned int, vector signed int); 5642vector unsigned int vec_xor (vector unsigned int, vector unsigned int); 5643vector signed short vec_xor (vector signed short, vector signed short); 5644vector unsigned short vec_xor (vector signed short, 5645 vector unsigned short); 5646vector unsigned short vec_xor (vector unsigned short, 5647 vector signed short); 5648vector unsigned short vec_xor (vector unsigned short, 5649 vector unsigned short); 5650vector signed char vec_xor (vector signed char, vector signed char); 5651vector unsigned char vec_xor (vector signed char, vector unsigned char); 5652 5653vector unsigned char vec_xor (vector unsigned char, vector signed char); 5654 5655vector unsigned char vec_xor (vector unsigned char, 5656 vector unsigned char); 5657 5658vector signed int vec_all_eq (vector signed char, vector unsigned char); 5659 5660vector signed int vec_all_eq (vector signed char, vector signed char); 5661vector signed int vec_all_eq (vector unsigned char, vector signed char); 5662 5663vector signed int vec_all_eq (vector unsigned char, 5664 vector unsigned char); 5665vector signed int vec_all_eq (vector signed short, 5666 vector unsigned short); 5667vector signed int vec_all_eq (vector signed short, vector signed short); 5668 5669vector signed int vec_all_eq (vector unsigned short, 5670 vector signed short); 5671vector signed int vec_all_eq (vector unsigned short, 5672 vector unsigned short); 5673vector signed int vec_all_eq (vector signed int, vector unsigned int); 5674vector signed int vec_all_eq (vector signed int, vector signed int); 5675vector signed int vec_all_eq (vector unsigned int, vector signed int); 5676vector signed int vec_all_eq (vector unsigned int, vector unsigned int); 5677 5678vector signed int vec_all_eq (vector float, vector float); 5679 5680vector signed int vec_all_ge (vector signed char, vector unsigned char); 5681 5682vector signed int vec_all_ge (vector unsigned char, vector signed char); 5683 5684vector signed int vec_all_ge (vector unsigned char, 5685 vector unsigned char); 5686vector signed int vec_all_ge (vector signed char, vector signed char); 5687vector signed int vec_all_ge (vector signed short, 5688 vector unsigned short); 5689vector signed int vec_all_ge (vector unsigned short, 5690 vector signed short); 5691vector signed int vec_all_ge (vector unsigned short, 5692 vector unsigned short); 5693vector signed int vec_all_ge (vector signed short, vector signed short); 5694 5695vector signed int vec_all_ge (vector signed int, vector unsigned int); 5696vector signed int vec_all_ge (vector unsigned int, vector signed int); 5697vector signed int vec_all_ge (vector unsigned int, vector unsigned int); 5698 5699vector signed int vec_all_ge (vector signed int, vector signed int); 5700vector signed int vec_all_ge (vector float, vector float); 5701 5702vector signed int vec_all_gt (vector signed char, vector unsigned char); 5703 5704vector signed int vec_all_gt (vector unsigned char, vector signed char); 5705 5706vector signed int vec_all_gt (vector unsigned char, 5707 vector unsigned char); 5708vector signed int vec_all_gt (vector signed char, vector signed char); 5709vector signed int vec_all_gt (vector signed short, 5710 vector unsigned short); 5711vector signed int vec_all_gt (vector unsigned short, 5712 vector signed short); 5713vector signed int vec_all_gt (vector unsigned short, 5714 vector unsigned short); 5715vector signed int vec_all_gt (vector signed short, vector signed short); 5716 5717vector signed int vec_all_gt (vector signed int, vector unsigned int); 5718vector signed int vec_all_gt (vector unsigned int, vector signed int); 5719vector signed int vec_all_gt (vector unsigned int, vector unsigned int); 5720 5721vector signed int vec_all_gt (vector signed int, vector signed int); 5722vector signed int vec_all_gt (vector float, vector float); 5723 5724vector signed int vec_all_in (vector float, vector float); 5725 5726vector signed int vec_all_le (vector signed char, vector unsigned char); 5727 5728vector signed int vec_all_le (vector unsigned char, vector signed char); 5729 5730vector signed int vec_all_le (vector unsigned char, 5731 vector unsigned char); 5732vector signed int vec_all_le (vector signed char, vector signed char); 5733vector signed int vec_all_le (vector signed short, 5734 vector unsigned short); 5735vector signed int vec_all_le (vector unsigned short, 5736 vector signed short); 5737vector signed int vec_all_le (vector unsigned short, 5738 vector unsigned short); 5739vector signed int vec_all_le (vector signed short, vector signed short); 5740 5741vector signed int vec_all_le (vector signed int, vector unsigned int); 5742vector signed int vec_all_le (vector unsigned int, vector signed int); 5743vector signed int vec_all_le (vector unsigned int, vector unsigned int); 5744 5745vector signed int vec_all_le (vector signed int, vector signed int); 5746vector signed int vec_all_le (vector float, vector float); 5747 5748vector signed int vec_all_lt (vector signed char, vector unsigned char); 5749 5750vector signed int vec_all_lt (vector unsigned char, vector signed char); 5751 5752vector signed int vec_all_lt (vector unsigned char, 5753 vector unsigned char); 5754vector signed int vec_all_lt (vector signed char, vector signed char); 5755vector signed int vec_all_lt (vector signed short, 5756 vector unsigned short); 5757vector signed int vec_all_lt (vector unsigned short, 5758 vector signed short); 5759vector signed int vec_all_lt (vector unsigned short, 5760 vector unsigned short); 5761vector signed int vec_all_lt (vector signed short, vector signed short); 5762 5763vector signed int vec_all_lt (vector signed int, vector unsigned int); 5764vector signed int vec_all_lt (vector unsigned int, vector signed int); 5765vector signed int vec_all_lt (vector unsigned int, vector unsigned int); 5766 5767vector signed int vec_all_lt (vector signed int, vector signed int); 5768vector signed int vec_all_lt (vector float, vector float); 5769 5770vector signed int vec_all_nan (vector float); 5771 5772vector signed int vec_all_ne (vector signed char, vector unsigned char); 5773 5774vector signed int vec_all_ne (vector signed char, vector signed char); 5775vector signed int vec_all_ne (vector unsigned char, vector signed char); 5776 5777vector signed int vec_all_ne (vector unsigned char, 5778 vector unsigned char); 5779vector signed int vec_all_ne (vector signed short, 5780 vector unsigned short); 5781vector signed int vec_all_ne (vector signed short, vector signed short); 5782 5783vector signed int vec_all_ne (vector unsigned short, 5784 vector signed short); 5785vector signed int vec_all_ne (vector unsigned short, 5786 vector unsigned short); 5787vector signed int vec_all_ne (vector signed int, vector unsigned int); 5788vector signed int vec_all_ne (vector signed int, vector signed int); 5789vector signed int vec_all_ne (vector unsigned int, vector signed int); 5790vector signed int vec_all_ne (vector unsigned int, vector unsigned int); 5791 5792vector signed int vec_all_ne (vector float, vector float); 5793 5794vector signed int vec_all_nge (vector float, vector float); 5795 5796vector signed int vec_all_ngt (vector float, vector float); 5797 5798vector signed int vec_all_nle (vector float, vector float); 5799 5800vector signed int vec_all_nlt (vector float, vector float); 5801 5802vector signed int vec_all_numeric (vector float); 5803 5804vector signed int vec_any_eq (vector signed char, vector unsigned char); 5805 5806vector signed int vec_any_eq (vector signed char, vector signed char); 5807vector signed int vec_any_eq (vector unsigned char, vector signed char); 5808 5809vector signed int vec_any_eq (vector unsigned char, 5810 vector unsigned char); 5811vector signed int vec_any_eq (vector signed short, 5812 vector unsigned short); 5813vector signed int vec_any_eq (vector signed short, vector signed short); 5814 5815vector signed int vec_any_eq (vector unsigned short, 5816 vector signed short); 5817vector signed int vec_any_eq (vector unsigned short, 5818 vector unsigned short); 5819vector signed int vec_any_eq (vector signed int, vector unsigned int); 5820vector signed int vec_any_eq (vector signed int, vector signed int); 5821vector signed int vec_any_eq (vector unsigned int, vector signed int); 5822vector signed int vec_any_eq (vector unsigned int, vector unsigned int); 5823 5824vector signed int vec_any_eq (vector float, vector float); 5825 5826vector signed int vec_any_ge (vector signed char, vector unsigned char); 5827 5828vector signed int vec_any_ge (vector unsigned char, vector signed char); 5829 5830vector signed int vec_any_ge (vector unsigned char, 5831 vector unsigned char); 5832vector signed int vec_any_ge (vector signed char, vector signed char); 5833vector signed int vec_any_ge (vector signed short, 5834 vector unsigned short); 5835vector signed int vec_any_ge (vector unsigned short, 5836 vector signed short); 5837vector signed int vec_any_ge (vector unsigned short, 5838 vector unsigned short); 5839vector signed int vec_any_ge (vector signed short, vector signed short); 5840 5841vector signed int vec_any_ge (vector signed int, vector unsigned int); 5842vector signed int vec_any_ge (vector unsigned int, vector signed int); 5843vector signed int vec_any_ge (vector unsigned int, vector unsigned int); 5844 5845vector signed int vec_any_ge (vector signed int, vector signed int); 5846vector signed int vec_any_ge (vector float, vector float); 5847 5848vector signed int vec_any_gt (vector signed char, vector unsigned char); 5849 5850vector signed int vec_any_gt (vector unsigned char, vector signed char); 5851 5852vector signed int vec_any_gt (vector unsigned char, 5853 vector unsigned char); 5854vector signed int vec_any_gt (vector signed char, vector signed char); 5855vector signed int vec_any_gt (vector signed short, 5856 vector unsigned short); 5857vector signed int vec_any_gt (vector unsigned short, 5858 vector signed short); 5859vector signed int vec_any_gt (vector unsigned short, 5860 vector unsigned short); 5861vector signed int vec_any_gt (vector signed short, vector signed short); 5862 5863vector signed int vec_any_gt (vector signed int, vector unsigned int); 5864vector signed int vec_any_gt (vector unsigned int, vector signed int); 5865vector signed int vec_any_gt (vector unsigned int, vector unsigned int); 5866 5867vector signed int vec_any_gt (vector signed int, vector signed int); 5868vector signed int vec_any_gt (vector float, vector float); 5869 5870vector signed int vec_any_le (vector signed char, vector unsigned char); 5871 5872vector signed int vec_any_le (vector unsigned char, vector signed char); 5873 5874vector signed int vec_any_le (vector unsigned char, 5875 vector unsigned char); 5876vector signed int vec_any_le (vector signed char, vector signed char); 5877vector signed int vec_any_le (vector signed short, 5878 vector unsigned short); 5879vector signed int vec_any_le (vector unsigned short, 5880 vector signed short); 5881vector signed int vec_any_le (vector unsigned short, 5882 vector unsigned short); 5883vector signed int vec_any_le (vector signed short, vector signed short); 5884 5885vector signed int vec_any_le (vector signed int, vector unsigned int); 5886vector signed int vec_any_le (vector unsigned int, vector signed int); 5887vector signed int vec_any_le (vector unsigned int, vector unsigned int); 5888 5889vector signed int vec_any_le (vector signed int, vector signed int); 5890vector signed int vec_any_le (vector float, vector float); 5891 5892vector signed int vec_any_lt (vector signed char, vector unsigned char); 5893 5894vector signed int vec_any_lt (vector unsigned char, vector signed char); 5895 5896vector signed int vec_any_lt (vector unsigned char, 5897 vector unsigned char); 5898vector signed int vec_any_lt (vector signed char, vector signed char); 5899vector signed int vec_any_lt (vector signed short, 5900 vector unsigned short); 5901vector signed int vec_any_lt (vector unsigned short, 5902 vector signed short); 5903vector signed int vec_any_lt (vector unsigned short, 5904 vector unsigned short); 5905vector signed int vec_any_lt (vector signed short, vector signed short); 5906 5907vector signed int vec_any_lt (vector signed int, vector unsigned int); 5908vector signed int vec_any_lt (vector unsigned int, vector signed int); 5909vector signed int vec_any_lt (vector unsigned int, vector unsigned int); 5910 5911vector signed int vec_any_lt (vector signed int, vector signed int); 5912vector signed int vec_any_lt (vector float, vector float); 5913 5914vector signed int vec_any_nan (vector float); 5915 5916vector signed int vec_any_ne (vector signed char, vector unsigned char); 5917 5918vector signed int vec_any_ne (vector signed char, vector signed char); 5919vector signed int vec_any_ne (vector unsigned char, vector signed char); 5920 5921vector signed int vec_any_ne (vector unsigned char, 5922 vector unsigned char); 5923vector signed int vec_any_ne (vector signed short, 5924 vector unsigned short); 5925vector signed int vec_any_ne (vector signed short, vector signed short); 5926 5927vector signed int vec_any_ne (vector unsigned short, 5928 vector signed short); 5929vector signed int vec_any_ne (vector unsigned short, 5930 vector unsigned short); 5931vector signed int vec_any_ne (vector signed int, vector unsigned int); 5932vector signed int vec_any_ne (vector signed int, vector signed int); 5933vector signed int vec_any_ne (vector unsigned int, vector signed int); 5934vector signed int vec_any_ne (vector unsigned int, vector unsigned int); 5935 5936vector signed int vec_any_ne (vector float, vector float); 5937 5938vector signed int vec_any_nge (vector float, vector float); 5939 5940vector signed int vec_any_ngt (vector float, vector float); 5941 5942vector signed int vec_any_nle (vector float, vector float); 5943 5944vector signed int vec_any_nlt (vector float, vector float); 5945 5946vector signed int vec_any_numeric (vector float); 5947 5948vector signed int vec_any_out (vector float, vector float); 5949@end smallexample 5950 5951@node Pragmas 5952@section Pragmas Accepted by GCC 5953@cindex pragmas 5954@cindex #pragma 5955 5956GCC supports several types of pragmas, primarily in order to compile 5957code originally written for other compilers. Note that in general 5958we do not recommend the use of pragmas; @xref{Function Attributes}, 5959for further explanation. 5960 5961@menu 5962* ARM Pragmas:: 5963* Darwin Pragmas:: 5964* Solaris Pragmas:: 5965* Tru64 Pragmas:: 5966@end menu 5967 5968@node ARM Pragmas 5969@subsection ARM Pragmas 5970 5971The ARM target defines pragmas for controlling the default addition of 5972@code{long_call} and @code{short_call} attributes to functions. 5973@xref{Function Attributes}, for information about the effects of these 5974attributes. 5975 5976@table @code 5977@item long_calls 5978@cindex pragma, long_calls 5979Set all subsequent functions to have the @code{long_call} attribute. 5980 5981@item no_long_calls 5982@cindex pragma, no_long_calls 5983Set all subsequent functions to have the @code{short_call} attribute. 5984 5985@item long_calls_off 5986@cindex pragma, long_calls_off 5987Do not affect the @code{long_call} or @code{short_call} attributes of 5988subsequent functions. 5989@end table 5990 5991@c Describe c4x pragmas here. 5992@c Describe h8300 pragmas here. 5993@c Describe i370 pragmas here. 5994@c Describe i960 pragmas here. 5995@c Describe sh pragmas here. 5996@c Describe v850 pragmas here. 5997 5998@node Darwin Pragmas 5999@subsection Darwin Pragmas 6000 6001The following pragmas are available for all architectures running the 6002Darwin operating system. These are useful for compatibility with other 6003MacOS compilers. 6004 6005@table @code 6006@item mark @var{tokens}@dots{} 6007@cindex pragma, mark 6008This pragma is accepted, but has no effect. 6009 6010@item options align=@var{alignment} 6011@cindex pragma, options align 6012This pragma sets the alignment of fields in structures. The values of 6013@var{alignment} may be @code{mac68k}, to emulate m68k alignment, or 6014@code{power}, to emulate PowerPC alignment. Uses of this pragma nest 6015properly; to restore the previous setting, use @code{reset} for the 6016@var{alignment}. 6017 6018@item segment @var{tokens}@dots{} 6019@cindex pragma, segment 6020This pragma is accepted, but has no effect. 6021 6022@item unused (@var{var} [, @var{var}]@dots{}) 6023@cindex pragma, unused 6024This pragma declares variables to be possibly unused. GCC will not 6025produce warnings for the listed variables. The effect is similar to 6026that of the @code{unused} attribute, except that this pragma may appear 6027anywhere within the variables' scopes. 6028@end table 6029 6030@node Solaris Pragmas 6031@subsection Solaris Pragmas 6032 6033For compatibility with the SunPRO compiler, the following pragma 6034is supported. 6035 6036@table @code 6037@item redefine_extname @var{oldname} @var{newname} 6038@cindex pragma, redefine_extname 6039 6040This pragma gives the C function @var{oldname} the assembler label 6041@var{newname}. The pragma must appear before the function declaration. 6042This pragma is equivalent to the asm labels extension (@pxref{Asm 6043Labels}). The preprocessor defines @code{__PRAGMA_REDEFINE_EXTNAME} 6044if the pragma is available. 6045@end table 6046 6047@node Tru64 Pragmas 6048@subsection Tru64 Pragmas 6049 6050For compatibility with the Compaq C compiler, the following pragma 6051is supported. 6052 6053@table @code 6054@item extern_prefix @var{string} 6055@cindex pragma, extern_prefix 6056 6057This pragma renames all subsequent function and variable declarations 6058such that @var{string} is prepended to the name. This effect may be 6059terminated by using another @code{extern_prefix} pragma with the 6060empty string. 6061 6062This pragma is similar in intent to to the asm labels extension 6063(@pxref{Asm Labels}) in that the system programmer wants to change 6064the assembly-level ABI without changing the source-level API. The 6065preprocessor defines @code{__EXTERN_PREFIX} if the pragma is available. 6066@end table 6067 6068@node Unnamed Fields 6069@section Unnamed struct/union fields within structs/unions. 6070@cindex struct 6071@cindex union 6072 6073For compatibility with other compilers, GCC allows you to define 6074a structure or union that contains, as fields, structures and unions 6075without names. For example: 6076 6077@example 6078struct @{ 6079 int a; 6080 union @{ 6081 int b; 6082 float c; 6083 @}; 6084 int d; 6085@} foo; 6086@end example 6087 6088In this example, the user would be able to access members of the unnamed 6089union with code like @samp{foo.b}. Note that only unnamed structs and 6090unions are allowed, you may not have, for example, an unnamed 6091@code{int}. 6092 6093You must never create such structures that cause ambiguous field definitions. 6094For example, this structure: 6095 6096@example 6097struct @{ 6098 int a; 6099 struct @{ 6100 int a; 6101 @}; 6102@} foo; 6103@end example 6104 6105It is ambiguous which @code{a} is being referred to with @samp{foo.a}. 6106Such constructs are not supported and must be avoided. In the future, 6107such constructs may be detected and treated as compilation errors. 6108 6109@node C++ Extensions 6110@chapter Extensions to the C++ Language 6111@cindex extensions, C++ language 6112@cindex C++ language extensions 6113 6114The GNU compiler provides these extensions to the C++ language (and you 6115can also use most of the C language extensions in your C++ programs). If you 6116want to write code that checks whether these features are available, you can 6117test for the GNU compiler the same way as for C programs: check for a 6118predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to 6119test specifically for GNU C++ (@pxref{Standard Predefined,,Standard 6120Predefined Macros,cpp.info,The C Preprocessor}). 6121 6122@menu 6123* Min and Max:: C++ Minimum and maximum operators. 6124* Volatiles:: What constitutes an access to a volatile object. 6125* Restricted Pointers:: C99 restricted pointers and references. 6126* Vague Linkage:: Where G++ puts inlines, vtables and such. 6127* C++ Interface:: You can use a single C++ header file for both 6128 declarations and definitions. 6129* Template Instantiation:: Methods for ensuring that exactly one copy of 6130 each needed template instantiation is emitted. 6131* Bound member functions:: You can extract a function pointer to the 6132 method denoted by a @samp{->*} or @samp{.*} expression. 6133* C++ Attributes:: Variable, function, and type attributes for C++ only. 6134* Java Exceptions:: Tweaking exception handling to work with Java. 6135* Deprecated Features:: Things might disappear from g++. 6136* Backwards Compatibility:: Compatibilities with earlier definitions of C++. 6137@end menu 6138 6139@node Min and Max 6140@section Minimum and Maximum Operators in C++ 6141 6142It is very convenient to have operators which return the ``minimum'' or the 6143``maximum'' of two arguments. In GNU C++ (but not in GNU C), 6144 6145@table @code 6146@item @var{a} <? @var{b} 6147@findex <? 6148@cindex minimum operator 6149is the @dfn{minimum}, returning the smaller of the numeric values 6150@var{a} and @var{b}; 6151 6152@item @var{a} >? @var{b} 6153@findex >? 6154@cindex maximum operator 6155is the @dfn{maximum}, returning the larger of the numeric values @var{a} 6156and @var{b}. 6157@end table 6158 6159These operations are not primitive in ordinary C++, since you can 6160use a macro to return the minimum of two things in C++, as in the 6161following example. 6162 6163@example 6164#define MIN(X,Y) ((X) < (Y) ? : (X) : (Y)) 6165@end example 6166 6167@noindent 6168You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to 6169the minimum value of variables @var{i} and @var{j}. 6170 6171However, side effects in @code{X} or @code{Y} may cause unintended 6172behavior. For example, @code{MIN (i++, j++)} will fail, incrementing 6173the smaller counter twice. A GNU C extension allows you to write safe 6174macros that avoid this kind of problem (@pxref{Naming Types,,Naming an 6175Expression's Type}). However, writing @code{MIN} and @code{MAX} as 6176macros also forces you to use function-call notation for a 6177fundamental arithmetic operation. Using GNU C++ extensions, you can 6178write @w{@samp{int min = i <? j;}} instead. 6179 6180Since @code{<?} and @code{>?} are built into the compiler, they properly 6181handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}} 6182works correctly. 6183 6184@node Volatiles 6185@section When is a Volatile Object Accessed? 6186@cindex accessing volatiles 6187@cindex volatile read 6188@cindex volatile write 6189@cindex volatile access 6190 6191Both the C and C++ standard have the concept of volatile objects. These 6192are normally accessed by pointers and used for accessing hardware. The 6193standards encourage compilers to refrain from optimizations 6194concerning accesses to volatile objects that it might perform on 6195non-volatile objects. The C standard leaves it implementation defined 6196as to what constitutes a volatile access. The C++ standard omits to 6197specify this, except to say that C++ should behave in a similar manner 6198to C with respect to volatiles, where possible. The minimum either 6199standard specifies is that at a sequence point all previous accesses to 6200volatile objects have stabilized and no subsequent accesses have 6201occurred. Thus an implementation is free to reorder and combine 6202volatile accesses which occur between sequence points, but cannot do so 6203for accesses across a sequence point. The use of volatiles does not 6204allow you to violate the restriction on updating objects multiple times 6205within a sequence point. 6206 6207In most expressions, it is intuitively obvious what is a read and what is 6208a write. For instance 6209 6210@example 6211volatile int *dst = @var{somevalue}; 6212volatile int *src = @var{someothervalue}; 6213*dst = *src; 6214@end example 6215 6216@noindent 6217will cause a read of the volatile object pointed to by @var{src} and stores the 6218value into the volatile object pointed to by @var{dst}. There is no 6219guarantee that these reads and writes are atomic, especially for objects 6220larger than @code{int}. 6221 6222Less obvious expressions are where something which looks like an access 6223is used in a void context. An example would be, 6224 6225@example 6226volatile int *src = @var{somevalue}; 6227*src; 6228@end example 6229 6230With C, such expressions are rvalues, and as rvalues cause a read of 6231the object, GCC interprets this as a read of the volatile being pointed 6232to. The C++ standard specifies that such expressions do not undergo 6233lvalue to rvalue conversion, and that the type of the dereferenced 6234object may be incomplete. The C++ standard does not specify explicitly 6235that it is this lvalue to rvalue conversion which is responsible for 6236causing an access. However, there is reason to believe that it is, 6237because otherwise certain simple expressions become undefined. However, 6238because it would surprise most programmers, G++ treats dereferencing a 6239pointer to volatile object of complete type in a void context as a read 6240of the object. When the object has incomplete type, G++ issues a 6241warning. 6242 6243@example 6244struct S; 6245struct T @{int m;@}; 6246volatile S *ptr1 = @var{somevalue}; 6247volatile T *ptr2 = @var{somevalue}; 6248*ptr1; 6249*ptr2; 6250@end example 6251 6252In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2} 6253causes a read of the object pointed to. If you wish to force an error on 6254the first case, you must force a conversion to rvalue with, for instance 6255a static cast, @code{static_cast<S>(*ptr1)}. 6256 6257When using a reference to volatile, G++ does not treat equivalent 6258expressions as accesses to volatiles, but instead issues a warning that 6259no volatile is accessed. The rationale for this is that otherwise it 6260becomes difficult to determine where volatile access occur, and not 6261possible to ignore the return value from functions returning volatile 6262references. Again, if you wish to force a read, cast the reference to 6263an rvalue. 6264 6265@node Restricted Pointers 6266@section Restricting Pointer Aliasing 6267@cindex restricted pointers 6268@cindex restricted references 6269@cindex restricted this pointer 6270 6271As with gcc, g++ understands the C99 feature of restricted pointers, 6272specified with the @code{__restrict__}, or @code{__restrict} type 6273qualifier. Because you cannot compile C++ by specifying the @option{-std=c99} 6274language flag, @code{restrict} is not a keyword in C++. 6275 6276In addition to allowing restricted pointers, you can specify restricted 6277references, which indicate that the reference is not aliased in the local 6278context. 6279 6280@example 6281void fn (int *__restrict__ rptr, int &__restrict__ rref) 6282@{ 6283 @dots{} 6284@} 6285@end example 6286 6287@noindent 6288In the body of @code{fn}, @var{rptr} points to an unaliased integer and 6289@var{rref} refers to a (different) unaliased integer. 6290 6291You may also specify whether a member function's @var{this} pointer is 6292unaliased by using @code{__restrict__} as a member function qualifier. 6293 6294@example 6295void T::fn () __restrict__ 6296@{ 6297 @dots{} 6298@} 6299@end example 6300 6301@noindent 6302Within the body of @code{T::fn}, @var{this} will have the effective 6303definition @code{T *__restrict__ const this}. Notice that the 6304interpretation of a @code{__restrict__} member function qualifier is 6305different to that of @code{const} or @code{volatile} qualifier, in that it 6306is applied to the pointer rather than the object. This is consistent with 6307other compilers which implement restricted pointers. 6308 6309As with all outermost parameter qualifiers, @code{__restrict__} is 6310ignored in function definition matching. This means you only need to 6311specify @code{__restrict__} in a function definition, rather than 6312in a function prototype as well. 6313 6314@node Vague Linkage 6315@section Vague Linkage 6316@cindex vague linkage 6317 6318There are several constructs in C++ which require space in the object 6319file but are not clearly tied to a single translation unit. We say that 6320these constructs have ``vague linkage''. Typically such constructs are 6321emitted wherever they are needed, though sometimes we can be more 6322clever. 6323 6324@table @asis 6325@item Inline Functions 6326Inline functions are typically defined in a header file which can be 6327included in many different compilations. Hopefully they can usually be 6328inlined, but sometimes an out-of-line copy is necessary, if the address 6329of the function is taken or if inlining fails. In general, we emit an 6330out-of-line copy in all translation units where one is needed. As an 6331exception, we only emit inline virtual functions with the vtable, since 6332it will always require a copy. 6333 6334Local static variables and string constants used in an inline function 6335are also considered to have vague linkage, since they must be shared 6336between all inlined and out-of-line instances of the function. 6337 6338@item VTables 6339@cindex vtable 6340C++ virtual functions are implemented in most compilers using a lookup 6341table, known as a vtable. The vtable contains pointers to the virtual 6342functions provided by a class, and each object of the class contains a 6343pointer to its vtable (or vtables, in some multiple-inheritance 6344situations). If the class declares any non-inline, non-pure virtual 6345functions, the first one is chosen as the ``key method'' for the class, 6346and the vtable is only emitted in the translation unit where the key 6347method is defined. 6348 6349@emph{Note:} If the chosen key method is later defined as inline, the 6350vtable will still be emitted in every translation unit which defines it. 6351Make sure that any inline virtuals are declared inline in the class 6352body, even if they are not defined there. 6353 6354@item type_info objects 6355@cindex type_info 6356@cindex RTTI 6357C++ requires information about types to be written out in order to 6358implement @samp{dynamic_cast}, @samp{typeid} and exception handling. 6359For polymorphic classes (classes with virtual functions), the type_info 6360object is written out along with the vtable so that @samp{dynamic_cast} 6361can determine the dynamic type of a class object at runtime. For all 6362other types, we write out the type_info object when it is used: when 6363applying @samp{typeid} to an expression, throwing an object, or 6364referring to a type in a catch clause or exception specification. 6365 6366@item Template Instantiations 6367Most everything in this section also applies to template instantiations, 6368but there are other options as well. 6369@xref{Template Instantiation,,Where's the Template?}. 6370 6371@end table 6372 6373When used with GNU ld version 2.8 or later on an ELF system such as 6374Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of 6375these constructs will be discarded at link time. This is known as 6376COMDAT support. 6377 6378On targets that don't support COMDAT, but do support weak symbols, GCC 6379will use them. This way one copy will override all the others, but 6380the unused copies will still take up space in the executable. 6381 6382For targets which do not support either COMDAT or weak symbols, 6383most entities with vague linkage will be emitted as local symbols to 6384avoid duplicate definition errors from the linker. This will not happen 6385for local statics in inlines, however, as having multiple copies will 6386almost certainly break things. 6387 6388@xref{C++ Interface,,Declarations and Definitions in One Header}, for 6389another way to control placement of these constructs. 6390 6391@node C++ Interface 6392@section Declarations and Definitions in One Header 6393 6394@cindex interface and implementation headers, C++ 6395@cindex C++ interface and implementation headers 6396C++ object definitions can be quite complex. In principle, your source 6397code will need two kinds of things for each object that you use across 6398more than one source file. First, you need an @dfn{interface} 6399specification, describing its structure with type declarations and 6400function prototypes. Second, you need the @dfn{implementation} itself. 6401It can be tedious to maintain a separate interface description in a 6402header file, in parallel to the actual implementation. It is also 6403dangerous, since separate interface and implementation definitions may 6404not remain parallel. 6405 6406@cindex pragmas, interface and implementation 6407With GNU C++, you can use a single header file for both purposes. 6408 6409@quotation 6410@emph{Warning:} The mechanism to specify this is in transition. For the 6411nonce, you must use one of two @code{#pragma} commands; in a future 6412release of GNU C++, an alternative mechanism will make these 6413@code{#pragma} commands unnecessary. 6414@end quotation 6415 6416The header file contains the full definitions, but is marked with 6417@samp{#pragma interface} in the source code. This allows the compiler 6418to use the header file only as an interface specification when ordinary 6419source files incorporate it with @code{#include}. In the single source 6420file where the full implementation belongs, you can use either a naming 6421convention or @samp{#pragma implementation} to indicate this alternate 6422use of the header file. 6423 6424@table @code 6425@item #pragma interface 6426@itemx #pragma interface "@var{subdir}/@var{objects}.h" 6427@kindex #pragma interface 6428Use this directive in @emph{header files} that define object classes, to save 6429space in most of the object files that use those classes. Normally, 6430local copies of certain information (backup copies of inline member 6431functions, debugging information, and the internal tables that implement 6432virtual functions) must be kept in each object file that includes class 6433definitions. You can use this pragma to avoid such duplication. When a 6434header file containing @samp{#pragma interface} is included in a 6435compilation, this auxiliary information will not be generated (unless 6436the main input source file itself uses @samp{#pragma implementation}). 6437Instead, the object files will contain references to be resolved at link 6438time. 6439 6440The second form of this directive is useful for the case where you have 6441multiple headers with the same name in different directories. If you 6442use this form, you must specify the same string to @samp{#pragma 6443implementation}. 6444 6445@item #pragma implementation 6446@itemx #pragma implementation "@var{objects}.h" 6447@kindex #pragma implementation 6448Use this pragma in a @emph{main input file}, when you want full output from 6449included header files to be generated (and made globally visible). The 6450included header file, in turn, should use @samp{#pragma interface}. 6451Backup copies of inline member functions, debugging information, and the 6452internal tables used to implement virtual functions are all generated in 6453implementation files. 6454 6455@cindex implied @code{#pragma implementation} 6456@cindex @code{#pragma implementation}, implied 6457@cindex naming convention, implementation headers 6458If you use @samp{#pragma implementation} with no argument, it applies to 6459an include file with the same basename@footnote{A file's @dfn{basename} 6460was the name stripped of all leading path information and of trailing 6461suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source 6462file. For example, in @file{allclass.cc}, giving just 6463@samp{#pragma implementation} 6464by itself is equivalent to @samp{#pragma implementation "allclass.h"}. 6465 6466In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as 6467an implementation file whenever you would include it from 6468@file{allclass.cc} even if you never specified @samp{#pragma 6469implementation}. This was deemed to be more trouble than it was worth, 6470however, and disabled. 6471 6472If you use an explicit @samp{#pragma implementation}, it must appear in 6473your source file @emph{before} you include the affected header files. 6474 6475Use the string argument if you want a single implementation file to 6476include code from multiple header files. (You must also use 6477@samp{#include} to include the header file; @samp{#pragma 6478implementation} only specifies how to use the file---it doesn't actually 6479include it.) 6480 6481There is no way to split up the contents of a single header file into 6482multiple implementation files. 6483@end table 6484 6485@cindex inlining and C++ pragmas 6486@cindex C++ pragmas, effect on inlining 6487@cindex pragmas in C++, effect on inlining 6488@samp{#pragma implementation} and @samp{#pragma interface} also have an 6489effect on function inlining. 6490 6491If you define a class in a header file marked with @samp{#pragma 6492interface}, the effect on a function defined in that class is similar to 6493an explicit @code{extern} declaration---the compiler emits no code at 6494all to define an independent version of the function. Its definition 6495is used only for inlining with its callers. 6496 6497@opindex fno-implement-inlines 6498Conversely, when you include the same header file in a main source file 6499that declares it as @samp{#pragma implementation}, the compiler emits 6500code for the function itself; this defines a version of the function 6501that can be found via pointers (or by callers compiled without 6502inlining). If all calls to the function can be inlined, you can avoid 6503emitting the function by compiling with @option{-fno-implement-inlines}. 6504If any calls were not inlined, you will get linker errors. 6505 6506@node Template Instantiation 6507@section Where's the Template? 6508 6509@cindex template instantiation 6510 6511C++ templates are the first language feature to require more 6512intelligence from the environment than one usually finds on a UNIX 6513system. Somehow the compiler and linker have to make sure that each 6514template instance occurs exactly once in the executable if it is needed, 6515and not at all otherwise. There are two basic approaches to this 6516problem, which I will refer to as the Borland model and the Cfront model. 6517 6518@table @asis 6519@item Borland model 6520Borland C++ solved the template instantiation problem by adding the code 6521equivalent of common blocks to their linker; the compiler emits template 6522instances in each translation unit that uses them, and the linker 6523collapses them together. The advantage of this model is that the linker 6524only has to consider the object files themselves; there is no external 6525complexity to worry about. This disadvantage is that compilation time 6526is increased because the template code is being compiled repeatedly. 6527Code written for this model tends to include definitions of all 6528templates in the header file, since they must be seen to be 6529instantiated. 6530 6531@item Cfront model 6532The AT&T C++ translator, Cfront, solved the template instantiation 6533problem by creating the notion of a template repository, an 6534automatically maintained place where template instances are stored. A 6535more modern version of the repository works as follows: As individual 6536object files are built, the compiler places any template definitions and 6537instantiations encountered in the repository. At link time, the link 6538wrapper adds in the objects in the repository and compiles any needed 6539instances that were not previously emitted. The advantages of this 6540model are more optimal compilation speed and the ability to use the 6541system linker; to implement the Borland model a compiler vendor also 6542needs to replace the linker. The disadvantages are vastly increased 6543complexity, and thus potential for error; for some code this can be 6544just as transparent, but in practice it can been very difficult to build 6545multiple programs in one directory and one program in multiple 6546directories. Code written for this model tends to separate definitions 6547of non-inline member templates into a separate file, which should be 6548compiled separately. 6549@end table 6550 6551When used with GNU ld version 2.8 or later on an ELF system such as 6552Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the 6553Borland model. On other systems, g++ implements neither automatic 6554model. 6555 6556A future version of g++ will support a hybrid model whereby the compiler 6557will emit any instantiations for which the template definition is 6558included in the compile, and store template definitions and 6559instantiation context information into the object file for the rest. 6560The link wrapper will extract that information as necessary and invoke 6561the compiler to produce the remaining instantiations. The linker will 6562then combine duplicate instantiations. 6563 6564In the mean time, you have the following options for dealing with 6565template instantiations: 6566 6567@enumerate 6568@item 6569@opindex frepo 6570Compile your template-using code with @option{-frepo}. The compiler will 6571generate files with the extension @samp{.rpo} listing all of the 6572template instantiations used in the corresponding object files which 6573could be instantiated there; the link wrapper, @samp{collect2}, will 6574then update the @samp{.rpo} files to tell the compiler where to place 6575those instantiations and rebuild any affected object files. The 6576link-time overhead is negligible after the first pass, as the compiler 6577will continue to place the instantiations in the same files. 6578 6579This is your best option for application code written for the Borland 6580model, as it will just work. Code written for the Cfront model will 6581need to be modified so that the template definitions are available at 6582one or more points of instantiation; usually this is as simple as adding 6583@code{#include <tmethods.cc>} to the end of each template header. 6584 6585For library code, if you want the library to provide all of the template 6586instantiations it needs, just try to link all of its object files 6587together; the link will fail, but cause the instantiations to be 6588generated as a side effect. Be warned, however, that this may cause 6589conflicts if multiple libraries try to provide the same instantiations. 6590For greater control, use explicit instantiation as described in the next 6591option. 6592 6593@item 6594@opindex fno-implicit-templates 6595Compile your code with @option{-fno-implicit-templates} to disable the 6596implicit generation of template instances, and explicitly instantiate 6597all the ones you use. This approach requires more knowledge of exactly 6598which instances you need than do the others, but it's less 6599mysterious and allows greater control. You can scatter the explicit 6600instantiations throughout your program, perhaps putting them in the 6601translation units where the instances are used or the translation units 6602that define the templates themselves; you can put all of the explicit 6603instantiations you need into one big file; or you can create small files 6604like 6605 6606@example 6607#include "Foo.h" 6608#include "Foo.cc" 6609 6610template class Foo<int>; 6611template ostream& operator << 6612 (ostream&, const Foo<int>&); 6613@end example 6614 6615for each of the instances you need, and create a template instantiation 6616library from those. 6617 6618If you are using Cfront-model code, you can probably get away with not 6619using @option{-fno-implicit-templates} when compiling files that don't 6620@samp{#include} the member template definitions. 6621 6622If you use one big file to do the instantiations, you may want to 6623compile it without @option{-fno-implicit-templates} so you get all of the 6624instances required by your explicit instantiations (but not by any 6625other files) without having to specify them as well. 6626 6627g++ has extended the template instantiation syntax outlined in the 6628Working Paper to allow forward declaration of explicit instantiations 6629(with @code{extern}), instantiation of the compiler support data for a 6630template class (i.e.@: the vtable) without instantiating any of its 6631members (with @code{inline}), and instantiation of only the static data 6632members of a template class, without the support data or member 6633functions (with (@code{static}): 6634 6635@example 6636extern template int max (int, int); 6637inline template class Foo<int>; 6638static template class Foo<int>; 6639@end example 6640 6641@item 6642Do nothing. Pretend g++ does implement automatic instantiation 6643management. Code written for the Borland model will work fine, but 6644each translation unit will contain instances of each of the templates it 6645uses. In a large program, this can lead to an unacceptable amount of code 6646duplication. 6647 6648@item 6649@opindex fexternal-templates 6650Add @samp{#pragma interface} to all files containing template 6651definitions. For each of these files, add @samp{#pragma implementation 6652"@var{filename}"} to the top of some @samp{.C} file which 6653@samp{#include}s it. Then compile everything with 6654@option{-fexternal-templates}. The templates will then only be expanded 6655in the translation unit which implements them (i.e.@: has a @samp{#pragma 6656implementation} line for the file where they live); all other files will 6657use external references. If you're lucky, everything should work 6658properly. If you get undefined symbol errors, you need to make sure 6659that each template instance which is used in the program is used in the 6660file which implements that template. If you don't have any use for a 6661particular instance in that file, you can just instantiate it 6662explicitly, using the syntax from the latest C++ working paper: 6663 6664@example 6665template class A<int>; 6666template ostream& operator << (ostream&, const A<int>&); 6667@end example 6668 6669This strategy will work with code written for either model. If you are 6670using code written for the Cfront model, the file containing a class 6671template and the file containing its member templates should be 6672implemented in the same translation unit. 6673 6674@item 6675@opindex falt-external-templates 6676A slight variation on this approach is to use the flag 6677@option{-falt-external-templates} instead. This flag causes template 6678instances to be emitted in the translation unit that implements the 6679header where they are first instantiated, rather than the one which 6680implements the file where the templates are defined. This header must 6681be the same in all translation units, or things are likely to break. 6682 6683@xref{C++ Interface,,Declarations and Definitions in One Header}, for 6684more discussion of these pragmas. 6685@end enumerate 6686 6687@node Bound member functions 6688@section Extracting the function pointer from a bound pointer to member function 6689 6690@cindex pmf 6691@cindex pointer to member function 6692@cindex bound pointer to member function 6693 6694In C++, pointer to member functions (PMFs) are implemented using a wide 6695pointer of sorts to handle all the possible call mechanisms; the PMF 6696needs to store information about how to adjust the @samp{this} pointer, 6697and if the function pointed to is virtual, where to find the vtable, and 6698where in the vtable to look for the member function. If you are using 6699PMFs in an inner loop, you should really reconsider that decision. If 6700that is not an option, you can extract the pointer to the function that 6701would be called for a given object/PMF pair and call it directly inside 6702the inner loop, to save a bit of time. 6703 6704Note that you will still be paying the penalty for the call through a 6705function pointer; on most modern architectures, such a call defeats the 6706branch prediction features of the CPU@. This is also true of normal 6707virtual function calls. 6708 6709The syntax for this extension is 6710 6711@example 6712extern A a; 6713extern int (A::*fp)(); 6714typedef int (*fptr)(A *); 6715 6716fptr p = (fptr)(a.*fp); 6717@end example 6718 6719For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}), 6720no object is needed to obtain the address of the function. They can be 6721converted to function pointers directly: 6722 6723@example 6724fptr p1 = (fptr)(&A::foo); 6725@end example 6726 6727@opindex Wno-pmf-conversions 6728You must specify @option{-Wno-pmf-conversions} to use this extension. 6729 6730@node C++ Attributes 6731@section C++-Specific Variable, Function, and Type Attributes 6732 6733Some attributes only make sense for C++ programs. 6734 6735@table @code 6736@item init_priority (@var{priority}) 6737@cindex init_priority attribute 6738 6739 6740In Standard C++, objects defined at namespace scope are guaranteed to be 6741initialized in an order in strict accordance with that of their definitions 6742@emph{in a given translation unit}. No guarantee is made for initializations 6743across translation units. However, GNU C++ allows users to control the 6744order of initialization of objects defined at namespace scope with the 6745@code{init_priority} attribute by specifying a relative @var{priority}, 6746a constant integral expression currently bounded between 101 and 65535 6747inclusive. Lower numbers indicate a higher priority. 6748 6749In the following example, @code{A} would normally be created before 6750@code{B}, but the @code{init_priority} attribute has reversed that order: 6751 6752@example 6753Some_Class A __attribute__ ((init_priority (2000))); 6754Some_Class B __attribute__ ((init_priority (543))); 6755@end example 6756 6757@noindent 6758Note that the particular values of @var{priority} do not matter; only their 6759relative ordering. 6760 6761@item java_interface 6762@cindex java_interface attribute 6763 6764This type attribute informs C++ that the class is a Java interface. It may 6765only be applied to classes declared within an @code{extern "Java"} block. 6766Calls to methods declared in this interface will be dispatched using GCJ's 6767interface table mechanism, instead of regular virtual table dispatch. 6768 6769@end table 6770 6771@node Java Exceptions 6772@section Java Exceptions 6773 6774The Java language uses a slightly different exception handling model 6775from C++. Normally, GNU C++ will automatically detect when you are 6776writing C++ code that uses Java exceptions, and handle them 6777appropriately. However, if C++ code only needs to execute destructors 6778when Java exceptions are thrown through it, GCC will guess incorrectly. 6779Sample problematic code is: 6780 6781@example 6782 struct S @{ ~S(); @}; 6783 extern void bar(); // is written in Java, and may throw exceptions 6784 void foo() 6785 @{ 6786 S s; 6787 bar(); 6788 @} 6789@end example 6790 6791@noindent 6792The usual effect of an incorrect guess is a link failure, complaining of 6793a missing routine called @samp{__gxx_personality_v0}. 6794 6795You can inform the compiler that Java exceptions are to be used in a 6796translation unit, irrespective of what it might think, by writing 6797@samp{@w{#pragma GCC java_exceptions}} at the head of the file. This 6798@samp{#pragma} must appear before any functions that throw or catch 6799exceptions, or run destructors when exceptions are thrown through them. 6800 6801You cannot mix Java and C++ exceptions in the same translation unit. It 6802is believed to be safe to throw a C++ exception from one file through 6803another file compiled for the Java exception model, or vice versa, but 6804there may be bugs in this area. 6805 6806@node Deprecated Features 6807@section Deprecated Features 6808 6809In the past, the GNU C++ compiler was extended to experiment with new 6810features, at a time when the C++ language was still evolving. Now that 6811the C++ standard is complete, some of those features are superseded by 6812superior alternatives. Using the old features might cause a warning in 6813some cases that the feature will be dropped in the future. In other 6814cases, the feature might be gone already. 6815 6816While the list below is not exhaustive, it documents some of the options 6817that are now deprecated: 6818 6819@table @code 6820@item -fexternal-templates 6821@itemx -falt-external-templates 6822These are two of the many ways for g++ to implement template 6823instantiation. @xref{Template Instantiation}. The C++ standard clearly 6824defines how template definitions have to be organized across 6825implementation units. g++ has an implicit instantiation mechanism that 6826should work just fine for standard-conforming code. 6827 6828@item -fstrict-prototype 6829@itemx -fno-strict-prototype 6830Previously it was possible to use an empty prototype parameter list to 6831indicate an unspecified number of parameters (like C), rather than no 6832parameters, as C++ demands. This feature has been removed, except where 6833it is required for backwards compatibility @xref{Backwards Compatibility}. 6834@end table 6835 6836The named return value extension has been deprecated, and is now 6837removed from g++. 6838 6839The use of initializer lists with new expressions has been deprecated, 6840and is now removed from g++. 6841 6842Floating and complex non-type template parameters have been deprecated, 6843and are now removed from g++. 6844 6845The implicit typename extension has been deprecated and will be removed 6846from g++ at some point. In some cases g++ determines that a dependant 6847type such as @code{TPL<T>::X} is a type without needing a 6848@code{typename} keyword, contrary to the standard. 6849 6850@node Backwards Compatibility 6851@section Backwards Compatibility 6852@cindex Backwards Compatibility 6853@cindex ARM [Annotated C++ Reference Manual] 6854 6855Now that there is a definitive ISO standard C++, G++ has a specification 6856to adhere to. The C++ language evolved over time, and features that 6857used to be acceptable in previous drafts of the standard, such as the ARM 6858[Annotated C++ Reference Manual], are no longer accepted. In order to allow 6859compilation of C++ written to such drafts, G++ contains some backwards 6860compatibilities. @emph{All such backwards compatibility features are 6861liable to disappear in future versions of G++.} They should be considered 6862deprecated @xref{Deprecated Features}. 6863 6864@table @code 6865@item For scope 6866If a variable is declared at for scope, it used to remain in scope until 6867the end of the scope which contained the for statement (rather than just 6868within the for scope). G++ retains this, but issues a warning, if such a 6869variable is accessed outside the for scope. 6870 6871@item Implicit C language 6872Old C system header files did not contain an @code{extern "C" @{@dots{}@}} 6873scope to set the language. On such systems, all header files are 6874implicitly scoped inside a C language scope. Also, an empty prototype 6875@code{()} will be treated as an unspecified number of arguments, rather 6876than no arguments, as C++ demands. 6877@end table 6878