extend.texi revision 122180
1@c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001,2002, 2003 2@c Free Software Foundation, Inc. 3@c This is part of the GCC manual. 4@c For copying conditions, see the file gcc.texi. 5 6@node C Implementation 7@chapter C Implementation-defined behavior 8@cindex implementation-defined behavior, C language 9 10A conforming implementation of ISO C is required to document its 11choice of behavior in each of the areas that are designated 12``implementation defined.'' The following lists all such areas, 13along with the section number from the ISO/IEC 9899:1999 standard. 14 15@menu 16* Translation implementation:: 17* Environment implementation:: 18* Identifiers implementation:: 19* Characters implementation:: 20* Integers implementation:: 21* Floating point implementation:: 22* Arrays and pointers implementation:: 23* Hints implementation:: 24* Structures unions enumerations and bit-fields implementation:: 25* Qualifiers implementation:: 26* Preprocessing directives implementation:: 27* Library functions implementation:: 28* Architecture implementation:: 29* Locale-specific behavior implementation:: 30@end menu 31 32@node Translation implementation 33@section Translation 34 35@itemize @bullet 36@item 37@cite{How a diagnostic is identified (3.10, 5.1.1.3).} 38 39Diagnostics consist of all the output sent to stderr by GCC. 40 41@item 42@cite{Whether each nonempty sequence of white-space characters other than 43new-line is retained or replaced by one space character in translation 44phase 3 (5.1.1.2).} 45@end itemize 46 47@node Environment implementation 48@section Environment 49 50The behavior of these points are dependent on the implementation 51of the C library, and are not defined by GCC itself. 52 53@node Identifiers implementation 54@section Identifiers 55 56@itemize @bullet 57@item 58@cite{Which additional multibyte characters may appear in identifiers 59and their correspondence to universal character names (6.4.2).} 60 61@item 62@cite{The number of significant initial characters in an identifier 63(5.2.4.1, 6.4.2).} 64 65For internal names, all characters are significant. For external names, 66the number of significant characters are defined by the linker; for 67almost all targets, all characters are significant. 68 69@end itemize 70 71@node Characters implementation 72@section Characters 73 74@itemize @bullet 75@item 76@cite{The number of bits in a byte (3.6).} 77 78@item 79@cite{The values of the members of the execution character set (5.2.1).} 80 81@item 82@cite{The unique value of the member of the execution character set produced 83for each of the standard alphabetic escape sequences (5.2.2).} 84 85@item 86@cite{The value of a @code{char} object into which has been stored any 87character other than a member of the basic execution character set (6.2.5).} 88 89@item 90@cite{Which of @code{signed char} or @code{unsigned char} has the same range, 91representation, and behavior as ``plain'' @code{char} (6.2.5, 6.3.1.1).} 92 93@item 94@cite{The mapping of members of the source character set (in character 95constants and string literals) to members of the execution character 96set (6.4.4.4, 5.1.1.2).} 97 98@item 99@cite{The value of an integer character constant containing more than one 100character or containing a character or escape sequence that does not map 101to a single-byte execution character (6.4.4.4).} 102 103@item 104@cite{The value of a wide character constant containing more than one 105multibyte character, or containing a multibyte character or escape 106sequence not represented in the extended execution character set (6.4.4.4).} 107 108@item 109@cite{The current locale used to convert a wide character constant consisting 110of a single multibyte character that maps to a member of the extended 111execution character set into a corresponding wide character code (6.4.4.4).} 112 113@item 114@cite{The current locale used to convert a wide string literal into 115corresponding wide character codes (6.4.5).} 116 117@item 118@cite{The value of a string literal containing a multibyte character or escape 119sequence not represented in the execution character set (6.4.5).} 120@end itemize 121 122@node Integers implementation 123@section Integers 124 125@itemize @bullet 126@item 127@cite{Any extended integer types that exist in the implementation (6.2.5).} 128 129@item 130@cite{Whether signed integer types are represented using sign and magnitude, 131two's complement, or one's complement, and whether the extraordinary value 132is a trap representation or an ordinary value (6.2.6.2).} 133 134GCC supports only two's complement integer types, and all bit patterns 135are ordinary values. 136 137@item 138@cite{The rank of any extended integer type relative to another extended 139integer type with the same precision (6.3.1.1).} 140 141@item 142@cite{The result of, or the signal raised by, converting an integer to a 143signed integer type when the value cannot be represented in an object of 144that type (6.3.1.3).} 145 146@item 147@cite{The results of some bitwise operations on signed integers (6.5).} 148@end itemize 149 150@node Floating point implementation 151@section Floating point 152 153@itemize @bullet 154@item 155@cite{The accuracy of the floating-point operations and of the library 156functions in @code{<math.h>} and @code{<complex.h>} that return floating-point 157results (5.2.4.2.2).} 158 159@item 160@cite{The rounding behaviors characterized by non-standard values 161of @code{FLT_ROUNDS} @gol 162(5.2.4.2.2).} 163 164@item 165@cite{The evaluation methods characterized by non-standard negative 166values of @code{FLT_EVAL_METHOD} (5.2.4.2.2).} 167 168@item 169@cite{The direction of rounding when an integer is converted to a 170floating-point number that cannot exactly represent the original 171value (6.3.1.4).} 172 173@item 174@cite{The direction of rounding when a floating-point number is 175converted to a narrower floating-point number (6.3.1.5).} 176 177@item 178@cite{How the nearest representable value or the larger or smaller 179representable value immediately adjacent to the nearest representable 180value is chosen for certain floating constants (6.4.4.2).} 181 182@item 183@cite{Whether and how floating expressions are contracted when not 184disallowed by the @code{FP_CONTRACT} pragma (6.5).} 185 186@item 187@cite{The default state for the @code{FENV_ACCESS} pragma (7.6.1).} 188 189@item 190@cite{Additional floating-point exceptions, rounding modes, environments, 191and classifications, and their macro names (7.6, 7.12).} 192 193@item 194@cite{The default state for the @code{FP_CONTRACT} pragma (7.12.2).} 195 196@item 197@cite{Whether the ``inexact'' floating-point exception can be raised 198when the rounded result actually does equal the mathematical result 199in an IEC 60559 conformant implementation (F.9).} 200 201@item 202@cite{Whether the ``underflow'' (and ``inexact'') floating-point 203exception can be raised when a result is tiny but not inexact in an 204IEC 60559 conformant implementation (F.9).} 205 206@end itemize 207 208@node Arrays and pointers implementation 209@section Arrays and pointers 210 211@itemize @bullet 212@item 213@cite{The result of converting a pointer to an integer or 214vice versa (6.3.2.3).} 215 216A cast from pointer to integer discards most-significant bits if the 217pointer representation is larger than the integer type, 218sign-extends@footnote{Future versions of GCC may zero-extend, or use 219a target-defined @code{ptr_extend} pattern. Do not rely on sign extension.} 220if the pointer representation is smaller than the integer type, otherwise 221the bits are unchanged. 222@c ??? We've always claimed that pointers were unsigned entities. 223@c Shouldn't we therefore be doing zero-extension? If so, the bug 224@c is in convert_to_integer, where we call type_for_size and request 225@c a signed integral type. On the other hand, it might be most useful 226@c for the target if we extend according to POINTERS_EXTEND_UNSIGNED. 227 228A cast from integer to pointer discards most-significant bits if the 229pointer representation is smaller than the integer type, extends according 230to the signedness of the integer type if the pointer representation 231is larger than the integer type, otherwise the bits are unchanged. 232 233When casting from pointer to integer and back again, the resulting 234pointer must reference the same object as the original pointer, otherwise 235the behavior is undefined. That is, one may not use integer arithmetic to 236avoid the undefined behavior of pointer arithmetic as proscribed in 6.5.6/8. 237 238@item 239@cite{The size of the result of subtracting two pointers to elements 240of the same array (6.5.6).} 241 242@end itemize 243 244@node Hints implementation 245@section Hints 246 247@itemize @bullet 248@item 249@cite{The extent to which suggestions made by using the @code{register} 250storage-class specifier are effective (6.7.1).} 251 252The @code{register} specifier affects code generation only in these ways: 253 254@itemize @bullet 255@item 256When used as part of the register variable extension, see 257@ref{Explicit Reg Vars}. 258 259@item 260When @option{-O0} is in use, the compiler allocates distinct stack 261memory for all variables that do not have the @code{register} 262storage-class specifier; if @code{register} is specified, the variable 263may have a shorter lifespan than the code would indicate and may never 264be placed in memory. 265 266@item 267On some rare x86 targets, @code{setjmp} doesn't save the registers in 268all circumstances. In those cases, GCC doesn't allocate any variables 269in registers unless they are marked @code{register}. 270 271@end itemize 272 273@item 274@cite{The extent to which suggestions made by using the inline function 275specifier are effective (6.7.4).} 276 277GCC will not inline any functions if the @option{-fno-inline} option is 278used or if @option{-O0} is used. Otherwise, GCC may still be unable to 279inline a function for many reasons; the @option{-Winline} option may be 280used to determine if a function has not been inlined and why not. 281 282@end itemize 283 284@node Structures unions enumerations and bit-fields implementation 285@section Structures, unions, enumerations, and bit-fields 286 287@itemize @bullet 288@item 289@cite{Whether a ``plain'' int bit-field is treated as a @code{signed int} 290bit-field or as an @code{unsigned int} bit-field (6.7.2, 6.7.2.1).} 291 292@item 293@cite{Allowable bit-field types other than @code{_Bool}, @code{signed int}, 294and @code{unsigned int} (6.7.2.1).} 295 296@item 297@cite{Whether a bit-field can straddle a storage-unit boundary (6.7.2.1).} 298 299@item 300@cite{The order of allocation of bit-fields within a unit (6.7.2.1).} 301 302@item 303@cite{The alignment of non-bit-field members of structures (6.7.2.1).} 304 305@item 306@cite{The integer type compatible with each enumerated type (6.7.2.2).} 307 308@end itemize 309 310@node Qualifiers implementation 311@section Qualifiers 312 313@itemize @bullet 314@item 315@cite{What constitutes an access to an object that has volatile-qualified 316type (6.7.3).} 317 318@end itemize 319 320@node Preprocessing directives implementation 321@section Preprocessing directives 322 323@itemize @bullet 324@item 325@cite{How sequences in both forms of header names are mapped to headers 326or external source file names (6.4.7).} 327 328@item 329@cite{Whether the value of a character constant in a constant expression 330that controls conditional inclusion matches the value of the same character 331constant in the execution character set (6.10.1).} 332 333@item 334@cite{Whether the value of a single-character character constant in a 335constant expression that controls conditional inclusion may have a 336negative value (6.10.1).} 337 338@item 339@cite{The places that are searched for an included @samp{<>} delimited 340header, and how the places are specified or the header is 341identified (6.10.2).} 342 343@item 344@cite{How the named source file is searched for in an included @samp{""} 345delimited header (6.10.2).} 346 347@item 348@cite{The method by which preprocessing tokens (possibly resulting from 349macro expansion) in a @code{#include} directive are combined into a header 350name (6.10.2).} 351 352@item 353@cite{The nesting limit for @code{#include} processing (6.10.2).} 354 355GCC imposes a limit of 200 nested @code{#include}s. 356 357@item 358@cite{Whether the @samp{#} operator inserts a @samp{\} character before 359the @samp{\} character that begins a universal character name in a 360character constant or string literal (6.10.3.2).} 361 362@item 363@cite{The behavior on each recognized non-@code{STDC #pragma} 364directive (6.10.6).} 365 366@item 367@cite{The definitions for @code{__DATE__} and @code{__TIME__} when 368respectively, the date and time of translation are not available (6.10.8).} 369 370If the date and time are not available, @code{__DATE__} expands to 371@code{@w{"??? ?? ????"}} and @code{__TIME__} expands to 372@code{"??:??:??"}. 373 374@end itemize 375 376@node Library functions implementation 377@section Library functions 378 379The behavior of these points are dependent on the implementation 380of the C library, and are not defined by GCC itself. 381 382@node Architecture implementation 383@section Architecture 384 385@itemize @bullet 386@item 387@cite{The values or expressions assigned to the macros specified in the 388headers @code{<float.h>}, @code{<limits.h>}, and @code{<stdint.h>} 389(5.2.4.2, 7.18.2, 7.18.3).} 390 391@item 392@cite{The number, order, and encoding of bytes in any object 393(when not explicitly specified in this International Standard) (6.2.6.1).} 394 395@item 396@cite{The value of the result of the sizeof operator (6.5.3.4).} 397 398@end itemize 399 400@node Locale-specific behavior implementation 401@section Locale-specific behavior 402 403The behavior of these points are dependent on the implementation 404of the C library, and are not defined by GCC itself. 405 406@node C Extensions 407@chapter Extensions to the C Language Family 408@cindex extensions, C language 409@cindex C language extensions 410 411@opindex pedantic 412GNU C provides several language features not found in ISO standard C@. 413(The @option{-pedantic} option directs GCC to print a warning message if 414any of these features is used.) To test for the availability of these 415features in conditional compilation, check for a predefined macro 416@code{__GNUC__}, which is always defined under GCC@. 417 418These extensions are available in C and Objective-C@. Most of them are 419also available in C++. @xref{C++ Extensions,,Extensions to the 420C++ Language}, for extensions that apply @emph{only} to C++. 421 422Some features that are in ISO C99 but not C89 or C++ are also, as 423extensions, accepted by GCC in C89 mode and in C++. 424 425@menu 426* Statement Exprs:: Putting statements and declarations inside expressions. 427* Local Labels:: Labels local to a statement-expression. 428* Labels as Values:: Getting pointers to labels, and computed gotos. 429* Nested Functions:: As in Algol and Pascal, lexical scoping of functions. 430* Constructing Calls:: Dispatching a call to another function. 431* Typeof:: @code{typeof}: referring to the type of an expression. 432* Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues. 433* Conditionals:: Omitting the middle operand of a @samp{?:} expression. 434* Long Long:: Double-word integers---@code{long long int}. 435* Complex:: Data types for complex numbers. 436* Hex Floats:: Hexadecimal floating-point constants. 437* Zero Length:: Zero-length arrays. 438* Variable Length:: Arrays whose length is computed at run time. 439* Empty Structures:: Structures with no members. 440* Variadic Macros:: Macros with a variable number of arguments. 441* Escaped Newlines:: Slightly looser rules for escaped newlines. 442* Subscripting:: Any array can be subscripted, even if not an lvalue. 443* Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. 444* Initializers:: Non-constant initializers. 445* Compound Literals:: Compound literals give structures, unions 446 or arrays as values. 447* Designated Inits:: Labeling elements of initializers. 448* Cast to Union:: Casting to union type from any member of the union. 449* Case Ranges:: `case 1 ... 9' and such. 450* Mixed Declarations:: Mixing declarations and code. 451* Function Attributes:: Declaring that functions have no side effects, 452 or that they can never return. 453* Attribute Syntax:: Formal syntax for attributes. 454* Function Prototypes:: Prototype declarations and old-style definitions. 455* C++ Comments:: C++ comments are recognized. 456* Dollar Signs:: Dollar sign is allowed in identifiers. 457* Character Escapes:: @samp{\e} stands for the character @key{ESC}. 458* Variable Attributes:: Specifying attributes of variables. 459* Type Attributes:: Specifying attributes of types. 460* Alignment:: Inquiring about the alignment of a type or variable. 461* Inline:: Defining inline functions (as fast as macros). 462* Extended Asm:: Assembler instructions with C expressions as operands. 463 (With them you can define ``built-in'' functions.) 464* Constraints:: Constraints for asm operands 465* Asm Labels:: Specifying the assembler name to use for a C symbol. 466* Explicit Reg Vars:: Defining variables residing in specified registers. 467* Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. 468* Incomplete Enums:: @code{enum foo;}, with details to follow. 469* Function Names:: Printable strings which are the name of the current 470 function. 471* Return Address:: Getting the return or frame address of a function. 472* Vector Extensions:: Using vector instructions through built-in functions. 473* Other Builtins:: Other built-in functions. 474* Target Builtins:: Built-in functions specific to particular targets. 475* Pragmas:: Pragmas accepted by GCC. 476* Unnamed Fields:: Unnamed struct/union fields within structs/unions. 477* Thread-Local:: Per-thread variables. 478@end menu 479 480@node Statement Exprs 481@section Statements and Declarations in Expressions 482@cindex statements inside expressions 483@cindex declarations inside expressions 484@cindex expressions containing statements 485@cindex macros, statements in expressions 486 487@c the above section title wrapped and causes an underfull hbox.. i 488@c changed it from "within" to "in". --mew 4feb93 489A compound statement enclosed in parentheses may appear as an expression 490in GNU C@. This allows you to use loops, switches, and local variables 491within an expression. 492 493Recall that a compound statement is a sequence of statements surrounded 494by braces; in this construct, parentheses go around the braces. For 495example: 496 497@example 498(@{ int y = foo (); int z; 499 if (y > 0) z = y; 500 else z = - y; 501 z; @}) 502@end example 503 504@noindent 505is a valid (though slightly more complex than necessary) expression 506for the absolute value of @code{foo ()}. 507 508The last thing in the compound statement should be an expression 509followed by a semicolon; the value of this subexpression serves as the 510value of the entire construct. (If you use some other kind of statement 511last within the braces, the construct has type @code{void}, and thus 512effectively no value.) 513 514This feature is especially useful in making macro definitions ``safe'' (so 515that they evaluate each operand exactly once). For example, the 516``maximum'' function is commonly defined as a macro in standard C as 517follows: 518 519@example 520#define max(a,b) ((a) > (b) ? (a) : (b)) 521@end example 522 523@noindent 524@cindex side effects, macro argument 525But this definition computes either @var{a} or @var{b} twice, with bad 526results if the operand has side effects. In GNU C, if you know the 527type of the operands (here let's assume @code{int}), you can define 528the macro safely as follows: 529 530@example 531#define maxint(a,b) \ 532 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @}) 533@end example 534 535Embedded statements are not allowed in constant expressions, such as 536the value of an enumeration constant, the width of a bit-field, or 537the initial value of a static variable. 538 539If you don't know the type of the operand, you can still do this, but you 540must use @code{typeof} (@pxref{Typeof}). 541 542Statement expressions are not supported fully in G++, and their fate 543there is unclear. (It is possible that they will become fully supported 544at some point, or that they will be deprecated, or that the bugs that 545are present will continue to exist indefinitely.) Presently, statement 546expressions do not work well as default arguments. 547 548In addition, there are semantic issues with statement-expressions in 549C++. If you try to use statement-expressions instead of inline 550functions in C++, you may be surprised at the way object destruction is 551handled. For example: 552 553@example 554#define foo(a) (@{int b = (a); b + 3; @}) 555@end example 556 557@noindent 558does not work the same way as: 559 560@example 561inline int foo(int a) @{ int b = a; return b + 3; @} 562@end example 563 564@noindent 565In particular, if the expression passed into @code{foo} involves the 566creation of temporaries, the destructors for those temporaries will be 567run earlier in the case of the macro than in the case of the function. 568 569These considerations mean that it is probably a bad idea to use 570statement-expressions of this form in header files that are designed to 571work with C++. (Note that some versions of the GNU C Library contained 572header files using statement-expression that lead to precisely this 573bug.) 574 575@node Local Labels 576@section Locally Declared Labels 577@cindex local labels 578@cindex macros, local labels 579 580Each statement expression is a scope in which @dfn{local labels} can be 581declared. A local label is simply an identifier; you can jump to it 582with an ordinary @code{goto} statement, but only from within the 583statement expression it belongs to. 584 585A local label declaration looks like this: 586 587@example 588__label__ @var{label}; 589@end example 590 591@noindent 592or 593 594@example 595__label__ @var{label1}, @var{label2}, /* @r{@dots{}} */; 596@end example 597 598Local label declarations must come at the beginning of the statement 599expression, right after the @samp{(@{}, before any ordinary 600declarations. 601 602The label declaration defines the label @emph{name}, but does not define 603the label itself. You must do this in the usual way, with 604@code{@var{label}:}, within the statements of the statement expression. 605 606The local label feature is useful because statement expressions are 607often used in macros. If the macro contains nested loops, a @code{goto} 608can be useful for breaking out of them. However, an ordinary label 609whose scope is the whole function cannot be used: if the macro can be 610expanded several times in one function, the label will be multiply 611defined in that function. A local label avoids this problem. For 612example: 613 614@example 615#define SEARCH(array, target) \ 616(@{ \ 617 __label__ found; \ 618 typeof (target) _SEARCH_target = (target); \ 619 typeof (*(array)) *_SEARCH_array = (array); \ 620 int i, j; \ 621 int value; \ 622 for (i = 0; i < max; i++) \ 623 for (j = 0; j < max; j++) \ 624 if (_SEARCH_array[i][j] == _SEARCH_target) \ 625 @{ value = i; goto found; @} \ 626 value = -1; \ 627 found: \ 628 value; \ 629@}) 630@end example 631 632@node Labels as Values 633@section Labels as Values 634@cindex labels as values 635@cindex computed gotos 636@cindex goto with computed label 637@cindex address of a label 638 639You can get the address of a label defined in the current function 640(or a containing function) with the unary operator @samp{&&}. The 641value has type @code{void *}. This value is a constant and can be used 642wherever a constant of that type is valid. For example: 643 644@example 645void *ptr; 646/* @r{@dots{}} */ 647ptr = &&foo; 648@end example 649 650To use these values, you need to be able to jump to one. This is done 651with the computed goto statement@footnote{The analogous feature in 652Fortran is called an assigned goto, but that name seems inappropriate in 653C, where one can do more than simply store label addresses in label 654variables.}, @code{goto *@var{exp};}. For example, 655 656@example 657goto *ptr; 658@end example 659 660@noindent 661Any expression of type @code{void *} is allowed. 662 663One way of using these constants is in initializing a static array that 664will serve as a jump table: 665 666@example 667static void *array[] = @{ &&foo, &&bar, &&hack @}; 668@end example 669 670Then you can select a label with indexing, like this: 671 672@example 673goto *array[i]; 674@end example 675 676@noindent 677Note that this does not check whether the subscript is in bounds---array 678indexing in C never does that. 679 680Such an array of label values serves a purpose much like that of the 681@code{switch} statement. The @code{switch} statement is cleaner, so 682use that rather than an array unless the problem does not fit a 683@code{switch} statement very well. 684 685Another use of label values is in an interpreter for threaded code. 686The labels within the interpreter function can be stored in the 687threaded code for super-fast dispatching. 688 689You may not use this mechanism to jump to code in a different function. 690If you do that, totally unpredictable things will happen. The best way to 691avoid this is to store the label address only in automatic variables and 692never pass it as an argument. 693 694An alternate way to write the above example is 695 696@example 697static const int array[] = @{ &&foo - &&foo, &&bar - &&foo, 698 &&hack - &&foo @}; 699goto *(&&foo + array[i]); 700@end example 701 702@noindent 703This is more friendly to code living in shared libraries, as it reduces 704the number of dynamic relocations that are needed, and by consequence, 705allows the data to be read-only. 706 707@node Nested Functions 708@section Nested Functions 709@cindex nested functions 710@cindex downward funargs 711@cindex thunks 712 713A @dfn{nested function} is a function defined inside another function. 714(Nested functions are not supported for GNU C++.) The nested function's 715name is local to the block where it is defined. For example, here we 716define a nested function named @code{square}, and call it twice: 717 718@example 719@group 720foo (double a, double b) 721@{ 722 double square (double z) @{ return z * z; @} 723 724 return square (a) + square (b); 725@} 726@end group 727@end example 728 729The nested function can access all the variables of the containing 730function that are visible at the point of its definition. This is 731called @dfn{lexical scoping}. For example, here we show a nested 732function which uses an inherited variable named @code{offset}: 733 734@example 735@group 736bar (int *array, int offset, int size) 737@{ 738 int access (int *array, int index) 739 @{ return array[index + offset]; @} 740 int i; 741 /* @r{@dots{}} */ 742 for (i = 0; i < size; i++) 743 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */ 744@} 745@end group 746@end example 747 748Nested function definitions are permitted within functions in the places 749where variable definitions are allowed; that is, in any block, before 750the first statement in the block. 751 752It is possible to call the nested function from outside the scope of its 753name by storing its address or passing the address to another function: 754 755@example 756hack (int *array, int size) 757@{ 758 void store (int index, int value) 759 @{ array[index] = value; @} 760 761 intermediate (store, size); 762@} 763@end example 764 765Here, the function @code{intermediate} receives the address of 766@code{store} as an argument. If @code{intermediate} calls @code{store}, 767the arguments given to @code{store} are used to store into @code{array}. 768But this technique works only so long as the containing function 769(@code{hack}, in this example) does not exit. 770 771If you try to call the nested function through its address after the 772containing function has exited, all hell will break loose. If you try 773to call it after a containing scope level has exited, and if it refers 774to some of the variables that are no longer in scope, you may be lucky, 775but it's not wise to take the risk. If, however, the nested function 776does not refer to anything that has gone out of scope, you should be 777safe. 778 779GCC implements taking the address of a nested function using a technique 780called @dfn{trampolines}. A paper describing them is available as 781 782@noindent 783@uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}. 784 785A nested function can jump to a label inherited from a containing 786function, provided the label was explicitly declared in the containing 787function (@pxref{Local Labels}). Such a jump returns instantly to the 788containing function, exiting the nested function which did the 789@code{goto} and any intermediate functions as well. Here is an example: 790 791@example 792@group 793bar (int *array, int offset, int size) 794@{ 795 __label__ failure; 796 int access (int *array, int index) 797 @{ 798 if (index > size) 799 goto failure; 800 return array[index + offset]; 801 @} 802 int i; 803 /* @r{@dots{}} */ 804 for (i = 0; i < size; i++) 805 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */ 806 /* @r{@dots{}} */ 807 return 0; 808 809 /* @r{Control comes here from @code{access} 810 if it detects an error.} */ 811 failure: 812 return -1; 813@} 814@end group 815@end example 816 817A nested function always has internal linkage. Declaring one with 818@code{extern} is erroneous. If you need to declare the nested function 819before its definition, use @code{auto} (which is otherwise meaningless 820for function declarations). 821 822@example 823bar (int *array, int offset, int size) 824@{ 825 __label__ failure; 826 auto int access (int *, int); 827 /* @r{@dots{}} */ 828 int access (int *array, int index) 829 @{ 830 if (index > size) 831 goto failure; 832 return array[index + offset]; 833 @} 834 /* @r{@dots{}} */ 835@} 836@end example 837 838@node Constructing Calls 839@section Constructing Function Calls 840@cindex constructing calls 841@cindex forwarding calls 842 843Using the built-in functions described below, you can record 844the arguments a function received, and call another function 845with the same arguments, without knowing the number or types 846of the arguments. 847 848You can also record the return value of that function call, 849and later return that value, without knowing what data type 850the function tried to return (as long as your caller expects 851that data type). 852 853@deftypefn {Built-in Function} {void *} __builtin_apply_args () 854This built-in function returns a pointer to data 855describing how to perform a call with the same arguments as were passed 856to the current function. 857 858The function saves the arg pointer register, structure value address, 859and all registers that might be used to pass arguments to a function 860into a block of memory allocated on the stack. Then it returns the 861address of that block. 862@end deftypefn 863 864@deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size}) 865This built-in function invokes @var{function} 866with a copy of the parameters described by @var{arguments} 867and @var{size}. 868 869The value of @var{arguments} should be the value returned by 870@code{__builtin_apply_args}. The argument @var{size} specifies the size 871of the stack argument data, in bytes. 872 873This function returns a pointer to data describing 874how to return whatever value was returned by @var{function}. The data 875is saved in a block of memory allocated on the stack. 876 877It is not always simple to compute the proper value for @var{size}. The 878value is used by @code{__builtin_apply} to compute the amount of data 879that should be pushed on the stack and copied from the incoming argument 880area. 881@end deftypefn 882 883@deftypefn {Built-in Function} {void} __builtin_return (void *@var{result}) 884This built-in function returns the value described by @var{result} from 885the containing function. You should specify, for @var{result}, a value 886returned by @code{__builtin_apply}. 887@end deftypefn 888 889@node Typeof 890@section Referring to a Type with @code{typeof} 891@findex typeof 892@findex sizeof 893@cindex macros, types of arguments 894 895Another way to refer to the type of an expression is with @code{typeof}. 896The syntax of using of this keyword looks like @code{sizeof}, but the 897construct acts semantically like a type name defined with @code{typedef}. 898 899There are two ways of writing the argument to @code{typeof}: with an 900expression or with a type. Here is an example with an expression: 901 902@example 903typeof (x[0](1)) 904@end example 905 906@noindent 907This assumes that @code{x} is an array of pointers to functions; 908the type described is that of the values of the functions. 909 910Here is an example with a typename as the argument: 911 912@example 913typeof (int *) 914@end example 915 916@noindent 917Here the type described is that of pointers to @code{int}. 918 919If you are writing a header file that must work when included in ISO C 920programs, write @code{__typeof__} instead of @code{typeof}. 921@xref{Alternate Keywords}. 922 923A @code{typeof}-construct can be used anywhere a typedef name could be 924used. For example, you can use it in a declaration, in a cast, or inside 925of @code{sizeof} or @code{typeof}. 926 927@code{typeof} is often useful in conjunction with the 928statements-within-expressions feature. Here is how the two together can 929be used to define a safe ``maximum'' macro that operates on any 930arithmetic type and evaluates each of its arguments exactly once: 931 932@example 933#define max(a,b) \ 934 (@{ typeof (a) _a = (a); \ 935 typeof (b) _b = (b); \ 936 _a > _b ? _a : _b; @}) 937@end example 938 939@cindex underscores in variables in macros 940@cindex @samp{_} in variables in macros 941@cindex local variables in macros 942@cindex variables, local, in macros 943@cindex macros, local variables in 944 945The reason for using names that start with underscores for the local 946variables is to avoid conflicts with variable names that occur within the 947expressions that are substituted for @code{a} and @code{b}. Eventually we 948hope to design a new form of declaration syntax that allows you to declare 949variables whose scopes start only after their initializers; this will be a 950more reliable way to prevent such conflicts. 951 952@noindent 953Some more examples of the use of @code{typeof}: 954 955@itemize @bullet 956@item 957This declares @code{y} with the type of what @code{x} points to. 958 959@example 960typeof (*x) y; 961@end example 962 963@item 964This declares @code{y} as an array of such values. 965 966@example 967typeof (*x) y[4]; 968@end example 969 970@item 971This declares @code{y} as an array of pointers to characters: 972 973@example 974typeof (typeof (char *)[4]) y; 975@end example 976 977@noindent 978It is equivalent to the following traditional C declaration: 979 980@example 981char *y[4]; 982@end example 983 984To see the meaning of the declaration using @code{typeof}, and why it 985might be a useful way to write, let's rewrite it with these macros: 986 987@example 988#define pointer(T) typeof(T *) 989#define array(T, N) typeof(T [N]) 990@end example 991 992@noindent 993Now the declaration can be rewritten this way: 994 995@example 996array (pointer (char), 4) y; 997@end example 998 999@noindent 1000Thus, @code{array (pointer (char), 4)} is the type of arrays of 4 1001pointers to @code{char}. 1002@end itemize 1003 1004@emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported 1005a more limited extension which permitted one to write 1006 1007@example 1008typedef @var{T} = @var{expr}; 1009@end example 1010 1011@noindent 1012with the effect of declaring @var{T} to have the type of the expression 1013@var{expr}. This extension does not work with GCC 3 (versions between 10143.0 and 3.2 will crash; 3.2.1 and later give an error). Code which 1015relies on it should be rewritten to use @code{typeof}: 1016 1017@example 1018typedef typeof(@var{expr}) @var{T}; 1019@end example 1020 1021@noindent 1022This will work with all versions of GCC@. 1023 1024@node Lvalues 1025@section Generalized Lvalues 1026@cindex compound expressions as lvalues 1027@cindex expressions, compound, as lvalues 1028@cindex conditional expressions as lvalues 1029@cindex expressions, conditional, as lvalues 1030@cindex casts as lvalues 1031@cindex generalized lvalues 1032@cindex lvalues, generalized 1033@cindex extensions, @code{?:} 1034@cindex @code{?:} extensions 1035 1036Compound expressions, conditional expressions and casts are allowed as 1037lvalues provided their operands are lvalues. This means that you can take 1038their addresses or store values into them. 1039 1040Standard C++ allows compound expressions and conditional expressions as 1041lvalues, and permits casts to reference type, so use of this extension 1042is deprecated for C++ code. 1043 1044For example, a compound expression can be assigned, provided the last 1045expression in the sequence is an lvalue. These two expressions are 1046equivalent: 1047 1048@example 1049(a, b) += 5 1050a, (b += 5) 1051@end example 1052 1053Similarly, the address of the compound expression can be taken. These two 1054expressions are equivalent: 1055 1056@example 1057&(a, b) 1058a, &b 1059@end example 1060 1061A conditional expression is a valid lvalue if its type is not void and the 1062true and false branches are both valid lvalues. For example, these two 1063expressions are equivalent: 1064 1065@example 1066(a ? b : c) = 5 1067(a ? b = 5 : (c = 5)) 1068@end example 1069 1070A cast is a valid lvalue if its operand is an lvalue. A simple 1071assignment whose left-hand side is a cast works by converting the 1072right-hand side first to the specified type, then to the type of the 1073inner left-hand side expression. After this is stored, the value is 1074converted back to the specified type to become the value of the 1075assignment. Thus, if @code{a} has type @code{char *}, the following two 1076expressions are equivalent: 1077 1078@example 1079(int)a = 5 1080(int)(a = (char *)(int)5) 1081@end example 1082 1083An assignment-with-arithmetic operation such as @samp{+=} applied to a cast 1084performs the arithmetic using the type resulting from the cast, and then 1085continues as in the previous case. Therefore, these two expressions are 1086equivalent: 1087 1088@example 1089(int)a += 5 1090(int)(a = (char *)(int) ((int)a + 5)) 1091@end example 1092 1093You cannot take the address of an lvalue cast, because the use of its 1094address would not work out coherently. Suppose that @code{&(int)f} were 1095permitted, where @code{f} has type @code{float}. Then the following 1096statement would try to store an integer bit-pattern where a floating 1097point number belongs: 1098 1099@example 1100*&(int)f = 1; 1101@end example 1102 1103This is quite different from what @code{(int)f = 1} would do---that 1104would convert 1 to floating point and store it. Rather than cause this 1105inconsistency, we think it is better to prohibit use of @samp{&} on a cast. 1106 1107If you really do want an @code{int *} pointer with the address of 1108@code{f}, you can simply write @code{(int *)&f}. 1109 1110@node Conditionals 1111@section Conditionals with Omitted Operands 1112@cindex conditional expressions, extensions 1113@cindex omitted middle-operands 1114@cindex middle-operands, omitted 1115@cindex extensions, @code{?:} 1116@cindex @code{?:} extensions 1117 1118The middle operand in a conditional expression may be omitted. Then 1119if the first operand is nonzero, its value is the value of the conditional 1120expression. 1121 1122Therefore, the expression 1123 1124@example 1125x ? : y 1126@end example 1127 1128@noindent 1129has the value of @code{x} if that is nonzero; otherwise, the value of 1130@code{y}. 1131 1132This example is perfectly equivalent to 1133 1134@example 1135x ? x : y 1136@end example 1137 1138@cindex side effect in ?: 1139@cindex ?: side effect 1140@noindent 1141In this simple case, the ability to omit the middle operand is not 1142especially useful. When it becomes useful is when the first operand does, 1143or may (if it is a macro argument), contain a side effect. Then repeating 1144the operand in the middle would perform the side effect twice. Omitting 1145the middle operand uses the value already computed without the undesirable 1146effects of recomputing it. 1147 1148@node Long Long 1149@section Double-Word Integers 1150@cindex @code{long long} data types 1151@cindex double-word arithmetic 1152@cindex multiprecision arithmetic 1153@cindex @code{LL} integer suffix 1154@cindex @code{ULL} integer suffix 1155 1156ISO C99 supports data types for integers that are at least 64 bits wide, 1157and as an extension GCC supports them in C89 mode and in C++. 1158Simply write @code{long long int} for a signed integer, or 1159@code{unsigned long long int} for an unsigned integer. To make an 1160integer constant of type @code{long long int}, add the suffix @samp{LL} 1161to the integer. To make an integer constant of type @code{unsigned long 1162long int}, add the suffix @samp{ULL} to the integer. 1163 1164You can use these types in arithmetic like any other integer types. 1165Addition, subtraction, and bitwise boolean operations on these types 1166are open-coded on all types of machines. Multiplication is open-coded 1167if the machine supports fullword-to-doubleword a widening multiply 1168instruction. Division and shifts are open-coded only on machines that 1169provide special support. The operations that are not open-coded use 1170special library routines that come with GCC@. 1171 1172There may be pitfalls when you use @code{long long} types for function 1173arguments, unless you declare function prototypes. If a function 1174expects type @code{int} for its argument, and you pass a value of type 1175@code{long long int}, confusion will result because the caller and the 1176subroutine will disagree about the number of bytes for the argument. 1177Likewise, if the function expects @code{long long int} and you pass 1178@code{int}. The best way to avoid such problems is to use prototypes. 1179 1180@node Complex 1181@section Complex Numbers 1182@cindex complex numbers 1183@cindex @code{_Complex} keyword 1184@cindex @code{__complex__} keyword 1185 1186ISO C99 supports complex floating data types, and as an extension GCC 1187supports them in C89 mode and in C++, and supports complex integer data 1188types which are not part of ISO C99. You can declare complex types 1189using the keyword @code{_Complex}. As an extension, the older GNU 1190keyword @code{__complex__} is also supported. 1191 1192For example, @samp{_Complex double x;} declares @code{x} as a 1193variable whose real part and imaginary part are both of type 1194@code{double}. @samp{_Complex short int y;} declares @code{y} to 1195have real and imaginary parts of type @code{short int}; this is not 1196likely to be useful, but it shows that the set of complex types is 1197complete. 1198 1199To write a constant with a complex data type, use the suffix @samp{i} or 1200@samp{j} (either one; they are equivalent). For example, @code{2.5fi} 1201has type @code{_Complex float} and @code{3i} has type 1202@code{_Complex int}. Such a constant always has a pure imaginary 1203value, but you can form any complex value you like by adding one to a 1204real constant. This is a GNU extension; if you have an ISO C99 1205conforming C library (such as GNU libc), and want to construct complex 1206constants of floating type, you should include @code{<complex.h>} and 1207use the macros @code{I} or @code{_Complex_I} instead. 1208 1209@cindex @code{__real__} keyword 1210@cindex @code{__imag__} keyword 1211To extract the real part of a complex-valued expression @var{exp}, write 1212@code{__real__ @var{exp}}. Likewise, use @code{__imag__} to 1213extract the imaginary part. This is a GNU extension; for values of 1214floating type, you should use the ISO C99 functions @code{crealf}, 1215@code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and 1216@code{cimagl}, declared in @code{<complex.h>} and also provided as 1217built-in functions by GCC@. 1218 1219@cindex complex conjugation 1220The operator @samp{~} performs complex conjugation when used on a value 1221with a complex type. This is a GNU extension; for values of 1222floating type, you should use the ISO C99 functions @code{conjf}, 1223@code{conj} and @code{conjl}, declared in @code{<complex.h>} and also 1224provided as built-in functions by GCC@. 1225 1226GCC can allocate complex automatic variables in a noncontiguous 1227fashion; it's even possible for the real part to be in a register while 1228the imaginary part is on the stack (or vice-versa). Only the DWARF2 1229debug info format can represent this, so use of DWARF2 is recommended. 1230If you are using the stabs debug info format, GCC describes a noncontiguous 1231complex variable as if it were two separate variables of noncomplex type. 1232If the variable's actual name is @code{foo}, the two fictitious 1233variables are named @code{foo$real} and @code{foo$imag}. You can 1234examine and set these two fictitious variables with your debugger. 1235 1236@node Hex Floats 1237@section Hex Floats 1238@cindex hex floats 1239 1240ISO C99 supports floating-point numbers written not only in the usual 1241decimal notation, such as @code{1.55e1}, but also numbers such as 1242@code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC 1243supports this in C89 mode (except in some cases when strictly 1244conforming) and in C++. In that format the 1245@samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are 1246mandatory. The exponent is a decimal number that indicates the power of 12472 by which the significant part will be multiplied. Thus @samp{0x1.f} is 1248@tex 1249$1 {15\over16}$, 1250@end tex 1251@ifnottex 12521 15/16, 1253@end ifnottex 1254@samp{p3} multiplies it by 8, and the value of @code{0x1.fp3} 1255is the same as @code{1.55e1}. 1256 1257Unlike for floating-point numbers in the decimal notation the exponent 1258is always required in the hexadecimal notation. Otherwise the compiler 1259would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This 1260could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the 1261extension for floating-point constants of type @code{float}. 1262 1263@node Zero Length 1264@section Arrays of Length Zero 1265@cindex arrays of length zero 1266@cindex zero-length arrays 1267@cindex length-zero arrays 1268@cindex flexible array members 1269 1270Zero-length arrays are allowed in GNU C@. They are very useful as the 1271last element of a structure which is really a header for a variable-length 1272object: 1273 1274@example 1275struct line @{ 1276 int length; 1277 char contents[0]; 1278@}; 1279 1280struct line *thisline = (struct line *) 1281 malloc (sizeof (struct line) + this_length); 1282thisline->length = this_length; 1283@end example 1284 1285In ISO C90, you would have to give @code{contents} a length of 1, which 1286means either you waste space or complicate the argument to @code{malloc}. 1287 1288In ISO C99, you would use a @dfn{flexible array member}, which is 1289slightly different in syntax and semantics: 1290 1291@itemize @bullet 1292@item 1293Flexible array members are written as @code{contents[]} without 1294the @code{0}. 1295 1296@item 1297Flexible array members have incomplete type, and so the @code{sizeof} 1298operator may not be applied. As a quirk of the original implementation 1299of zero-length arrays, @code{sizeof} evaluates to zero. 1300 1301@item 1302Flexible array members may only appear as the last member of a 1303@code{struct} that is otherwise non-empty. 1304 1305@item 1306A structure containing a flexible array member, or a union containing 1307such a structure (possibly recursively), may not be a member of a 1308structure or an element of an array. (However, these uses are 1309permitted by GCC as extensions.) 1310@end itemize 1311 1312GCC versions before 3.0 allowed zero-length arrays to be statically 1313initialized, as if they were flexible arrays. In addition to those 1314cases that were useful, it also allowed initializations in situations 1315that would corrupt later data. Non-empty initialization of zero-length 1316arrays is now treated like any case where there are more initializer 1317elements than the array holds, in that a suitable warning about "excess 1318elements in array" is given, and the excess elements (all of them, in 1319this case) are ignored. 1320 1321Instead GCC allows static initialization of flexible array members. 1322This is equivalent to defining a new structure containing the original 1323structure followed by an array of sufficient size to contain the data. 1324I.e.@: in the following, @code{f1} is constructed as if it were declared 1325like @code{f2}. 1326 1327@example 1328struct f1 @{ 1329 int x; int y[]; 1330@} f1 = @{ 1, @{ 2, 3, 4 @} @}; 1331 1332struct f2 @{ 1333 struct f1 f1; int data[3]; 1334@} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @}; 1335@end example 1336 1337@noindent 1338The convenience of this extension is that @code{f1} has the desired 1339type, eliminating the need to consistently refer to @code{f2.f1}. 1340 1341This has symmetry with normal static arrays, in that an array of 1342unknown size is also written with @code{[]}. 1343 1344Of course, this extension only makes sense if the extra data comes at 1345the end of a top-level object, as otherwise we would be overwriting 1346data at subsequent offsets. To avoid undue complication and confusion 1347with initialization of deeply nested arrays, we simply disallow any 1348non-empty initialization except when the structure is the top-level 1349object. For example: 1350 1351@example 1352struct foo @{ int x; int y[]; @}; 1353struct bar @{ struct foo z; @}; 1354 1355struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.} 1356struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 1357struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.} 1358struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 1359@end example 1360 1361@node Empty Structures 1362@section Structures With No Members 1363@cindex empty structures 1364@cindex zero-size structures 1365 1366GCC permits a C structure to have no members: 1367 1368@example 1369struct empty @{ 1370@}; 1371@end example 1372 1373The structure will have size zero. In C++, empty structures are part 1374of the language. G++ treats empty structures as if they had a single 1375member of type @code{char}. 1376 1377@node Variable Length 1378@section Arrays of Variable Length 1379@cindex variable-length arrays 1380@cindex arrays of variable length 1381@cindex VLAs 1382 1383Variable-length automatic arrays are allowed in ISO C99, and as an 1384extension GCC accepts them in C89 mode and in C++. (However, GCC's 1385implementation of variable-length arrays does not yet conform in detail 1386to the ISO C99 standard.) These arrays are 1387declared like any other automatic arrays, but with a length that is not 1388a constant expression. The storage is allocated at the point of 1389declaration and deallocated when the brace-level is exited. For 1390example: 1391 1392@example 1393FILE * 1394concat_fopen (char *s1, char *s2, char *mode) 1395@{ 1396 char str[strlen (s1) + strlen (s2) + 1]; 1397 strcpy (str, s1); 1398 strcat (str, s2); 1399 return fopen (str, mode); 1400@} 1401@end example 1402 1403@cindex scope of a variable length array 1404@cindex variable-length array scope 1405@cindex deallocating variable length arrays 1406Jumping or breaking out of the scope of the array name deallocates the 1407storage. Jumping into the scope is not allowed; you get an error 1408message for it. 1409 1410@cindex @code{alloca} vs variable-length arrays 1411You can use the function @code{alloca} to get an effect much like 1412variable-length arrays. The function @code{alloca} is available in 1413many other C implementations (but not in all). On the other hand, 1414variable-length arrays are more elegant. 1415 1416There are other differences between these two methods. Space allocated 1417with @code{alloca} exists until the containing @emph{function} returns. 1418The space for a variable-length array is deallocated as soon as the array 1419name's scope ends. (If you use both variable-length arrays and 1420@code{alloca} in the same function, deallocation of a variable-length array 1421will also deallocate anything more recently allocated with @code{alloca}.) 1422 1423You can also use variable-length arrays as arguments to functions: 1424 1425@example 1426struct entry 1427tester (int len, char data[len][len]) 1428@{ 1429 /* @r{@dots{}} */ 1430@} 1431@end example 1432 1433The length of an array is computed once when the storage is allocated 1434and is remembered for the scope of the array in case you access it with 1435@code{sizeof}. 1436 1437If you want to pass the array first and the length afterward, you can 1438use a forward declaration in the parameter list---another GNU extension. 1439 1440@example 1441struct entry 1442tester (int len; char data[len][len], int len) 1443@{ 1444 /* @r{@dots{}} */ 1445@} 1446@end example 1447 1448@cindex parameter forward declaration 1449The @samp{int len} before the semicolon is a @dfn{parameter forward 1450declaration}, and it serves the purpose of making the name @code{len} 1451known when the declaration of @code{data} is parsed. 1452 1453You can write any number of such parameter forward declarations in the 1454parameter list. They can be separated by commas or semicolons, but the 1455last one must end with a semicolon, which is followed by the ``real'' 1456parameter declarations. Each forward declaration must match a ``real'' 1457declaration in parameter name and data type. ISO C99 does not support 1458parameter forward declarations. 1459 1460@node Variadic Macros 1461@section Macros with a Variable Number of Arguments. 1462@cindex variable number of arguments 1463@cindex macro with variable arguments 1464@cindex rest argument (in macro) 1465@cindex variadic macros 1466 1467In the ISO C standard of 1999, a macro can be declared to accept a 1468variable number of arguments much as a function can. The syntax for 1469defining the macro is similar to that of a function. Here is an 1470example: 1471 1472@smallexample 1473#define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) 1474@end smallexample 1475 1476Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of 1477such a macro, it represents the zero or more tokens until the closing 1478parenthesis that ends the invocation, including any commas. This set of 1479tokens replaces the identifier @code{__VA_ARGS__} in the macro body 1480wherever it appears. See the CPP manual for more information. 1481 1482GCC has long supported variadic macros, and used a different syntax that 1483allowed you to give a name to the variable arguments just like any other 1484argument. Here is an example: 1485 1486@example 1487#define debug(format, args...) fprintf (stderr, format, args) 1488@end example 1489 1490This is in all ways equivalent to the ISO C example above, but arguably 1491more readable and descriptive. 1492 1493GNU CPP has two further variadic macro extensions, and permits them to 1494be used with either of the above forms of macro definition. 1495 1496In standard C, you are not allowed to leave the variable argument out 1497entirely; but you are allowed to pass an empty argument. For example, 1498this invocation is invalid in ISO C, because there is no comma after 1499the string: 1500 1501@example 1502debug ("A message") 1503@end example 1504 1505GNU CPP permits you to completely omit the variable arguments in this 1506way. In the above examples, the compiler would complain, though since 1507the expansion of the macro still has the extra comma after the format 1508string. 1509 1510To help solve this problem, CPP behaves specially for variable arguments 1511used with the token paste operator, @samp{##}. If instead you write 1512 1513@smallexample 1514#define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) 1515@end smallexample 1516 1517and if the variable arguments are omitted or empty, the @samp{##} 1518operator causes the preprocessor to remove the comma before it. If you 1519do provide some variable arguments in your macro invocation, GNU CPP 1520does not complain about the paste operation and instead places the 1521variable arguments after the comma. Just like any other pasted macro 1522argument, these arguments are not macro expanded. 1523 1524@node Escaped Newlines 1525@section Slightly Looser Rules for Escaped Newlines 1526@cindex escaped newlines 1527@cindex newlines (escaped) 1528 1529Recently, the preprocessor has relaxed its treatment of escaped 1530newlines. Previously, the newline had to immediately follow a 1531backslash. The current implementation allows whitespace in the form 1532of spaces, horizontal and vertical tabs, and form feeds between the 1533backslash and the subsequent newline. The preprocessor issues a 1534warning, but treats it as a valid escaped newline and combines the two 1535lines to form a single logical line. This works within comments and 1536tokens, as well as between tokens. Comments are @emph{not} treated as 1537whitespace for the purposes of this relaxation, since they have not 1538yet been replaced with spaces. 1539 1540@node Subscripting 1541@section Non-Lvalue Arrays May Have Subscripts 1542@cindex subscripting 1543@cindex arrays, non-lvalue 1544 1545@cindex subscripting and function values 1546In ISO C99, arrays that are not lvalues still decay to pointers, and 1547may be subscripted, although they may not be modified or used after 1548the next sequence point and the unary @samp{&} operator may not be 1549applied to them. As an extension, GCC allows such arrays to be 1550subscripted in C89 mode, though otherwise they do not decay to 1551pointers outside C99 mode. For example, 1552this is valid in GNU C though not valid in C89: 1553 1554@example 1555@group 1556struct foo @{int a[4];@}; 1557 1558struct foo f(); 1559 1560bar (int index) 1561@{ 1562 return f().a[index]; 1563@} 1564@end group 1565@end example 1566 1567@node Pointer Arith 1568@section Arithmetic on @code{void}- and Function-Pointers 1569@cindex void pointers, arithmetic 1570@cindex void, size of pointer to 1571@cindex function pointers, arithmetic 1572@cindex function, size of pointer to 1573 1574In GNU C, addition and subtraction operations are supported on pointers to 1575@code{void} and on pointers to functions. This is done by treating the 1576size of a @code{void} or of a function as 1. 1577 1578A consequence of this is that @code{sizeof} is also allowed on @code{void} 1579and on function types, and returns 1. 1580 1581@opindex Wpointer-arith 1582The option @option{-Wpointer-arith} requests a warning if these extensions 1583are used. 1584 1585@node Initializers 1586@section Non-Constant Initializers 1587@cindex initializers, non-constant 1588@cindex non-constant initializers 1589 1590As in standard C++ and ISO C99, the elements of an aggregate initializer for an 1591automatic variable are not required to be constant expressions in GNU C@. 1592Here is an example of an initializer with run-time varying elements: 1593 1594@example 1595foo (float f, float g) 1596@{ 1597 float beat_freqs[2] = @{ f-g, f+g @}; 1598 /* @r{@dots{}} */ 1599@} 1600@end example 1601 1602@node Compound Literals 1603@section Compound Literals 1604@cindex constructor expressions 1605@cindex initializations in expressions 1606@cindex structures, constructor expression 1607@cindex expressions, constructor 1608@cindex compound literals 1609@c The GNU C name for what C99 calls compound literals was "constructor expressions". 1610 1611ISO C99 supports compound literals. A compound literal looks like 1612a cast containing an initializer. Its value is an object of the 1613type specified in the cast, containing the elements specified in 1614the initializer; it is an lvalue. As an extension, GCC supports 1615compound literals in C89 mode and in C++. 1616 1617Usually, the specified type is a structure. Assume that 1618@code{struct foo} and @code{structure} are declared as shown: 1619 1620@example 1621struct foo @{int a; char b[2];@} structure; 1622@end example 1623 1624@noindent 1625Here is an example of constructing a @code{struct foo} with a compound literal: 1626 1627@example 1628structure = ((struct foo) @{x + y, 'a', 0@}); 1629@end example 1630 1631@noindent 1632This is equivalent to writing the following: 1633 1634@example 1635@{ 1636 struct foo temp = @{x + y, 'a', 0@}; 1637 structure = temp; 1638@} 1639@end example 1640 1641You can also construct an array. If all the elements of the compound literal 1642are (made up of) simple constant expressions, suitable for use in 1643initializers of objects of static storage duration, then the compound 1644literal can be coerced to a pointer to its first element and used in 1645such an initializer, as shown here: 1646 1647@example 1648char **foo = (char *[]) @{ "x", "y", "z" @}; 1649@end example 1650 1651Compound literals for scalar types and union types are is 1652also allowed, but then the compound literal is equivalent 1653to a cast. 1654 1655As a GNU extension, GCC allows initialization of objects with static storage 1656duration by compound literals (which is not possible in ISO C99, because 1657the initializer is not a constant). 1658It is handled as if the object was initialized only with the bracket 1659enclosed list if compound literal's and object types match. 1660The initializer list of the compound literal must be constant. 1661If the object being initialized has array type of unknown size, the size is 1662determined by compound literal size. 1663 1664@example 1665static struct foo x = (struct foo) @{1, 'a', 'b'@}; 1666static int y[] = (int []) @{1, 2, 3@}; 1667static int z[] = (int [3]) @{1@}; 1668@end example 1669 1670@noindent 1671The above lines are equivalent to the following: 1672@example 1673static struct foo x = @{1, 'a', 'b'@}; 1674static int y[] = @{1, 2, 3@}; 1675static int z[] = @{1, 0, 0@}; 1676@end example 1677 1678@node Designated Inits 1679@section Designated Initializers 1680@cindex initializers with labeled elements 1681@cindex labeled elements in initializers 1682@cindex case labels in initializers 1683@cindex designated initializers 1684 1685Standard C89 requires the elements of an initializer to appear in a fixed 1686order, the same as the order of the elements in the array or structure 1687being initialized. 1688 1689In ISO C99 you can give the elements in any order, specifying the array 1690indices or structure field names they apply to, and GNU C allows this as 1691an extension in C89 mode as well. This extension is not 1692implemented in GNU C++. 1693 1694To specify an array index, write 1695@samp{[@var{index}] =} before the element value. For example, 1696 1697@example 1698int a[6] = @{ [4] = 29, [2] = 15 @}; 1699@end example 1700 1701@noindent 1702is equivalent to 1703 1704@example 1705int a[6] = @{ 0, 0, 15, 0, 29, 0 @}; 1706@end example 1707 1708@noindent 1709The index values must be constant expressions, even if the array being 1710initialized is automatic. 1711 1712An alternative syntax for this which has been obsolete since GCC 2.5 but 1713GCC still accepts is to write @samp{[@var{index}]} before the element 1714value, with no @samp{=}. 1715 1716To initialize a range of elements to the same value, write 1717@samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU 1718extension. For example, 1719 1720@example 1721int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @}; 1722@end example 1723 1724@noindent 1725If the value in it has side-effects, the side-effects will happen only once, 1726not for each initialized field by the range initializer. 1727 1728@noindent 1729Note that the length of the array is the highest value specified 1730plus one. 1731 1732In a structure initializer, specify the name of a field to initialize 1733with @samp{.@var{fieldname} =} before the element value. For example, 1734given the following structure, 1735 1736@example 1737struct point @{ int x, y; @}; 1738@end example 1739 1740@noindent 1741the following initialization 1742 1743@example 1744struct point p = @{ .y = yvalue, .x = xvalue @}; 1745@end example 1746 1747@noindent 1748is equivalent to 1749 1750@example 1751struct point p = @{ xvalue, yvalue @}; 1752@end example 1753 1754Another syntax which has the same meaning, obsolete since GCC 2.5, is 1755@samp{@var{fieldname}:}, as shown here: 1756 1757@example 1758struct point p = @{ y: yvalue, x: xvalue @}; 1759@end example 1760 1761@cindex designators 1762The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a 1763@dfn{designator}. You can also use a designator (or the obsolete colon 1764syntax) when initializing a union, to specify which element of the union 1765should be used. For example, 1766 1767@example 1768union foo @{ int i; double d; @}; 1769 1770union foo f = @{ .d = 4 @}; 1771@end example 1772 1773@noindent 1774will convert 4 to a @code{double} to store it in the union using 1775the second element. By contrast, casting 4 to type @code{union foo} 1776would store it into the union as the integer @code{i}, since it is 1777an integer. (@xref{Cast to Union}.) 1778 1779You can combine this technique of naming elements with ordinary C 1780initialization of successive elements. Each initializer element that 1781does not have a designator applies to the next consecutive element of the 1782array or structure. For example, 1783 1784@example 1785int a[6] = @{ [1] = v1, v2, [4] = v4 @}; 1786@end example 1787 1788@noindent 1789is equivalent to 1790 1791@example 1792int a[6] = @{ 0, v1, v2, 0, v4, 0 @}; 1793@end example 1794 1795Labeling the elements of an array initializer is especially useful 1796when the indices are characters or belong to an @code{enum} type. 1797For example: 1798 1799@example 1800int whitespace[256] 1801 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1, 1802 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @}; 1803@end example 1804 1805@cindex designator lists 1806You can also write a series of @samp{.@var{fieldname}} and 1807@samp{[@var{index}]} designators before an @samp{=} to specify a 1808nested subobject to initialize; the list is taken relative to the 1809subobject corresponding to the closest surrounding brace pair. For 1810example, with the @samp{struct point} declaration above: 1811 1812@smallexample 1813struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @}; 1814@end smallexample 1815 1816@noindent 1817If the same field is initialized multiple times, it will have value from 1818the last initialization. If any such overridden initialization has 1819side-effect, it is unspecified whether the side-effect happens or not. 1820Currently, gcc will discard them and issue a warning. 1821 1822@node Case Ranges 1823@section Case Ranges 1824@cindex case ranges 1825@cindex ranges in case statements 1826 1827You can specify a range of consecutive values in a single @code{case} label, 1828like this: 1829 1830@example 1831case @var{low} ... @var{high}: 1832@end example 1833 1834@noindent 1835This has the same effect as the proper number of individual @code{case} 1836labels, one for each integer value from @var{low} to @var{high}, inclusive. 1837 1838This feature is especially useful for ranges of ASCII character codes: 1839 1840@example 1841case 'A' ... 'Z': 1842@end example 1843 1844@strong{Be careful:} Write spaces around the @code{...}, for otherwise 1845it may be parsed wrong when you use it with integer values. For example, 1846write this: 1847 1848@example 1849case 1 ... 5: 1850@end example 1851 1852@noindent 1853rather than this: 1854 1855@example 1856case 1...5: 1857@end example 1858 1859@node Cast to Union 1860@section Cast to a Union Type 1861@cindex cast to a union 1862@cindex union, casting to a 1863 1864A cast to union type is similar to other casts, except that the type 1865specified is a union type. You can specify the type either with 1866@code{union @var{tag}} or with a typedef name. A cast to union is actually 1867a constructor though, not a cast, and hence does not yield an lvalue like 1868normal casts. (@xref{Compound Literals}.) 1869 1870The types that may be cast to the union type are those of the members 1871of the union. Thus, given the following union and variables: 1872 1873@example 1874union foo @{ int i; double d; @}; 1875int x; 1876double y; 1877@end example 1878 1879@noindent 1880both @code{x} and @code{y} can be cast to type @code{union foo}. 1881 1882Using the cast as the right-hand side of an assignment to a variable of 1883union type is equivalent to storing in a member of the union: 1884 1885@example 1886union foo u; 1887/* @r{@dots{}} */ 1888u = (union foo) x @equiv{} u.i = x 1889u = (union foo) y @equiv{} u.d = y 1890@end example 1891 1892You can also use the union cast as a function argument: 1893 1894@example 1895void hack (union foo); 1896/* @r{@dots{}} */ 1897hack ((union foo) x); 1898@end example 1899 1900@node Mixed Declarations 1901@section Mixed Declarations and Code 1902@cindex mixed declarations and code 1903@cindex declarations, mixed with code 1904@cindex code, mixed with declarations 1905 1906ISO C99 and ISO C++ allow declarations and code to be freely mixed 1907within compound statements. As an extension, GCC also allows this in 1908C89 mode. For example, you could do: 1909 1910@example 1911int i; 1912/* @r{@dots{}} */ 1913i++; 1914int j = i + 2; 1915@end example 1916 1917Each identifier is visible from where it is declared until the end of 1918the enclosing block. 1919 1920@node Function Attributes 1921@section Declaring Attributes of Functions 1922@cindex function attributes 1923@cindex declaring attributes of functions 1924@cindex functions that never return 1925@cindex functions that have no side effects 1926@cindex functions in arbitrary sections 1927@cindex functions that behave like malloc 1928@cindex @code{volatile} applied to function 1929@cindex @code{const} applied to function 1930@cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments 1931@cindex functions with non-null pointer arguments 1932@cindex functions that are passed arguments in registers on the 386 1933@cindex functions that pop the argument stack on the 386 1934@cindex functions that do not pop the argument stack on the 386 1935 1936In GNU C, you declare certain things about functions called in your program 1937which help the compiler optimize function calls and check your code more 1938carefully. 1939 1940The keyword @code{__attribute__} allows you to specify special 1941attributes when making a declaration. This keyword is followed by an 1942attribute specification inside double parentheses. The following 1943attributes are currently defined for functions on all targets: 1944@code{noreturn}, @code{noinline}, @code{always_inline}, 1945@code{pure}, @code{const}, @code{nothrow}, 1946@code{format}, @code{format_arg}, @code{no_instrument_function}, 1947@code{section}, @code{constructor}, @code{destructor}, @code{used}, 1948@code{unused}, @code{deprecated}, @code{weak}, @code{malloc}, 1949@code{alias}, and @code{nonnull}. Several other attributes are defined 1950for functions on particular target systems. Other attributes, including 1951@code{section} are supported for variables declarations 1952(@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}). 1953 1954You may also specify attributes with @samp{__} preceding and following 1955each keyword. This allows you to use them in header files without 1956being concerned about a possible macro of the same name. For example, 1957you may use @code{__noreturn__} instead of @code{noreturn}. 1958 1959@xref{Attribute Syntax}, for details of the exact syntax for using 1960attributes. 1961 1962@table @code 1963@cindex @code{noreturn} function attribute 1964@item noreturn 1965A few standard library functions, such as @code{abort} and @code{exit}, 1966cannot return. GCC knows this automatically. Some programs define 1967their own functions that never return. You can declare them 1968@code{noreturn} to tell the compiler this fact. For example, 1969 1970@smallexample 1971@group 1972void fatal () __attribute__ ((noreturn)); 1973 1974void 1975fatal (/* @r{@dots{}} */) 1976@{ 1977 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */ 1978 exit (1); 1979@} 1980@end group 1981@end smallexample 1982 1983The @code{noreturn} keyword tells the compiler to assume that 1984@code{fatal} cannot return. It can then optimize without regard to what 1985would happen if @code{fatal} ever did return. This makes slightly 1986better code. More importantly, it helps avoid spurious warnings of 1987uninitialized variables. 1988 1989Do not assume that registers saved by the calling function are 1990restored before calling the @code{noreturn} function. 1991 1992It does not make sense for a @code{noreturn} function to have a return 1993type other than @code{void}. 1994 1995The attribute @code{noreturn} is not implemented in GCC versions 1996earlier than 2.5. An alternative way to declare that a function does 1997not return, which works in the current version and in some older 1998versions, is as follows: 1999 2000@smallexample 2001typedef void voidfn (); 2002 2003volatile voidfn fatal; 2004@end smallexample 2005 2006@cindex @code{noinline} function attribute 2007@item noinline 2008This function attribute prevents a function from being considered for 2009inlining. 2010 2011@cindex @code{always_inline} function attribute 2012@item always_inline 2013Generally, functions are not inlined unless optimization is specified. 2014For functions declared inline, this attribute inlines the function even 2015if no optimization level was specified. 2016 2017@cindex @code{pure} function attribute 2018@item pure 2019Many functions have no effects except the return value and their 2020return value depends only on the parameters and/or global variables. 2021Such a function can be subject 2022to common subexpression elimination and loop optimization just as an 2023arithmetic operator would be. These functions should be declared 2024with the attribute @code{pure}. For example, 2025 2026@smallexample 2027int square (int) __attribute__ ((pure)); 2028@end smallexample 2029 2030@noindent 2031says that the hypothetical function @code{square} is safe to call 2032fewer times than the program says. 2033 2034Some of common examples of pure functions are @code{strlen} or @code{memcmp}. 2035Interesting non-pure functions are functions with infinite loops or those 2036depending on volatile memory or other system resource, that may change between 2037two consecutive calls (such as @code{feof} in a multithreading environment). 2038 2039The attribute @code{pure} is not implemented in GCC versions earlier 2040than 2.96. 2041@cindex @code{const} function attribute 2042@item const 2043Many functions do not examine any values except their arguments, and 2044have no effects except the return value. Basically this is just slightly 2045more strict class than the @code{pure} attribute above, since function is not 2046allowed to read global memory. 2047 2048@cindex pointer arguments 2049Note that a function that has pointer arguments and examines the data 2050pointed to must @emph{not} be declared @code{const}. Likewise, a 2051function that calls a non-@code{const} function usually must not be 2052@code{const}. It does not make sense for a @code{const} function to 2053return @code{void}. 2054 2055The attribute @code{const} is not implemented in GCC versions earlier 2056than 2.5. An alternative way to declare that a function has no side 2057effects, which works in the current version and in some older versions, 2058is as follows: 2059 2060@smallexample 2061typedef int intfn (); 2062 2063extern const intfn square; 2064@end smallexample 2065 2066This approach does not work in GNU C++ from 2.6.0 on, since the language 2067specifies that the @samp{const} must be attached to the return value. 2068 2069@cindex @code{nothrow} function attribute 2070@item nothrow 2071The @code{nothrow} attribute is used to inform the compiler that a 2072function cannot throw an exception. For example, most functions in 2073the standard C library can be guaranteed not to throw an exception 2074with the notable exceptions of @code{qsort} and @code{bsearch} that 2075take function pointer arguments. The @code{nothrow} attribute is not 2076implemented in GCC versions earlier than 3.2. 2077 2078@item format (@var{archetype}, @var{string-index}, @var{first-to-check}) 2079@cindex @code{format} function attribute 2080@opindex Wformat 2081The @code{format} attribute specifies that a function takes @code{printf}, 2082@code{scanf}, @code{strftime} or @code{strfmon} style arguments which 2083should be type-checked against a format string. For example, the 2084declaration: 2085 2086@smallexample 2087extern int 2088my_printf (void *my_object, const char *my_format, ...) 2089 __attribute__ ((format (printf, 2, 3))); 2090@end smallexample 2091 2092@noindent 2093causes the compiler to check the arguments in calls to @code{my_printf} 2094for consistency with the @code{printf} style format string argument 2095@code{my_format}. 2096 2097The parameter @var{archetype} determines how the format string is 2098interpreted, and should be @code{printf}, @code{scanf}, @code{strftime} 2099or @code{strfmon}. (You can also use @code{__printf__}, 2100@code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The 2101parameter @var{string-index} specifies which argument is the format 2102string argument (starting from 1), while @var{first-to-check} is the 2103number of the first argument to check against the format string. For 2104functions where the arguments are not available to be checked (such as 2105@code{vprintf}), specify the third parameter as zero. In this case the 2106compiler only checks the format string for consistency. For 2107@code{strftime} formats, the third parameter is required to be zero. 2108Since non-static C++ methods have an implicit @code{this} argument, the 2109arguments of such methods should be counted from two, not one, when 2110giving values for @var{string-index} and @var{first-to-check}. 2111 2112In the example above, the format string (@code{my_format}) is the second 2113argument of the function @code{my_print}, and the arguments to check 2114start with the third argument, so the correct parameters for the format 2115attribute are 2 and 3. 2116 2117@opindex ffreestanding 2118The @code{format} attribute allows you to identify your own functions 2119which take format strings as arguments, so that GCC can check the 2120calls to these functions for errors. The compiler always (unless 2121@option{-ffreestanding} is used) checks formats 2122for the standard library functions @code{printf}, @code{fprintf}, 2123@code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime}, 2124@code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such 2125warnings are requested (using @option{-Wformat}), so there is no need to 2126modify the header file @file{stdio.h}. In C99 mode, the functions 2127@code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and 2128@code{vsscanf} are also checked. Except in strictly conforming C 2129standard modes, the X/Open function @code{strfmon} is also checked as 2130are @code{printf_unlocked} and @code{fprintf_unlocked}. 2131@xref{C Dialect Options,,Options Controlling C Dialect}. 2132 2133@item format_arg (@var{string-index}) 2134@cindex @code{format_arg} function attribute 2135@opindex Wformat-nonliteral 2136The @code{format_arg} attribute specifies that a function takes a format 2137string for a @code{printf}, @code{scanf}, @code{strftime} or 2138@code{strfmon} style function and modifies it (for example, to translate 2139it into another language), so the result can be passed to a 2140@code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style 2141function (with the remaining arguments to the format function the same 2142as they would have been for the unmodified string). For example, the 2143declaration: 2144 2145@smallexample 2146extern char * 2147my_dgettext (char *my_domain, const char *my_format) 2148 __attribute__ ((format_arg (2))); 2149@end smallexample 2150 2151@noindent 2152causes the compiler to check the arguments in calls to a @code{printf}, 2153@code{scanf}, @code{strftime} or @code{strfmon} type function, whose 2154format string argument is a call to the @code{my_dgettext} function, for 2155consistency with the format string argument @code{my_format}. If the 2156@code{format_arg} attribute had not been specified, all the compiler 2157could tell in such calls to format functions would be that the format 2158string argument is not constant; this would generate a warning when 2159@option{-Wformat-nonliteral} is used, but the calls could not be checked 2160without the attribute. 2161 2162The parameter @var{string-index} specifies which argument is the format 2163string argument (starting from one). Since non-static C++ methods have 2164an implicit @code{this} argument, the arguments of such methods should 2165be counted from two. 2166 2167The @code{format-arg} attribute allows you to identify your own 2168functions which modify format strings, so that GCC can check the 2169calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} 2170type function whose operands are a call to one of your own function. 2171The compiler always treats @code{gettext}, @code{dgettext}, and 2172@code{dcgettext} in this manner except when strict ISO C support is 2173requested by @option{-ansi} or an appropriate @option{-std} option, or 2174@option{-ffreestanding} is used. @xref{C Dialect Options,,Options 2175Controlling C Dialect}. 2176 2177@item nonnull (@var{arg-index}, @dots{}) 2178@cindex @code{nonnull} function attribute 2179The @code{nonnull} attribute specifies that some function parameters should 2180be non-null pointers. For instance, the declaration: 2181 2182@smallexample 2183extern void * 2184my_memcpy (void *dest, const void *src, size_t len) 2185 __attribute__((nonnull (1, 2))); 2186@end smallexample 2187 2188@noindent 2189causes the compiler to check that, in calls to @code{my_memcpy}, 2190arguments @var{dest} and @var{src} are non-null. If the compiler 2191determines that a null pointer is passed in an argument slot marked 2192as non-null, and the @option{-Wnonnull} option is enabled, a warning 2193is issued. The compiler may also choose to make optimizations based 2194on the knowledge that certain function arguments will not be null. 2195 2196If no argument index list is given to the @code{nonnull} attribute, 2197all pointer arguments are marked as non-null. To illustrate, the 2198following declaration is equivalent to the previous example: 2199 2200@smallexample 2201extern void * 2202my_memcpy (void *dest, const void *src, size_t len) 2203 __attribute__((nonnull)); 2204@end smallexample 2205 2206@item no_instrument_function 2207@cindex @code{no_instrument_function} function attribute 2208@opindex finstrument-functions 2209If @option{-finstrument-functions} is given, profiling function calls will 2210be generated at entry and exit of most user-compiled functions. 2211Functions with this attribute will not be so instrumented. 2212 2213@item section ("@var{section-name}") 2214@cindex @code{section} function attribute 2215Normally, the compiler places the code it generates in the @code{text} section. 2216Sometimes, however, you need additional sections, or you need certain 2217particular functions to appear in special sections. The @code{section} 2218attribute specifies that a function lives in a particular section. 2219For example, the declaration: 2220 2221@smallexample 2222extern void foobar (void) __attribute__ ((section ("bar"))); 2223@end smallexample 2224 2225@noindent 2226puts the function @code{foobar} in the @code{bar} section. 2227 2228Some file formats do not support arbitrary sections so the @code{section} 2229attribute is not available on all platforms. 2230If you need to map the entire contents of a module to a particular 2231section, consider using the facilities of the linker instead. 2232 2233@item constructor 2234@itemx destructor 2235@cindex @code{constructor} function attribute 2236@cindex @code{destructor} function attribute 2237The @code{constructor} attribute causes the function to be called 2238automatically before execution enters @code{main ()}. Similarly, the 2239@code{destructor} attribute causes the function to be called 2240automatically after @code{main ()} has completed or @code{exit ()} has 2241been called. Functions with these attributes are useful for 2242initializing data that will be used implicitly during the execution of 2243the program. 2244 2245These attributes are not currently implemented for Objective-C@. 2246 2247@cindex @code{unused} attribute. 2248@item unused 2249This attribute, attached to a function, means that the function is meant 2250to be possibly unused. GCC will not produce a warning for this 2251function. GNU C++ does not currently support this attribute as 2252definitions without parameters are valid in C++. 2253 2254@cindex @code{used} attribute. 2255@item used 2256This attribute, attached to a function, means that code must be emitted 2257for the function even if it appears that the function is not referenced. 2258This is useful, for example, when the function is referenced only in 2259inline assembly. 2260 2261@cindex @code{deprecated} attribute. 2262@item deprecated 2263The @code{deprecated} attribute results in a warning if the function 2264is used anywhere in the source file. This is useful when identifying 2265functions that are expected to be removed in a future version of a 2266program. The warning also includes the location of the declaration 2267of the deprecated function, to enable users to easily find further 2268information about why the function is deprecated, or what they should 2269do instead. Note that the warnings only occurs for uses: 2270 2271@smallexample 2272int old_fn () __attribute__ ((deprecated)); 2273int old_fn (); 2274int (*fn_ptr)() = old_fn; 2275@end smallexample 2276 2277results in a warning on line 3 but not line 2. 2278 2279The @code{deprecated} attribute can also be used for variables and 2280types (@pxref{Variable Attributes}, @pxref{Type Attributes}.) 2281 2282@item weak 2283@cindex @code{weak} attribute 2284The @code{weak} attribute causes the declaration to be emitted as a weak 2285symbol rather than a global. This is primarily useful in defining 2286library functions which can be overridden in user code, though it can 2287also be used with non-function declarations. Weak symbols are supported 2288for ELF targets, and also for a.out targets when using the GNU assembler 2289and linker. 2290 2291@item malloc 2292@cindex @code{malloc} attribute 2293The @code{malloc} attribute is used to tell the compiler that a function 2294may be treated as if it were the malloc function. The compiler assumes 2295that calls to malloc result in pointers that cannot alias anything. 2296This will often improve optimization. 2297 2298@item alias ("@var{target}") 2299@cindex @code{alias} attribute 2300The @code{alias} attribute causes the declaration to be emitted as an 2301alias for another symbol, which must be specified. For instance, 2302 2303@smallexample 2304void __f () @{ /* @r{Do something.} */; @} 2305void f () __attribute__ ((weak, alias ("__f"))); 2306@end smallexample 2307 2308declares @samp{f} to be a weak alias for @samp{__f}. In C++, the 2309mangled name for the target must be used. 2310 2311Not all target machines support this attribute. 2312 2313@item visibility ("@var{visibility_type}") 2314@cindex @code{visibility} attribute 2315The @code{visibility} attribute on ELF targets causes the declaration 2316to be emitted with default, hidden, protected or internal visibility. 2317 2318@smallexample 2319void __attribute__ ((visibility ("protected"))) 2320f () @{ /* @r{Do something.} */; @} 2321int i __attribute__ ((visibility ("hidden"))); 2322@end smallexample 2323 2324See the ELF gABI for complete details, but the short story is: 2325 2326@table @dfn 2327@item default 2328Default visibility is the normal case for ELF. This value is 2329available for the visibility attribute to override other options 2330that may change the assumed visibility of symbols. 2331 2332@item hidden 2333Hidden visibility indicates that the symbol will not be placed into 2334the dynamic symbol table, so no other @dfn{module} (executable or 2335shared library) can reference it directly. 2336 2337@item protected 2338Protected visibility indicates that the symbol will be placed in the 2339dynamic symbol table, but that references within the defining module 2340will bind to the local symbol. That is, the symbol cannot be overridden 2341by another module. 2342 2343@item internal 2344Internal visibility is like hidden visibility, but with additional 2345processor specific semantics. Unless otherwise specified by the psABI, 2346gcc defines internal visibility to mean that the function is @emph{never} 2347called from another module. Note that hidden symbols, while they cannot 2348be referenced directly by other modules, can be referenced indirectly via 2349function pointers. By indicating that a symbol cannot be called from 2350outside the module, gcc may for instance omit the load of a PIC register 2351since it is known that the calling function loaded the correct value. 2352@end table 2353 2354Not all ELF targets support this attribute. 2355 2356@item regparm (@var{number}) 2357@cindex @code{regparm} attribute 2358@cindex functions that are passed arguments in registers on the 386 2359On the Intel 386, the @code{regparm} attribute causes the compiler to 2360pass up to @var{number} integer arguments in registers EAX, 2361EDX, and ECX instead of on the stack. Functions that take a 2362variable number of arguments will continue to be passed all of their 2363arguments on the stack. 2364 2365Beware that on some ELF systems this attribute is unsuitable for 2366global functions in shared libraries with lazy binding (which is the 2367default). Lazy binding will send the first call via resolving code in 2368the loader, which might assume EAX, EDX and ECX can be clobbered, as 2369per the standard calling conventions. Solaris 8 is affected by this. 2370GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be 2371safe since the loaders there save all registers. (Lazy binding can be 2372disabled with the linker or the loader if desired, to avoid the 2373problem.) 2374 2375@item stdcall 2376@cindex functions that pop the argument stack on the 386 2377On the Intel 386, the @code{stdcall} attribute causes the compiler to 2378assume that the called function will pop off the stack space used to 2379pass arguments, unless it takes a variable number of arguments. 2380 2381 2382@item cdecl 2383@cindex functions that do pop the argument stack on the 386 2384@opindex mrtd 2385On the Intel 386, the @code{cdecl} attribute causes the compiler to 2386assume that the calling function will pop off the stack space used to 2387pass arguments. This is 2388useful to override the effects of the @option{-mrtd} switch. 2389 2390@item longcall/shortcall 2391@cindex functions called via pointer on the RS/6000 and PowerPC 2392On the RS/6000 and PowerPC, the @code{longcall} attribute causes the 2393compiler to always call this function via a pointer, just as it would if 2394the @option{-mlongcall} option had been specified. The @code{shortcall} 2395attribute causes the compiler not to do this. These attributes override 2396both the @option{-mlongcall} switch and the @code{#pragma longcall} 2397setting. 2398 2399@xref{RS/6000 and PowerPC Options}, for more information on whether long 2400calls are necessary. 2401 2402@item long_call/short_call 2403@cindex indirect calls on ARM 2404This attribute specifies how a particular function is called on 2405ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options}) 2406command line switch and @code{#pragma long_calls} settings. The 2407@code{long_call} attribute causes the compiler to always call the 2408function by first loading its address into a register and then using the 2409contents of that register. The @code{short_call} attribute always places 2410the offset to the function from the call site into the @samp{BL} 2411instruction directly. 2412 2413@item function_vector 2414@cindex calling functions through the function vector on the H8/300 processors 2415Use this attribute on the H8/300 and H8/300H to indicate that the specified 2416function should be called through the function vector. Calling a 2417function through the function vector will reduce code size, however; 2418the function vector has a limited size (maximum 128 entries on the H8/300 2419and 64 entries on the H8/300H) and shares space with the interrupt vector. 2420 2421You must use GAS and GLD from GNU binutils version 2.7 or later for 2422this attribute to work correctly. 2423 2424@item interrupt 2425@cindex interrupt handler functions 2426Use this attribute on the ARM, AVR, M32R/D and Xstormy16 ports to indicate 2427that the specified function is an interrupt handler. The compiler will 2428generate function entry and exit sequences suitable for use in an 2429interrupt handler when this attribute is present. 2430 2431Note, interrupt handlers for the H8/300, H8/300H and SH processors can 2432be specified via the @code{interrupt_handler} attribute. 2433 2434Note, on the AVR, interrupts will be enabled inside the function. 2435 2436Note, for the ARM, you can specify the kind of interrupt to be handled by 2437adding an optional parameter to the interrupt attribute like this: 2438 2439@smallexample 2440void f () __attribute__ ((interrupt ("IRQ"))); 2441@end smallexample 2442 2443Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@. 2444 2445@item interrupt_handler 2446@cindex interrupt handler functions on the H8/300 and SH processors 2447Use this attribute on the H8/300, H8/300H and SH to indicate that the 2448specified function is an interrupt handler. The compiler will generate 2449function entry and exit sequences suitable for use in an interrupt 2450handler when this attribute is present. 2451 2452@item sp_switch 2453Use this attribute on the SH to indicate an @code{interrupt_handler} 2454function should switch to an alternate stack. It expects a string 2455argument that names a global variable holding the address of the 2456alternate stack. 2457 2458@smallexample 2459void *alt_stack; 2460void f () __attribute__ ((interrupt_handler, 2461 sp_switch ("alt_stack"))); 2462@end smallexample 2463 2464@item trap_exit 2465Use this attribute on the SH for an @code{interrupt_handle} to return using 2466@code{trapa} instead of @code{rte}. This attribute expects an integer 2467argument specifying the trap number to be used. 2468 2469@item eightbit_data 2470@cindex eight bit data on the H8/300 and H8/300H 2471Use this attribute on the H8/300 and H8/300H to indicate that the specified 2472variable should be placed into the eight bit data section. 2473The compiler will generate more efficient code for certain operations 2474on data in the eight bit data area. Note the eight bit data area is limited to 2475256 bytes of data. 2476 2477You must use GAS and GLD from GNU binutils version 2.7 or later for 2478this attribute to work correctly. 2479 2480@item tiny_data 2481@cindex tiny data section on the H8/300H 2482Use this attribute on the H8/300H to indicate that the specified 2483variable should be placed into the tiny data section. 2484The compiler will generate more efficient code for loads and stores 2485on data in the tiny data section. Note the tiny data area is limited to 2486slightly under 32kbytes of data. 2487 2488@item signal 2489@cindex signal handler functions on the AVR processors 2490Use this attribute on the AVR to indicate that the specified 2491function is a signal handler. The compiler will generate function 2492entry and exit sequences suitable for use in a signal handler when this 2493attribute is present. Interrupts will be disabled inside the function. 2494 2495@item naked 2496@cindex function without a prologue/epilogue code 2497Use this attribute on the ARM, AVR and IP2K ports to indicate that the 2498specified function does not need prologue/epilogue sequences generated by 2499the compiler. It is up to the programmer to provide these sequences. 2500 2501@item model (@var{model-name}) 2502@cindex function addressability on the M32R/D 2503Use this attribute on the M32R/D to set the addressability of an object, 2504and of the code generated for a function. 2505The identifier @var{model-name} is one of @code{small}, @code{medium}, 2506or @code{large}, representing each of the code models. 2507 2508Small model objects live in the lower 16MB of memory (so that their 2509addresses can be loaded with the @code{ld24} instruction), and are 2510callable with the @code{bl} instruction. 2511 2512Medium model objects may live anywhere in the 32-bit address space (the 2513compiler will generate @code{seth/add3} instructions to load their addresses), 2514and are callable with the @code{bl} instruction. 2515 2516Large model objects may live anywhere in the 32-bit address space (the 2517compiler will generate @code{seth/add3} instructions to load their addresses), 2518and may not be reachable with the @code{bl} instruction (the compiler will 2519generate the much slower @code{seth/add3/jl} instruction sequence). 2520 2521@item far 2522@cindex functions which handle memory bank switching 2523On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to 2524use a calling convention that takes care of switching memory banks when 2525entering and leaving a function. This calling convention is also the 2526default when using the @option{-mlong-calls} option. 2527 2528On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions 2529to call and return from a function. 2530 2531On 68HC11 the compiler will generate a sequence of instructions 2532to invoke a board-specific routine to switch the memory bank and call the 2533real function. The board-specific routine simulates a @code{call}. 2534At the end of a function, it will jump to a board-specific routine 2535instead of using @code{rts}. The board-specific return routine simulates 2536the @code{rtc}. 2537 2538@item near 2539@cindex functions which do not handle memory bank switching on 68HC11/68HC12 2540On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to 2541use the normal calling convention based on @code{jsr} and @code{rts}. 2542This attribute can be used to cancel the effect of the @option{-mlong-calls} 2543option. 2544 2545@item dllimport 2546@cindex @code{__declspec(dllimport)} 2547On Windows targets, the @code{dllimport} attribute causes the compiler 2548to reference a function or variable via a global pointer to a pointer 2549that is set up by the Windows dll library. The pointer name is formed by 2550combining @code{_imp__} and the function or variable name. The attribute 2551implies @code{extern} storage. 2552 2553Currently, the attribute is ignored for inlined functions. If the 2554attribute is applied to a symbol @emph{definition}, an error is reported. 2555If a symbol previously declared @code{dllimport} is later defined, the 2556attribute is ignored in subsequent references, and a warning is emitted. 2557The attribute is also overridden by a subsequent declaration as 2558@code{dllexport}. 2559 2560When applied to C++ classes, the attribute marks non-inlined 2561member functions and static data members as imports. However, the 2562attribute is ignored for virtual methods to allow creation of vtables 2563using thunks. 2564 2565On cygwin, mingw and arm-pe targets, @code{__declspec(dllimport)} is 2566recognized as a synonym for @code{__attribute__ ((dllimport))} for 2567compatibility with other Windows compilers. 2568 2569The use of the @code{dllimport} attribute on functions is not necessary, 2570but provides a small performance benefit by eliminating a thunk in the 2571dll. The use of the @code{dllimport} attribute on imported variables was 2572required on older versions of GNU ld, but can now be avoided by passing 2573the @option{--enable-auto-import} switch to ld. As with functions, using 2574the attribute for a variable eliminates a thunk in the dll. 2575 2576One drawback to using this attribute is that a pointer to a function or 2577variable marked as dllimport cannot be used as a constant address. The 2578attribute can be disabled for functions by setting the 2579@option{-mnop-fun-dllimport} flag. 2580 2581@item dllexport 2582@cindex @code{__declspec(dllexport)} 2583On Windows targets the @code{dllexport} attribute causes the compiler to 2584provide a global pointer to a pointer in a dll, so that it can be 2585referenced with the @code{dllimport} attribute. The pointer name is 2586formed by combining @code{_imp__} and the function or variable name. 2587 2588Currently, the @code{dllexport}attribute is ignored for inlined 2589functions, but export can be forced by using the 2590@option{-fkeep-inline-functions} flag. The attribute is also ignored for 2591undefined symbols. 2592 2593When applied to C++ classes. the attribute marks defined non-inlined 2594member functions and static data members as exports. Static consts 2595initialized in-class are not marked unless they are also defined 2596out-of-class. 2597 2598On cygwin, mingw and arm-pe targets, @code{__declspec(dllexport)} is 2599recognized as a synonym for @code{__attribute__ ((dllexport))} for 2600compatibility with other Windows compilers. 2601 2602Alternative methods for including the symbol in the dll's export table 2603are to use a .def file with an @code{EXPORTS} section or, with GNU ld, 2604using the @option{--export-all} linker flag. 2605 2606@end table 2607 2608You can specify multiple attributes in a declaration by separating them 2609by commas within the double parentheses or by immediately following an 2610attribute declaration with another attribute declaration. 2611 2612@cindex @code{#pragma}, reason for not using 2613@cindex pragma, reason for not using 2614Some people object to the @code{__attribute__} feature, suggesting that 2615ISO C's @code{#pragma} should be used instead. At the time 2616@code{__attribute__} was designed, there were two reasons for not doing 2617this. 2618 2619@enumerate 2620@item 2621It is impossible to generate @code{#pragma} commands from a macro. 2622 2623@item 2624There is no telling what the same @code{#pragma} might mean in another 2625compiler. 2626@end enumerate 2627 2628These two reasons applied to almost any application that might have been 2629proposed for @code{#pragma}. It was basically a mistake to use 2630@code{#pragma} for @emph{anything}. 2631 2632The ISO C99 standard includes @code{_Pragma}, which now allows pragmas 2633to be generated from macros. In addition, a @code{#pragma GCC} 2634namespace is now in use for GCC-specific pragmas. However, it has been 2635found convenient to use @code{__attribute__} to achieve a natural 2636attachment of attributes to their corresponding declarations, whereas 2637@code{#pragma GCC} is of use for constructs that do not naturally form 2638part of the grammar. @xref{Other Directives,,Miscellaneous 2639Preprocessing Directives, cpp, The GNU C Preprocessor}. 2640 2641@node Attribute Syntax 2642@section Attribute Syntax 2643@cindex attribute syntax 2644 2645This section describes the syntax with which @code{__attribute__} may be 2646used, and the constructs to which attribute specifiers bind, for the C 2647language. Some details may vary for C++ and Objective-C@. Because of 2648infelicities in the grammar for attributes, some forms described here 2649may not be successfully parsed in all cases. 2650 2651There are some problems with the semantics of attributes in C++. For 2652example, there are no manglings for attributes, although they may affect 2653code generation, so problems may arise when attributed types are used in 2654conjunction with templates or overloading. Similarly, @code{typeid} 2655does not distinguish between types with different attributes. Support 2656for attributes in C++ may be restricted in future to attributes on 2657declarations only, but not on nested declarators. 2658 2659@xref{Function Attributes}, for details of the semantics of attributes 2660applying to functions. @xref{Variable Attributes}, for details of the 2661semantics of attributes applying to variables. @xref{Type Attributes}, 2662for details of the semantics of attributes applying to structure, union 2663and enumerated types. 2664 2665An @dfn{attribute specifier} is of the form 2666@code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list} 2667is a possibly empty comma-separated sequence of @dfn{attributes}, where 2668each attribute is one of the following: 2669 2670@itemize @bullet 2671@item 2672Empty. Empty attributes are ignored. 2673 2674@item 2675A word (which may be an identifier such as @code{unused}, or a reserved 2676word such as @code{const}). 2677 2678@item 2679A word, followed by, in parentheses, parameters for the attribute. 2680These parameters take one of the following forms: 2681 2682@itemize @bullet 2683@item 2684An identifier. For example, @code{mode} attributes use this form. 2685 2686@item 2687An identifier followed by a comma and a non-empty comma-separated list 2688of expressions. For example, @code{format} attributes use this form. 2689 2690@item 2691A possibly empty comma-separated list of expressions. For example, 2692@code{format_arg} attributes use this form with the list being a single 2693integer constant expression, and @code{alias} attributes use this form 2694with the list being a single string constant. 2695@end itemize 2696@end itemize 2697 2698An @dfn{attribute specifier list} is a sequence of one or more attribute 2699specifiers, not separated by any other tokens. 2700 2701An attribute specifier list may appear after the colon following a 2702label, other than a @code{case} or @code{default} label. The only 2703attribute it makes sense to use after a label is @code{unused}. This 2704feature is intended for code generated by programs which contains labels 2705that may be unused but which is compiled with @option{-Wall}. It would 2706not normally be appropriate to use in it human-written code, though it 2707could be useful in cases where the code that jumps to the label is 2708contained within an @code{#ifdef} conditional. 2709 2710An attribute specifier list may appear as part of a @code{struct}, 2711@code{union} or @code{enum} specifier. It may go either immediately 2712after the @code{struct}, @code{union} or @code{enum} keyword, or after 2713the closing brace. It is ignored if the content of the structure, union 2714or enumerated type is not defined in the specifier in which the 2715attribute specifier list is used---that is, in usages such as 2716@code{struct __attribute__((foo)) bar} with no following opening brace. 2717Where attribute specifiers follow the closing brace, they are considered 2718to relate to the structure, union or enumerated type defined, not to any 2719enclosing declaration the type specifier appears in, and the type 2720defined is not complete until after the attribute specifiers. 2721@c Otherwise, there would be the following problems: a shift/reduce 2722@c conflict between attributes binding the struct/union/enum and 2723@c binding to the list of specifiers/qualifiers; and "aligned" 2724@c attributes could use sizeof for the structure, but the size could be 2725@c changed later by "packed" attributes. 2726 2727Otherwise, an attribute specifier appears as part of a declaration, 2728counting declarations of unnamed parameters and type names, and relates 2729to that declaration (which may be nested in another declaration, for 2730example in the case of a parameter declaration), or to a particular declarator 2731within a declaration. Where an 2732attribute specifier is applied to a parameter declared as a function or 2733an array, it should apply to the function or array rather than the 2734pointer to which the parameter is implicitly converted, but this is not 2735yet correctly implemented. 2736 2737Any list of specifiers and qualifiers at the start of a declaration may 2738contain attribute specifiers, whether or not such a list may in that 2739context contain storage class specifiers. (Some attributes, however, 2740are essentially in the nature of storage class specifiers, and only make 2741sense where storage class specifiers may be used; for example, 2742@code{section}.) There is one necessary limitation to this syntax: the 2743first old-style parameter declaration in a function definition cannot 2744begin with an attribute specifier, because such an attribute applies to 2745the function instead by syntax described below (which, however, is not 2746yet implemented in this case). In some other cases, attribute 2747specifiers are permitted by this grammar but not yet supported by the 2748compiler. All attribute specifiers in this place relate to the 2749declaration as a whole. In the obsolescent usage where a type of 2750@code{int} is implied by the absence of type specifiers, such a list of 2751specifiers and qualifiers may be an attribute specifier list with no 2752other specifiers or qualifiers. 2753 2754An attribute specifier list may appear immediately before a declarator 2755(other than the first) in a comma-separated list of declarators in a 2756declaration of more than one identifier using a single list of 2757specifiers and qualifiers. Such attribute specifiers apply 2758only to the identifier before whose declarator they appear. For 2759example, in 2760 2761@smallexample 2762__attribute__((noreturn)) void d0 (void), 2763 __attribute__((format(printf, 1, 2))) d1 (const char *, ...), 2764 d2 (void) 2765@end smallexample 2766 2767@noindent 2768the @code{noreturn} attribute applies to all the functions 2769declared; the @code{format} attribute only applies to @code{d1}. 2770 2771An attribute specifier list may appear immediately before the comma, 2772@code{=} or semicolon terminating the declaration of an identifier other 2773than a function definition. At present, such attribute specifiers apply 2774to the declared object or function, but in future they may attach to the 2775outermost adjacent declarator. In simple cases there is no difference, 2776but, for example, in 2777 2778@smallexample 2779void (****f)(void) __attribute__((noreturn)); 2780@end smallexample 2781 2782@noindent 2783at present the @code{noreturn} attribute applies to @code{f}, which 2784causes a warning since @code{f} is not a function, but in future it may 2785apply to the function @code{****f}. The precise semantics of what 2786attributes in such cases will apply to are not yet specified. Where an 2787assembler name for an object or function is specified (@pxref{Asm 2788Labels}), at present the attribute must follow the @code{asm} 2789specification; in future, attributes before the @code{asm} specification 2790may apply to the adjacent declarator, and those after it to the declared 2791object or function. 2792 2793An attribute specifier list may, in future, be permitted to appear after 2794the declarator in a function definition (before any old-style parameter 2795declarations or the function body). 2796 2797Attribute specifiers may be mixed with type qualifiers appearing inside 2798the @code{[]} of a parameter array declarator, in the C99 construct by 2799which such qualifiers are applied to the pointer to which the array is 2800implicitly converted. Such attribute specifiers apply to the pointer, 2801not to the array, but at present this is not implemented and they are 2802ignored. 2803 2804An attribute specifier list may appear at the start of a nested 2805declarator. At present, there are some limitations in this usage: the 2806attributes correctly apply to the declarator, but for most individual 2807attributes the semantics this implies are not implemented. 2808When attribute specifiers follow the @code{*} of a pointer 2809declarator, they may be mixed with any type qualifiers present. 2810The following describes the formal semantics of this syntax. It will make the 2811most sense if you are familiar with the formal specification of 2812declarators in the ISO C standard. 2813 2814Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T 2815D1}, where @code{T} contains declaration specifiers that specify a type 2816@var{Type} (such as @code{int}) and @code{D1} is a declarator that 2817contains an identifier @var{ident}. The type specified for @var{ident} 2818for derived declarators whose type does not include an attribute 2819specifier is as in the ISO C standard. 2820 2821If @code{D1} has the form @code{( @var{attribute-specifier-list} D )}, 2822and the declaration @code{T D} specifies the type 2823``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 2824@code{T D1} specifies the type ``@var{derived-declarator-type-list} 2825@var{attribute-specifier-list} @var{Type}'' for @var{ident}. 2826 2827If @code{D1} has the form @code{* 2828@var{type-qualifier-and-attribute-specifier-list} D}, and the 2829declaration @code{T D} specifies the type 2830``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 2831@code{T D1} specifies the type ``@var{derived-declarator-type-list} 2832@var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for 2833@var{ident}. 2834 2835For example, 2836 2837@smallexample 2838void (__attribute__((noreturn)) ****f) (void); 2839@end smallexample 2840 2841@noindent 2842specifies the type ``pointer to pointer to pointer to pointer to 2843non-returning function returning @code{void}''. As another example, 2844 2845@smallexample 2846char *__attribute__((aligned(8))) *f; 2847@end smallexample 2848 2849@noindent 2850specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''. 2851Note again that this does not work with most attributes; for example, 2852the usage of @samp{aligned} and @samp{noreturn} attributes given above 2853is not yet supported. 2854 2855For compatibility with existing code written for compiler versions that 2856did not implement attributes on nested declarators, some laxity is 2857allowed in the placing of attributes. If an attribute that only applies 2858to types is applied to a declaration, it will be treated as applying to 2859the type of that declaration. If an attribute that only applies to 2860declarations is applied to the type of a declaration, it will be treated 2861as applying to that declaration; and, for compatibility with code 2862placing the attributes immediately before the identifier declared, such 2863an attribute applied to a function return type will be treated as 2864applying to the function type, and such an attribute applied to an array 2865element type will be treated as applying to the array type. If an 2866attribute that only applies to function types is applied to a 2867pointer-to-function type, it will be treated as applying to the pointer 2868target type; if such an attribute is applied to a function return type 2869that is not a pointer-to-function type, it will be treated as applying 2870to the function type. 2871 2872@node Function Prototypes 2873@section Prototypes and Old-Style Function Definitions 2874@cindex function prototype declarations 2875@cindex old-style function definitions 2876@cindex promotion of formal parameters 2877 2878GNU C extends ISO C to allow a function prototype to override a later 2879old-style non-prototype definition. Consider the following example: 2880 2881@example 2882/* @r{Use prototypes unless the compiler is old-fashioned.} */ 2883#ifdef __STDC__ 2884#define P(x) x 2885#else 2886#define P(x) () 2887#endif 2888 2889/* @r{Prototype function declaration.} */ 2890int isroot P((uid_t)); 2891 2892/* @r{Old-style function definition.} */ 2893int 2894isroot (x) /* ??? lossage here ??? */ 2895 uid_t x; 2896@{ 2897 return x == 0; 2898@} 2899@end example 2900 2901Suppose the type @code{uid_t} happens to be @code{short}. ISO C does 2902not allow this example, because subword arguments in old-style 2903non-prototype definitions are promoted. Therefore in this example the 2904function definition's argument is really an @code{int}, which does not 2905match the prototype argument type of @code{short}. 2906 2907This restriction of ISO C makes it hard to write code that is portable 2908to traditional C compilers, because the programmer does not know 2909whether the @code{uid_t} type is @code{short}, @code{int}, or 2910@code{long}. Therefore, in cases like these GNU C allows a prototype 2911to override a later old-style definition. More precisely, in GNU C, a 2912function prototype argument type overrides the argument type specified 2913by a later old-style definition if the former type is the same as the 2914latter type before promotion. Thus in GNU C the above example is 2915equivalent to the following: 2916 2917@example 2918int isroot (uid_t); 2919 2920int 2921isroot (uid_t x) 2922@{ 2923 return x == 0; 2924@} 2925@end example 2926 2927@noindent 2928GNU C++ does not support old-style function definitions, so this 2929extension is irrelevant. 2930 2931@node C++ Comments 2932@section C++ Style Comments 2933@cindex // 2934@cindex C++ comments 2935@cindex comments, C++ style 2936 2937In GNU C, you may use C++ style comments, which start with @samp{//} and 2938continue until the end of the line. Many other C implementations allow 2939such comments, and they are included in the 1999 C standard. However, 2940C++ style comments are not recognized if you specify an @option{-std} 2941option specifying a version of ISO C before C99, or @option{-ansi} 2942(equivalent to @option{-std=c89}). 2943 2944@node Dollar Signs 2945@section Dollar Signs in Identifier Names 2946@cindex $ 2947@cindex dollar signs in identifier names 2948@cindex identifier names, dollar signs in 2949 2950In GNU C, you may normally use dollar signs in identifier names. 2951This is because many traditional C implementations allow such identifiers. 2952However, dollar signs in identifiers are not supported on a few target 2953machines, typically because the target assembler does not allow them. 2954 2955@node Character Escapes 2956@section The Character @key{ESC} in Constants 2957 2958You can use the sequence @samp{\e} in a string or character constant to 2959stand for the ASCII character @key{ESC}. 2960 2961@node Alignment 2962@section Inquiring on Alignment of Types or Variables 2963@cindex alignment 2964@cindex type alignment 2965@cindex variable alignment 2966 2967The keyword @code{__alignof__} allows you to inquire about how an object 2968is aligned, or the minimum alignment usually required by a type. Its 2969syntax is just like @code{sizeof}. 2970 2971For example, if the target machine requires a @code{double} value to be 2972aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8. 2973This is true on many RISC machines. On more traditional machine 2974designs, @code{__alignof__ (double)} is 4 or even 2. 2975 2976Some machines never actually require alignment; they allow reference to any 2977data type even at an odd address. For these machines, @code{__alignof__} 2978reports the @emph{recommended} alignment of a type. 2979 2980If the operand of @code{__alignof__} is an lvalue rather than a type, 2981its value is the required alignment for its type, taking into account 2982any minimum alignment specified with GCC's @code{__attribute__} 2983extension (@pxref{Variable Attributes}). For example, after this 2984declaration: 2985 2986@example 2987struct foo @{ int x; char y; @} foo1; 2988@end example 2989 2990@noindent 2991the value of @code{__alignof__ (foo1.y)} is 1, even though its actual 2992alignment is probably 2 or 4, the same as @code{__alignof__ (int)}. 2993 2994It is an error to ask for the alignment of an incomplete type. 2995 2996@node Variable Attributes 2997@section Specifying Attributes of Variables 2998@cindex attribute of variables 2999@cindex variable attributes 3000 3001The keyword @code{__attribute__} allows you to specify special 3002attributes of variables or structure fields. This keyword is followed 3003by an attribute specification inside double parentheses. Some 3004attributes are currently defined generically for variables. 3005Other attributes are defined for variables on particular target 3006systems. Other attributes are available for functions 3007(@pxref{Function Attributes}) and for types (@pxref{Type Attributes}). 3008Other front ends might define more attributes 3009(@pxref{C++ Extensions,,Extensions to the C++ Language}). 3010 3011You may also specify attributes with @samp{__} preceding and following 3012each keyword. This allows you to use them in header files without 3013being concerned about a possible macro of the same name. For example, 3014you may use @code{__aligned__} instead of @code{aligned}. 3015 3016@xref{Attribute Syntax}, for details of the exact syntax for using 3017attributes. 3018 3019@table @code 3020@cindex @code{aligned} attribute 3021@item aligned (@var{alignment}) 3022This attribute specifies a minimum alignment for the variable or 3023structure field, measured in bytes. For example, the declaration: 3024 3025@smallexample 3026int x __attribute__ ((aligned (16))) = 0; 3027@end smallexample 3028 3029@noindent 3030causes the compiler to allocate the global variable @code{x} on a 303116-byte boundary. On a 68040, this could be used in conjunction with 3032an @code{asm} expression to access the @code{move16} instruction which 3033requires 16-byte aligned operands. 3034 3035You can also specify the alignment of structure fields. For example, to 3036create a double-word aligned @code{int} pair, you could write: 3037 3038@smallexample 3039struct foo @{ int x[2] __attribute__ ((aligned (8))); @}; 3040@end smallexample 3041 3042@noindent 3043This is an alternative to creating a union with a @code{double} member 3044that forces the union to be double-word aligned. 3045 3046As in the preceding examples, you can explicitly specify the alignment 3047(in bytes) that you wish the compiler to use for a given variable or 3048structure field. Alternatively, you can leave out the alignment factor 3049and just ask the compiler to align a variable or field to the maximum 3050useful alignment for the target machine you are compiling for. For 3051example, you could write: 3052 3053@smallexample 3054short array[3] __attribute__ ((aligned)); 3055@end smallexample 3056 3057Whenever you leave out the alignment factor in an @code{aligned} attribute 3058specification, the compiler automatically sets the alignment for the declared 3059variable or field to the largest alignment which is ever used for any data 3060type on the target machine you are compiling for. Doing this can often make 3061copy operations more efficient, because the compiler can use whatever 3062instructions copy the biggest chunks of memory when performing copies to 3063or from the variables or fields that you have aligned this way. 3064 3065The @code{aligned} attribute can only increase the alignment; but you 3066can decrease it by specifying @code{packed} as well. See below. 3067 3068Note that the effectiveness of @code{aligned} attributes may be limited 3069by inherent limitations in your linker. On many systems, the linker is 3070only able to arrange for variables to be aligned up to a certain maximum 3071alignment. (For some linkers, the maximum supported alignment may 3072be very very small.) If your linker is only able to align variables 3073up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} 3074in an @code{__attribute__} will still only provide you with 8 byte 3075alignment. See your linker documentation for further information. 3076 3077@item cleanup (@var{cleanup_function}) 3078@cindex @code{cleanup} attribute 3079The @code{cleanup} attribute runs a function when the variable goes 3080out of scope. This attribute can only be applied to auto function 3081scope variables; it may not be applied to parameters or variables 3082with static storage duration. The function must take one parameter, 3083a pointer to a type compatible with the variable. The return value 3084of the function (if any) is ignored. 3085 3086If @option{-fexceptions} is enabled, then @var{cleanup_function} 3087will be run during the stack unwinding that happens during the 3088processing of the exception. Note that the @code{cleanup} attribute 3089does not allow the exception to be caught, only to perform an action. 3090It is undefined what happens if @var{cleanup_function} does not 3091return normally. 3092 3093@item common 3094@itemx nocommon 3095@cindex @code{common} attribute 3096@cindex @code{nocommon} attribute 3097@opindex fcommon 3098@opindex fno-common 3099The @code{common} attribute requests GCC to place a variable in 3100``common'' storage. The @code{nocommon} attribute requests the 3101opposite -- to allocate space for it directly. 3102 3103These attributes override the default chosen by the 3104@option{-fno-common} and @option{-fcommon} flags respectively. 3105 3106@item deprecated 3107@cindex @code{deprecated} attribute 3108The @code{deprecated} attribute results in a warning if the variable 3109is used anywhere in the source file. This is useful when identifying 3110variables that are expected to be removed in a future version of a 3111program. The warning also includes the location of the declaration 3112of the deprecated variable, to enable users to easily find further 3113information about why the variable is deprecated, or what they should 3114do instead. Note that the warning only occurs for uses: 3115 3116@smallexample 3117extern int old_var __attribute__ ((deprecated)); 3118extern int old_var; 3119int new_fn () @{ return old_var; @} 3120@end smallexample 3121 3122results in a warning on line 3 but not line 2. 3123 3124The @code{deprecated} attribute can also be used for functions and 3125types (@pxref{Function Attributes}, @pxref{Type Attributes}.) 3126 3127@item mode (@var{mode}) 3128@cindex @code{mode} attribute 3129This attribute specifies the data type for the declaration---whichever 3130type corresponds to the mode @var{mode}. This in effect lets you 3131request an integer or floating point type according to its width. 3132 3133You may also specify a mode of @samp{byte} or @samp{__byte__} to 3134indicate the mode corresponding to a one-byte integer, @samp{word} or 3135@samp{__word__} for the mode of a one-word integer, and @samp{pointer} 3136or @samp{__pointer__} for the mode used to represent pointers. 3137 3138@item packed 3139@cindex @code{packed} attribute 3140The @code{packed} attribute specifies that a variable or structure field 3141should have the smallest possible alignment---one byte for a variable, 3142and one bit for a field, unless you specify a larger value with the 3143@code{aligned} attribute. 3144 3145Here is a structure in which the field @code{x} is packed, so that it 3146immediately follows @code{a}: 3147 3148@example 3149struct foo 3150@{ 3151 char a; 3152 int x[2] __attribute__ ((packed)); 3153@}; 3154@end example 3155 3156@item section ("@var{section-name}") 3157@cindex @code{section} variable attribute 3158Normally, the compiler places the objects it generates in sections like 3159@code{data} and @code{bss}. Sometimes, however, you need additional sections, 3160or you need certain particular variables to appear in special sections, 3161for example to map to special hardware. The @code{section} 3162attribute specifies that a variable (or function) lives in a particular 3163section. For example, this small program uses several specific section names: 3164 3165@smallexample 3166struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @}; 3167struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @}; 3168char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @}; 3169int init_data __attribute__ ((section ("INITDATA"))) = 0; 3170 3171main() 3172@{ 3173 /* Initialize stack pointer */ 3174 init_sp (stack + sizeof (stack)); 3175 3176 /* Initialize initialized data */ 3177 memcpy (&init_data, &data, &edata - &data); 3178 3179 /* Turn on the serial ports */ 3180 init_duart (&a); 3181 init_duart (&b); 3182@} 3183@end smallexample 3184 3185@noindent 3186Use the @code{section} attribute with an @emph{initialized} definition 3187of a @emph{global} variable, as shown in the example. GCC issues 3188a warning and otherwise ignores the @code{section} attribute in 3189uninitialized variable declarations. 3190 3191You may only use the @code{section} attribute with a fully initialized 3192global definition because of the way linkers work. The linker requires 3193each object be defined once, with the exception that uninitialized 3194variables tentatively go in the @code{common} (or @code{bss}) section 3195and can be multiply ``defined''. You can force a variable to be 3196initialized with the @option{-fno-common} flag or the @code{nocommon} 3197attribute. 3198 3199Some file formats do not support arbitrary sections so the @code{section} 3200attribute is not available on all platforms. 3201If you need to map the entire contents of a module to a particular 3202section, consider using the facilities of the linker instead. 3203 3204@item shared 3205@cindex @code{shared} variable attribute 3206On Windows, in addition to putting variable definitions in a named 3207section, the section can also be shared among all running copies of an 3208executable or DLL@. For example, this small program defines shared data 3209by putting it in a named section @code{shared} and marking the section 3210shareable: 3211 3212@smallexample 3213int foo __attribute__((section ("shared"), shared)) = 0; 3214 3215int 3216main() 3217@{ 3218 /* Read and write foo. All running 3219 copies see the same value. */ 3220 return 0; 3221@} 3222@end smallexample 3223 3224@noindent 3225You may only use the @code{shared} attribute along with @code{section} 3226attribute with a fully initialized global definition because of the way 3227linkers work. See @code{section} attribute for more information. 3228 3229The @code{shared} attribute is only available on Windows@. 3230 3231@item tls_model ("@var{tls_model}") 3232@cindex @code{tls_model} attribute 3233The @code{tls_model} attribute sets thread-local storage model 3234(@pxref{Thread-Local}) of a particular @code{__thread} variable, 3235overriding @code{-ftls-model=} command line switch on a per-variable 3236basis. 3237The @var{tls_model} argument should be one of @code{global-dynamic}, 3238@code{local-dynamic}, @code{initial-exec} or @code{local-exec}. 3239 3240Not all targets support this attribute. 3241 3242@item transparent_union 3243This attribute, attached to a function parameter which is a union, means 3244that the corresponding argument may have the type of any union member, 3245but the argument is passed as if its type were that of the first union 3246member. For more details see @xref{Type Attributes}. You can also use 3247this attribute on a @code{typedef} for a union data type; then it 3248applies to all function parameters with that type. 3249 3250@item unused 3251This attribute, attached to a variable, means that the variable is meant 3252to be possibly unused. GCC will not produce a warning for this 3253variable. 3254 3255@item vector_size (@var{bytes}) 3256This attribute specifies the vector size for the variable, measured in 3257bytes. For example, the declaration: 3258 3259@smallexample 3260int foo __attribute__ ((vector_size (16))); 3261@end smallexample 3262 3263@noindent 3264causes the compiler to set the mode for @code{foo}, to be 16 bytes, 3265divided into @code{int} sized units. Assuming a 32-bit int (a vector of 32664 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@. 3267 3268This attribute is only applicable to integral and float scalars, 3269although arrays, pointers, and function return values are allowed in 3270conjunction with this construct. 3271 3272Aggregates with this attribute are invalid, even if they are of the same 3273size as a corresponding scalar. For example, the declaration: 3274 3275@smallexample 3276struct S @{ int a; @}; 3277struct S __attribute__ ((vector_size (16))) foo; 3278@end smallexample 3279 3280@noindent 3281is invalid even if the size of the structure is the same as the size of 3282the @code{int}. 3283 3284@item weak 3285The @code{weak} attribute is described in @xref{Function Attributes}. 3286 3287@item model (@var{model-name}) 3288@cindex variable addressability on the M32R/D 3289Use this attribute on the M32R/D to set the addressability of an object. 3290The identifier @var{model-name} is one of @code{small}, @code{medium}, 3291or @code{large}, representing each of the code models. 3292 3293Small model objects live in the lower 16MB of memory (so that their 3294addresses can be loaded with the @code{ld24} instruction). 3295 3296Medium and large model objects may live anywhere in the 32-bit address space 3297(the compiler will generate @code{seth/add3} instructions to load their 3298addresses). 3299 3300@item dllimport 3301The @code{dllimport} attribute is described in @xref{Function Attributes}. 3302 3303@item dlexport 3304The @code{dllexport} attribute is described in @xref{Function Attributes}. 3305 3306@end table 3307 3308To specify multiple attributes, separate them by commas within the 3309double parentheses: for example, @samp{__attribute__ ((aligned (16), 3310packed))}. 3311 3312@node Type Attributes 3313@section Specifying Attributes of Types 3314@cindex attribute of types 3315@cindex type attributes 3316 3317The keyword @code{__attribute__} allows you to specify special 3318attributes of @code{struct} and @code{union} types when you define such 3319types. This keyword is followed by an attribute specification inside 3320double parentheses. Six attributes are currently defined for types: 3321@code{aligned}, @code{packed}, @code{transparent_union}, @code{unused}, 3322@code{deprecated} and @code{may_alias}. Other attributes are defined for 3323functions (@pxref{Function Attributes}) and for variables 3324(@pxref{Variable Attributes}). 3325 3326You may also specify any one of these attributes with @samp{__} 3327preceding and following its keyword. This allows you to use these 3328attributes in header files without being concerned about a possible 3329macro of the same name. For example, you may use @code{__aligned__} 3330instead of @code{aligned}. 3331 3332You may specify the @code{aligned} and @code{transparent_union} 3333attributes either in a @code{typedef} declaration or just past the 3334closing curly brace of a complete enum, struct or union type 3335@emph{definition} and the @code{packed} attribute only past the closing 3336brace of a definition. 3337 3338You may also specify attributes between the enum, struct or union 3339tag and the name of the type rather than after the closing brace. 3340 3341@xref{Attribute Syntax}, for details of the exact syntax for using 3342attributes. 3343 3344@table @code 3345@cindex @code{aligned} attribute 3346@item aligned (@var{alignment}) 3347This attribute specifies a minimum alignment (in bytes) for variables 3348of the specified type. For example, the declarations: 3349 3350@smallexample 3351struct S @{ short f[3]; @} __attribute__ ((aligned (8))); 3352typedef int more_aligned_int __attribute__ ((aligned (8))); 3353@end smallexample 3354 3355@noindent 3356force the compiler to insure (as far as it can) that each variable whose 3357type is @code{struct S} or @code{more_aligned_int} will be allocated and 3358aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all 3359variables of type @code{struct S} aligned to 8-byte boundaries allows 3360the compiler to use the @code{ldd} and @code{std} (doubleword load and 3361store) instructions when copying one variable of type @code{struct S} to 3362another, thus improving run-time efficiency. 3363 3364Note that the alignment of any given @code{struct} or @code{union} type 3365is required by the ISO C standard to be at least a perfect multiple of 3366the lowest common multiple of the alignments of all of the members of 3367the @code{struct} or @code{union} in question. This means that you @emph{can} 3368effectively adjust the alignment of a @code{struct} or @code{union} 3369type by attaching an @code{aligned} attribute to any one of the members 3370of such a type, but the notation illustrated in the example above is a 3371more obvious, intuitive, and readable way to request the compiler to 3372adjust the alignment of an entire @code{struct} or @code{union} type. 3373 3374As in the preceding example, you can explicitly specify the alignment 3375(in bytes) that you wish the compiler to use for a given @code{struct} 3376or @code{union} type. Alternatively, you can leave out the alignment factor 3377and just ask the compiler to align a type to the maximum 3378useful alignment for the target machine you are compiling for. For 3379example, you could write: 3380 3381@smallexample 3382struct S @{ short f[3]; @} __attribute__ ((aligned)); 3383@end smallexample 3384 3385Whenever you leave out the alignment factor in an @code{aligned} 3386attribute specification, the compiler automatically sets the alignment 3387for the type to the largest alignment which is ever used for any data 3388type on the target machine you are compiling for. Doing this can often 3389make copy operations more efficient, because the compiler can use 3390whatever instructions copy the biggest chunks of memory when performing 3391copies to or from the variables which have types that you have aligned 3392this way. 3393 3394In the example above, if the size of each @code{short} is 2 bytes, then 3395the size of the entire @code{struct S} type is 6 bytes. The smallest 3396power of two which is greater than or equal to that is 8, so the 3397compiler sets the alignment for the entire @code{struct S} type to 8 3398bytes. 3399 3400Note that although you can ask the compiler to select a time-efficient 3401alignment for a given type and then declare only individual stand-alone 3402objects of that type, the compiler's ability to select a time-efficient 3403alignment is primarily useful only when you plan to create arrays of 3404variables having the relevant (efficiently aligned) type. If you 3405declare or use arrays of variables of an efficiently-aligned type, then 3406it is likely that your program will also be doing pointer arithmetic (or 3407subscripting, which amounts to the same thing) on pointers to the 3408relevant type, and the code that the compiler generates for these 3409pointer arithmetic operations will often be more efficient for 3410efficiently-aligned types than for other types. 3411 3412The @code{aligned} attribute can only increase the alignment; but you 3413can decrease it by specifying @code{packed} as well. See below. 3414 3415Note that the effectiveness of @code{aligned} attributes may be limited 3416by inherent limitations in your linker. On many systems, the linker is 3417only able to arrange for variables to be aligned up to a certain maximum 3418alignment. (For some linkers, the maximum supported alignment may 3419be very very small.) If your linker is only able to align variables 3420up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} 3421in an @code{__attribute__} will still only provide you with 8 byte 3422alignment. See your linker documentation for further information. 3423 3424@item packed 3425This attribute, attached to an @code{enum}, @code{struct}, or 3426@code{union} type definition, specifies that the minimum required memory 3427be used to represent the type. 3428 3429@opindex fshort-enums 3430Specifying this attribute for @code{struct} and @code{union} types is 3431equivalent to specifying the @code{packed} attribute on each of the 3432structure or union members. Specifying the @option{-fshort-enums} 3433flag on the line is equivalent to specifying the @code{packed} 3434attribute on all @code{enum} definitions. 3435 3436You may only specify this attribute after a closing curly brace on an 3437@code{enum} definition, not in a @code{typedef} declaration, unless that 3438declaration also contains the definition of the @code{enum}. 3439 3440@item transparent_union 3441This attribute, attached to a @code{union} type definition, indicates 3442that any function parameter having that union type causes calls to that 3443function to be treated in a special way. 3444 3445First, the argument corresponding to a transparent union type can be of 3446any type in the union; no cast is required. Also, if the union contains 3447a pointer type, the corresponding argument can be a null pointer 3448constant or a void pointer expression; and if the union contains a void 3449pointer type, the corresponding argument can be any pointer expression. 3450If the union member type is a pointer, qualifiers like @code{const} on 3451the referenced type must be respected, just as with normal pointer 3452conversions. 3453 3454Second, the argument is passed to the function using the calling 3455conventions of the first member of the transparent union, not the calling 3456conventions of the union itself. All members of the union must have the 3457same machine representation; this is necessary for this argument passing 3458to work properly. 3459 3460Transparent unions are designed for library functions that have multiple 3461interfaces for compatibility reasons. For example, suppose the 3462@code{wait} function must accept either a value of type @code{int *} to 3463comply with Posix, or a value of type @code{union wait *} to comply with 3464the 4.1BSD interface. If @code{wait}'s parameter were @code{void *}, 3465@code{wait} would accept both kinds of arguments, but it would also 3466accept any other pointer type and this would make argument type checking 3467less useful. Instead, @code{<sys/wait.h>} might define the interface 3468as follows: 3469 3470@smallexample 3471typedef union 3472 @{ 3473 int *__ip; 3474 union wait *__up; 3475 @} wait_status_ptr_t __attribute__ ((__transparent_union__)); 3476 3477pid_t wait (wait_status_ptr_t); 3478@end smallexample 3479 3480This interface allows either @code{int *} or @code{union wait *} 3481arguments to be passed, using the @code{int *} calling convention. 3482The program can call @code{wait} with arguments of either type: 3483 3484@example 3485int w1 () @{ int w; return wait (&w); @} 3486int w2 () @{ union wait w; return wait (&w); @} 3487@end example 3488 3489With this interface, @code{wait}'s implementation might look like this: 3490 3491@example 3492pid_t wait (wait_status_ptr_t p) 3493@{ 3494 return waitpid (-1, p.__ip, 0); 3495@} 3496@end example 3497 3498@item unused 3499When attached to a type (including a @code{union} or a @code{struct}), 3500this attribute means that variables of that type are meant to appear 3501possibly unused. GCC will not produce a warning for any variables of 3502that type, even if the variable appears to do nothing. This is often 3503the case with lock or thread classes, which are usually defined and then 3504not referenced, but contain constructors and destructors that have 3505nontrivial bookkeeping functions. 3506 3507@item deprecated 3508The @code{deprecated} attribute results in a warning if the type 3509is used anywhere in the source file. This is useful when identifying 3510types that are expected to be removed in a future version of a program. 3511If possible, the warning also includes the location of the declaration 3512of the deprecated type, to enable users to easily find further 3513information about why the type is deprecated, or what they should do 3514instead. Note that the warnings only occur for uses and then only 3515if the type is being applied to an identifier that itself is not being 3516declared as deprecated. 3517 3518@smallexample 3519typedef int T1 __attribute__ ((deprecated)); 3520T1 x; 3521typedef T1 T2; 3522T2 y; 3523typedef T1 T3 __attribute__ ((deprecated)); 3524T3 z __attribute__ ((deprecated)); 3525@end smallexample 3526 3527results in a warning on line 2 and 3 but not lines 4, 5, or 6. No 3528warning is issued for line 4 because T2 is not explicitly 3529deprecated. Line 5 has no warning because T3 is explicitly 3530deprecated. Similarly for line 6. 3531 3532The @code{deprecated} attribute can also be used for functions and 3533variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.) 3534 3535@item may_alias 3536Accesses to objects with types with this attribute are not subjected to 3537type-based alias analysis, but are instead assumed to be able to alias 3538any other type of objects, just like the @code{char} type. See 3539@option{-fstrict-aliasing} for more information on aliasing issues. 3540 3541Example of use: 3542 3543@smallexample 3544typedef short __attribute__((__may_alias__)) short_a; 3545 3546int 3547main (void) 3548@{ 3549 int a = 0x12345678; 3550 short_a *b = (short_a *) &a; 3551 3552 b[1] = 0; 3553 3554 if (a == 0x12345678) 3555 abort(); 3556 3557 exit(0); 3558@} 3559@end smallexample 3560 3561If you replaced @code{short_a} with @code{short} in the variable 3562declaration, the above program would abort when compiled with 3563@option{-fstrict-aliasing}, which is on by default at @option{-O2} or 3564above in recent GCC versions. 3565@end table 3566 3567To specify multiple attributes, separate them by commas within the 3568double parentheses: for example, @samp{__attribute__ ((aligned (16), 3569packed))}. 3570 3571@node Inline 3572@section An Inline Function is As Fast As a Macro 3573@cindex inline functions 3574@cindex integrating function code 3575@cindex open coding 3576@cindex macros, inline alternative 3577 3578By declaring a function @code{inline}, you can direct GCC to 3579integrate that function's code into the code for its callers. This 3580makes execution faster by eliminating the function-call overhead; in 3581addition, if any of the actual argument values are constant, their known 3582values may permit simplifications at compile time so that not all of the 3583inline function's code needs to be included. The effect on code size is 3584less predictable; object code may be larger or smaller with function 3585inlining, depending on the particular case. Inlining of functions is an 3586optimization and it really ``works'' only in optimizing compilation. If 3587you don't use @option{-O}, no function is really inline. 3588 3589Inline functions are included in the ISO C99 standard, but there are 3590currently substantial differences between what GCC implements and what 3591the ISO C99 standard requires. 3592 3593To declare a function inline, use the @code{inline} keyword in its 3594declaration, like this: 3595 3596@example 3597inline int 3598inc (int *a) 3599@{ 3600 (*a)++; 3601@} 3602@end example 3603 3604(If you are writing a header file to be included in ISO C programs, write 3605@code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.) 3606You can also make all ``simple enough'' functions inline with the option 3607@option{-finline-functions}. 3608 3609@opindex Winline 3610Note that certain usages in a function definition can make it unsuitable 3611for inline substitution. Among these usages are: use of varargs, use of 3612alloca, use of variable sized data types (@pxref{Variable Length}), 3613use of computed goto (@pxref{Labels as Values}), use of nonlocal goto, 3614and nested functions (@pxref{Nested Functions}). Using @option{-Winline} 3615will warn when a function marked @code{inline} could not be substituted, 3616and will give the reason for the failure. 3617 3618Note that in C and Objective-C, unlike C++, the @code{inline} keyword 3619does not affect the linkage of the function. 3620 3621@cindex automatic @code{inline} for C++ member fns 3622@cindex @code{inline} automatic for C++ member fns 3623@cindex member fns, automatically @code{inline} 3624@cindex C++ member fns, automatically @code{inline} 3625@opindex fno-default-inline 3626GCC automatically inlines member functions defined within the class 3627body of C++ programs even if they are not explicitly declared 3628@code{inline}. (You can override this with @option{-fno-default-inline}; 3629@pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.) 3630 3631@cindex inline functions, omission of 3632@opindex fkeep-inline-functions 3633When a function is both inline and @code{static}, if all calls to the 3634function are integrated into the caller, and the function's address is 3635never used, then the function's own assembler code is never referenced. 3636In this case, GCC does not actually output assembler code for the 3637function, unless you specify the option @option{-fkeep-inline-functions}. 3638Some calls cannot be integrated for various reasons (in particular, 3639calls that precede the function's definition cannot be integrated, and 3640neither can recursive calls within the definition). If there is a 3641nonintegrated call, then the function is compiled to assembler code as 3642usual. The function must also be compiled as usual if the program 3643refers to its address, because that can't be inlined. 3644 3645@cindex non-static inline function 3646When an inline function is not @code{static}, then the compiler must assume 3647that there may be calls from other source files; since a global symbol can 3648be defined only once in any program, the function must not be defined in 3649the other source files, so the calls therein cannot be integrated. 3650Therefore, a non-@code{static} inline function is always compiled on its 3651own in the usual fashion. 3652 3653If you specify both @code{inline} and @code{extern} in the function 3654definition, then the definition is used only for inlining. In no case 3655is the function compiled on its own, not even if you refer to its 3656address explicitly. Such an address becomes an external reference, as 3657if you had only declared the function, and had not defined it. 3658 3659This combination of @code{inline} and @code{extern} has almost the 3660effect of a macro. The way to use it is to put a function definition in 3661a header file with these keywords, and put another copy of the 3662definition (lacking @code{inline} and @code{extern}) in a library file. 3663The definition in the header file will cause most calls to the function 3664to be inlined. If any uses of the function remain, they will refer to 3665the single copy in the library. 3666 3667Since GCC eventually will implement ISO C99 semantics for 3668inline functions, it is best to use @code{static inline} only 3669to guarantee compatibility. (The 3670existing semantics will remain available when @option{-std=gnu89} is 3671specified, but eventually the default will be @option{-std=gnu99} and 3672that will implement the C99 semantics, though it does not do so yet.) 3673 3674GCC does not inline any functions when not optimizing unless you specify 3675the @samp{always_inline} attribute for the function, like this: 3676 3677@example 3678/* Prototype. */ 3679inline void foo (const char) __attribute__((always_inline)); 3680@end example 3681 3682@node Extended Asm 3683@section Assembler Instructions with C Expression Operands 3684@cindex extended @code{asm} 3685@cindex @code{asm} expressions 3686@cindex assembler instructions 3687@cindex registers 3688 3689In an assembler instruction using @code{asm}, you can specify the 3690operands of the instruction using C expressions. This means you need not 3691guess which registers or memory locations will contain the data you want 3692to use. 3693 3694You must specify an assembler instruction template much like what 3695appears in a machine description, plus an operand constraint string for 3696each operand. 3697 3698For example, here is how to use the 68881's @code{fsinx} instruction: 3699 3700@example 3701asm ("fsinx %1,%0" : "=f" (result) : "f" (angle)); 3702@end example 3703 3704@noindent 3705Here @code{angle} is the C expression for the input operand while 3706@code{result} is that of the output operand. Each has @samp{"f"} as its 3707operand constraint, saying that a floating point register is required. 3708The @samp{=} in @samp{=f} indicates that the operand is an output; all 3709output operands' constraints must use @samp{=}. The constraints use the 3710same language used in the machine description (@pxref{Constraints}). 3711 3712Each operand is described by an operand-constraint string followed by 3713the C expression in parentheses. A colon separates the assembler 3714template from the first output operand and another separates the last 3715output operand from the first input, if any. Commas separate the 3716operands within each group. The total number of operands is currently 3717limited to 30; this limitation may be lifted in some future version of 3718GCC. 3719 3720If there are no output operands but there are input operands, you must 3721place two consecutive colons surrounding the place where the output 3722operands would go. 3723 3724As of GCC version 3.1, it is also possible to specify input and output 3725operands using symbolic names which can be referenced within the 3726assembler code. These names are specified inside square brackets 3727preceding the constraint string, and can be referenced inside the 3728assembler code using @code{%[@var{name}]} instead of a percentage sign 3729followed by the operand number. Using named operands the above example 3730could look like: 3731 3732@example 3733asm ("fsinx %[angle],%[output]" 3734 : [output] "=f" (result) 3735 : [angle] "f" (angle)); 3736@end example 3737 3738@noindent 3739Note that the symbolic operand names have no relation whatsoever to 3740other C identifiers. You may use any name you like, even those of 3741existing C symbols, but you must ensure that no two operands within the same 3742assembler construct use the same symbolic name. 3743 3744Output operand expressions must be lvalues; the compiler can check this. 3745The input operands need not be lvalues. The compiler cannot check 3746whether the operands have data types that are reasonable for the 3747instruction being executed. It does not parse the assembler instruction 3748template and does not know what it means or even whether it is valid 3749assembler input. The extended @code{asm} feature is most often used for 3750machine instructions the compiler itself does not know exist. If 3751the output expression cannot be directly addressed (for example, it is a 3752bit-field), your constraint must allow a register. In that case, GCC 3753will use the register as the output of the @code{asm}, and then store 3754that register into the output. 3755 3756The ordinary output operands must be write-only; GCC will assume that 3757the values in these operands before the instruction are dead and need 3758not be generated. Extended asm supports input-output or read-write 3759operands. Use the constraint character @samp{+} to indicate such an 3760operand and list it with the output operands. 3761 3762When the constraints for the read-write operand (or the operand in which 3763only some of the bits are to be changed) allows a register, you may, as 3764an alternative, logically split its function into two separate operands, 3765one input operand and one write-only output operand. The connection 3766between them is expressed by constraints which say they need to be in 3767the same location when the instruction executes. You can use the same C 3768expression for both operands, or different expressions. For example, 3769here we write the (fictitious) @samp{combine} instruction with 3770@code{bar} as its read-only source operand and @code{foo} as its 3771read-write destination: 3772 3773@example 3774asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar)); 3775@end example 3776 3777@noindent 3778The constraint @samp{"0"} for operand 1 says that it must occupy the 3779same location as operand 0. A number in constraint is allowed only in 3780an input operand and it must refer to an output operand. 3781 3782Only a number in the constraint can guarantee that one operand will be in 3783the same place as another. The mere fact that @code{foo} is the value 3784of both operands is not enough to guarantee that they will be in the 3785same place in the generated assembler code. The following would not 3786work reliably: 3787 3788@example 3789asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar)); 3790@end example 3791 3792Various optimizations or reloading could cause operands 0 and 1 to be in 3793different registers; GCC knows no reason not to do so. For example, the 3794compiler might find a copy of the value of @code{foo} in one register and 3795use it for operand 1, but generate the output operand 0 in a different 3796register (copying it afterward to @code{foo}'s own address). Of course, 3797since the register for operand 1 is not even mentioned in the assembler 3798code, the result will not work, but GCC can't tell that. 3799 3800As of GCC version 3.1, one may write @code{[@var{name}]} instead of 3801the operand number for a matching constraint. For example: 3802 3803@example 3804asm ("cmoveq %1,%2,%[result]" 3805 : [result] "=r"(result) 3806 : "r" (test), "r"(new), "[result]"(old)); 3807@end example 3808 3809Some instructions clobber specific hard registers. To describe this, 3810write a third colon after the input operands, followed by the names of 3811the clobbered hard registers (given as strings). Here is a realistic 3812example for the VAX: 3813 3814@example 3815asm volatile ("movc3 %0,%1,%2" 3816 : /* no outputs */ 3817 : "g" (from), "g" (to), "g" (count) 3818 : "r0", "r1", "r2", "r3", "r4", "r5"); 3819@end example 3820 3821You may not write a clobber description in a way that overlaps with an 3822input or output operand. For example, you may not have an operand 3823describing a register class with one member if you mention that register 3824in the clobber list. Variables declared to live in specific registers 3825(@pxref{Explicit Reg Vars}), and used as asm input or output operands must 3826have no part mentioned in the clobber description. 3827There is no way for you to specify that an input 3828operand is modified without also specifying it as an output 3829operand. Note that if all the output operands you specify are for this 3830purpose (and hence unused), you will then also need to specify 3831@code{volatile} for the @code{asm} construct, as described below, to 3832prevent GCC from deleting the @code{asm} statement as unused. 3833 3834If you refer to a particular hardware register from the assembler code, 3835you will probably have to list the register after the third colon to 3836tell the compiler the register's value is modified. In some assemblers, 3837the register names begin with @samp{%}; to produce one @samp{%} in the 3838assembler code, you must write @samp{%%} in the input. 3839 3840If your assembler instruction can alter the condition code register, add 3841@samp{cc} to the list of clobbered registers. GCC on some machines 3842represents the condition codes as a specific hardware register; 3843@samp{cc} serves to name this register. On other machines, the 3844condition code is handled differently, and specifying @samp{cc} has no 3845effect. But it is valid no matter what the machine. 3846 3847If your assembler instruction modifies memory in an unpredictable 3848fashion, add @samp{memory} to the list of clobbered registers. This 3849will cause GCC to not keep memory values cached in registers across 3850the assembler instruction. You will also want to add the 3851@code{volatile} keyword if the memory affected is not listed in the 3852inputs or outputs of the @code{asm}, as the @samp{memory} clobber does 3853not count as a side-effect of the @code{asm}. 3854 3855You can put multiple assembler instructions together in a single 3856@code{asm} template, separated by the characters normally used in assembly 3857code for the system. A combination that works in most places is a newline 3858to break the line, plus a tab character to move to the instruction field 3859(written as @samp{\n\t}). Sometimes semicolons can be used, if the 3860assembler allows semicolons as a line-breaking character. Note that some 3861assembler dialects use semicolons to start a comment. 3862The input operands are guaranteed not to use any of the clobbered 3863registers, and neither will the output operands' addresses, so you can 3864read and write the clobbered registers as many times as you like. Here 3865is an example of multiple instructions in a template; it assumes the 3866subroutine @code{_foo} accepts arguments in registers 9 and 10: 3867 3868@example 3869asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo" 3870 : /* no outputs */ 3871 : "g" (from), "g" (to) 3872 : "r9", "r10"); 3873@end example 3874 3875Unless an output operand has the @samp{&} constraint modifier, GCC 3876may allocate it in the same register as an unrelated input operand, on 3877the assumption the inputs are consumed before the outputs are produced. 3878This assumption may be false if the assembler code actually consists of 3879more than one instruction. In such a case, use @samp{&} for each output 3880operand that may not overlap an input. @xref{Modifiers}. 3881 3882If you want to test the condition code produced by an assembler 3883instruction, you must include a branch and a label in the @code{asm} 3884construct, as follows: 3885 3886@example 3887asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:" 3888 : "g" (result) 3889 : "g" (input)); 3890@end example 3891 3892@noindent 3893This assumes your assembler supports local labels, as the GNU assembler 3894and most Unix assemblers do. 3895 3896Speaking of labels, jumps from one @code{asm} to another are not 3897supported. The compiler's optimizers do not know about these jumps, and 3898therefore they cannot take account of them when deciding how to 3899optimize. 3900 3901@cindex macros containing @code{asm} 3902Usually the most convenient way to use these @code{asm} instructions is to 3903encapsulate them in macros that look like functions. For example, 3904 3905@example 3906#define sin(x) \ 3907(@{ double __value, __arg = (x); \ 3908 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \ 3909 __value; @}) 3910@end example 3911 3912@noindent 3913Here the variable @code{__arg} is used to make sure that the instruction 3914operates on a proper @code{double} value, and to accept only those 3915arguments @code{x} which can convert automatically to a @code{double}. 3916 3917Another way to make sure the instruction operates on the correct data 3918type is to use a cast in the @code{asm}. This is different from using a 3919variable @code{__arg} in that it converts more different types. For 3920example, if the desired type were @code{int}, casting the argument to 3921@code{int} would accept a pointer with no complaint, while assigning the 3922argument to an @code{int} variable named @code{__arg} would warn about 3923using a pointer unless the caller explicitly casts it. 3924 3925If an @code{asm} has output operands, GCC assumes for optimization 3926purposes the instruction has no side effects except to change the output 3927operands. This does not mean instructions with a side effect cannot be 3928used, but you must be careful, because the compiler may eliminate them 3929if the output operands aren't used, or move them out of loops, or 3930replace two with one if they constitute a common subexpression. Also, 3931if your instruction does have a side effect on a variable that otherwise 3932appears not to change, the old value of the variable may be reused later 3933if it happens to be found in a register. 3934 3935You can prevent an @code{asm} instruction from being deleted, moved 3936significantly, or combined, by writing the keyword @code{volatile} after 3937the @code{asm}. For example: 3938 3939@example 3940#define get_and_set_priority(new) \ 3941(@{ int __old; \ 3942 asm volatile ("get_and_set_priority %0, %1" \ 3943 : "=g" (__old) : "g" (new)); \ 3944 __old; @}) 3945@end example 3946 3947@noindent 3948If you write an @code{asm} instruction with no outputs, GCC will know 3949the instruction has side-effects and will not delete the instruction or 3950move it outside of loops. 3951 3952The @code{volatile} keyword indicates that the instruction has 3953important side-effects. GCC will not delete a volatile @code{asm} if 3954it is reachable. (The instruction can still be deleted if GCC can 3955prove that control-flow will never reach the location of the 3956instruction.) In addition, GCC will not reschedule instructions 3957across a volatile @code{asm} instruction. For example: 3958 3959@example 3960*(volatile int *)addr = foo; 3961asm volatile ("eieio" : : ); 3962@end example 3963 3964@noindent 3965Assume @code{addr} contains the address of a memory mapped device 3966register. The PowerPC @code{eieio} instruction (Enforce In-order 3967Execution of I/O) tells the CPU to make sure that the store to that 3968device register happens before it issues any other I/O@. 3969 3970Note that even a volatile @code{asm} instruction can be moved in ways 3971that appear insignificant to the compiler, such as across jump 3972instructions. You can't expect a sequence of volatile @code{asm} 3973instructions to remain perfectly consecutive. If you want consecutive 3974output, use a single @code{asm}. Also, GCC will perform some 3975optimizations across a volatile @code{asm} instruction; GCC does not 3976``forget everything'' when it encounters a volatile @code{asm} 3977instruction the way some other compilers do. 3978 3979An @code{asm} instruction without any operands or clobbers (an ``old 3980style'' @code{asm}) will be treated identically to a volatile 3981@code{asm} instruction. 3982 3983It is a natural idea to look for a way to give access to the condition 3984code left by the assembler instruction. However, when we attempted to 3985implement this, we found no way to make it work reliably. The problem 3986is that output operands might need reloading, which would result in 3987additional following ``store'' instructions. On most machines, these 3988instructions would alter the condition code before there was time to 3989test it. This problem doesn't arise for ordinary ``test'' and 3990``compare'' instructions because they don't have any output operands. 3991 3992For reasons similar to those described above, it is not possible to give 3993an assembler instruction access to the condition code left by previous 3994instructions. 3995 3996If you are writing a header file that should be includable in ISO C 3997programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate 3998Keywords}. 3999 4000@subsection i386 floating point asm operands 4001 4002There are several rules on the usage of stack-like regs in 4003asm_operands insns. These rules apply only to the operands that are 4004stack-like regs: 4005 4006@enumerate 4007@item 4008Given a set of input regs that die in an asm_operands, it is 4009necessary to know which are implicitly popped by the asm, and 4010which must be explicitly popped by gcc. 4011 4012An input reg that is implicitly popped by the asm must be 4013explicitly clobbered, unless it is constrained to match an 4014output operand. 4015 4016@item 4017For any input reg that is implicitly popped by an asm, it is 4018necessary to know how to adjust the stack to compensate for the pop. 4019If any non-popped input is closer to the top of the reg-stack than 4020the implicitly popped reg, it would not be possible to know what the 4021stack looked like---it's not clear how the rest of the stack ``slides 4022up''. 4023 4024All implicitly popped input regs must be closer to the top of 4025the reg-stack than any input that is not implicitly popped. 4026 4027It is possible that if an input dies in an insn, reload might 4028use the input reg for an output reload. Consider this example: 4029 4030@example 4031asm ("foo" : "=t" (a) : "f" (b)); 4032@end example 4033 4034This asm says that input B is not popped by the asm, and that 4035the asm pushes a result onto the reg-stack, i.e., the stack is one 4036deeper after the asm than it was before. But, it is possible that 4037reload will think that it can use the same reg for both the input and 4038the output, if input B dies in this insn. 4039 4040If any input operand uses the @code{f} constraint, all output reg 4041constraints must use the @code{&} earlyclobber. 4042 4043The asm above would be written as 4044 4045@example 4046asm ("foo" : "=&t" (a) : "f" (b)); 4047@end example 4048 4049@item 4050Some operands need to be in particular places on the stack. All 4051output operands fall in this category---there is no other way to 4052know which regs the outputs appear in unless the user indicates 4053this in the constraints. 4054 4055Output operands must specifically indicate which reg an output 4056appears in after an asm. @code{=f} is not allowed: the operand 4057constraints must select a class with a single reg. 4058 4059@item 4060Output operands may not be ``inserted'' between existing stack regs. 4061Since no 387 opcode uses a read/write operand, all output operands 4062are dead before the asm_operands, and are pushed by the asm_operands. 4063It makes no sense to push anywhere but the top of the reg-stack. 4064 4065Output operands must start at the top of the reg-stack: output 4066operands may not ``skip'' a reg. 4067 4068@item 4069Some asm statements may need extra stack space for internal 4070calculations. This can be guaranteed by clobbering stack registers 4071unrelated to the inputs and outputs. 4072 4073@end enumerate 4074 4075Here are a couple of reasonable asms to want to write. This asm 4076takes one input, which is internally popped, and produces two outputs. 4077 4078@example 4079asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); 4080@end example 4081 4082This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode, 4083and replaces them with one output. The user must code the @code{st(1)} 4084clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs. 4085 4086@example 4087asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); 4088@end example 4089 4090@include md.texi 4091 4092@node Asm Labels 4093@section Controlling Names Used in Assembler Code 4094@cindex assembler names for identifiers 4095@cindex names used in assembler code 4096@cindex identifiers, names in assembler code 4097 4098You can specify the name to be used in the assembler code for a C 4099function or variable by writing the @code{asm} (or @code{__asm__}) 4100keyword after the declarator as follows: 4101 4102@example 4103int foo asm ("myfoo") = 2; 4104@end example 4105 4106@noindent 4107This specifies that the name to be used for the variable @code{foo} in 4108the assembler code should be @samp{myfoo} rather than the usual 4109@samp{_foo}. 4110 4111On systems where an underscore is normally prepended to the name of a C 4112function or variable, this feature allows you to define names for the 4113linker that do not start with an underscore. 4114 4115It does not make sense to use this feature with a non-static local 4116variable since such variables do not have assembler names. If you are 4117trying to put the variable in a particular register, see @ref{Explicit 4118Reg Vars}. GCC presently accepts such code with a warning, but will 4119probably be changed to issue an error, rather than a warning, in the 4120future. 4121 4122You cannot use @code{asm} in this way in a function @emph{definition}; but 4123you can get the same effect by writing a declaration for the function 4124before its definition and putting @code{asm} there, like this: 4125 4126@example 4127extern func () asm ("FUNC"); 4128 4129func (x, y) 4130 int x, y; 4131/* @r{@dots{}} */ 4132@end example 4133 4134It is up to you to make sure that the assembler names you choose do not 4135conflict with any other assembler symbols. Also, you must not use a 4136register name; that would produce completely invalid assembler code. GCC 4137does not as yet have the ability to store static variables in registers. 4138Perhaps that will be added. 4139 4140@node Explicit Reg Vars 4141@section Variables in Specified Registers 4142@cindex explicit register variables 4143@cindex variables in specified registers 4144@cindex specified registers 4145@cindex registers, global allocation 4146 4147GNU C allows you to put a few global variables into specified hardware 4148registers. You can also specify the register in which an ordinary 4149register variable should be allocated. 4150 4151@itemize @bullet 4152@item 4153Global register variables reserve registers throughout the program. 4154This may be useful in programs such as programming language 4155interpreters which have a couple of global variables that are accessed 4156very often. 4157 4158@item 4159Local register variables in specific registers do not reserve the 4160registers. The compiler's data flow analysis is capable of determining 4161where the specified registers contain live values, and where they are 4162available for other uses. Stores into local register variables may be deleted 4163when they appear to be dead according to dataflow analysis. References 4164to local register variables may be deleted or moved or simplified. 4165 4166These local variables are sometimes convenient for use with the extended 4167@code{asm} feature (@pxref{Extended Asm}), if you want to write one 4168output of the assembler instruction directly into a particular register. 4169(This will work provided the register you specify fits the constraints 4170specified for that operand in the @code{asm}.) 4171@end itemize 4172 4173@menu 4174* Global Reg Vars:: 4175* Local Reg Vars:: 4176@end menu 4177 4178@node Global Reg Vars 4179@subsection Defining Global Register Variables 4180@cindex global register variables 4181@cindex registers, global variables in 4182 4183You can define a global register variable in GNU C like this: 4184 4185@example 4186register int *foo asm ("a5"); 4187@end example 4188 4189@noindent 4190Here @code{a5} is the name of the register which should be used. Choose a 4191register which is normally saved and restored by function calls on your 4192machine, so that library routines will not clobber it. 4193 4194Naturally the register name is cpu-dependent, so you would need to 4195conditionalize your program according to cpu type. The register 4196@code{a5} would be a good choice on a 68000 for a variable of pointer 4197type. On machines with register windows, be sure to choose a ``global'' 4198register that is not affected magically by the function call mechanism. 4199 4200In addition, operating systems on one type of cpu may differ in how they 4201name the registers; then you would need additional conditionals. For 4202example, some 68000 operating systems call this register @code{%a5}. 4203 4204Eventually there may be a way of asking the compiler to choose a register 4205automatically, but first we need to figure out how it should choose and 4206how to enable you to guide the choice. No solution is evident. 4207 4208Defining a global register variable in a certain register reserves that 4209register entirely for this use, at least within the current compilation. 4210The register will not be allocated for any other purpose in the functions 4211in the current compilation. The register will not be saved and restored by 4212these functions. Stores into this register are never deleted even if they 4213would appear to be dead, but references may be deleted or moved or 4214simplified. 4215 4216It is not safe to access the global register variables from signal 4217handlers, or from more than one thread of control, because the system 4218library routines may temporarily use the register for other things (unless 4219you recompile them specially for the task at hand). 4220 4221@cindex @code{qsort}, and global register variables 4222It is not safe for one function that uses a global register variable to 4223call another such function @code{foo} by way of a third function 4224@code{lose} that was compiled without knowledge of this variable (i.e.@: in a 4225different source file in which the variable wasn't declared). This is 4226because @code{lose} might save the register and put some other value there. 4227For example, you can't expect a global register variable to be available in 4228the comparison-function that you pass to @code{qsort}, since @code{qsort} 4229might have put something else in that register. (If you are prepared to 4230recompile @code{qsort} with the same global register variable, you can 4231solve this problem.) 4232 4233If you want to recompile @code{qsort} or other source files which do not 4234actually use your global register variable, so that they will not use that 4235register for any other purpose, then it suffices to specify the compiler 4236option @option{-ffixed-@var{reg}}. You need not actually add a global 4237register declaration to their source code. 4238 4239A function which can alter the value of a global register variable cannot 4240safely be called from a function compiled without this variable, because it 4241could clobber the value the caller expects to find there on return. 4242Therefore, the function which is the entry point into the part of the 4243program that uses the global register variable must explicitly save and 4244restore the value which belongs to its caller. 4245 4246@cindex register variable after @code{longjmp} 4247@cindex global register after @code{longjmp} 4248@cindex value after @code{longjmp} 4249@findex longjmp 4250@findex setjmp 4251On most machines, @code{longjmp} will restore to each global register 4252variable the value it had at the time of the @code{setjmp}. On some 4253machines, however, @code{longjmp} will not change the value of global 4254register variables. To be portable, the function that called @code{setjmp} 4255should make other arrangements to save the values of the global register 4256variables, and to restore them in a @code{longjmp}. This way, the same 4257thing will happen regardless of what @code{longjmp} does. 4258 4259All global register variable declarations must precede all function 4260definitions. If such a declaration could appear after function 4261definitions, the declaration would be too late to prevent the register from 4262being used for other purposes in the preceding functions. 4263 4264Global register variables may not have initial values, because an 4265executable file has no means to supply initial contents for a register. 4266 4267On the SPARC, there are reports that g3 @dots{} g7 are suitable 4268registers, but certain library functions, such as @code{getwd}, as well 4269as the subroutines for division and remainder, modify g3 and g4. g1 and 4270g2 are local temporaries. 4271 4272On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7. 4273Of course, it will not do to use more than a few of those. 4274 4275@node Local Reg Vars 4276@subsection Specifying Registers for Local Variables 4277@cindex local variables, specifying registers 4278@cindex specifying registers for local variables 4279@cindex registers for local variables 4280 4281You can define a local register variable with a specified register 4282like this: 4283 4284@example 4285register int *foo asm ("a5"); 4286@end example 4287 4288@noindent 4289Here @code{a5} is the name of the register which should be used. Note 4290that this is the same syntax used for defining global register 4291variables, but for a local variable it would appear within a function. 4292 4293Naturally the register name is cpu-dependent, but this is not a 4294problem, since specific registers are most often useful with explicit 4295assembler instructions (@pxref{Extended Asm}). Both of these things 4296generally require that you conditionalize your program according to 4297cpu type. 4298 4299In addition, operating systems on one type of cpu may differ in how they 4300name the registers; then you would need additional conditionals. For 4301example, some 68000 operating systems call this register @code{%a5}. 4302 4303Defining such a register variable does not reserve the register; it 4304remains available for other uses in places where flow control determines 4305the variable's value is not live. However, these registers are made 4306unavailable for use in the reload pass; excessive use of this feature 4307leaves the compiler too few available registers to compile certain 4308functions. 4309 4310This option does not guarantee that GCC will generate code that has 4311this variable in the register you specify at all times. You may not 4312code an explicit reference to this register in an @code{asm} statement 4313and assume it will always refer to this variable. 4314 4315Stores into local register variables may be deleted when they appear to be dead 4316according to dataflow analysis. References to local register variables may 4317be deleted or moved or simplified. 4318 4319@node Alternate Keywords 4320@section Alternate Keywords 4321@cindex alternate keywords 4322@cindex keywords, alternate 4323 4324@option{-ansi} and the various @option{-std} options disable certain 4325keywords. This causes trouble when you want to use GNU C extensions, or 4326a general-purpose header file that should be usable by all programs, 4327including ISO C programs. The keywords @code{asm}, @code{typeof} and 4328@code{inline} are not available in programs compiled with 4329@option{-ansi} or @option{-std} (although @code{inline} can be used in a 4330program compiled with @option{-std=c99}). The ISO C99 keyword 4331@code{restrict} is only available when @option{-std=gnu99} (which will 4332eventually be the default) or @option{-std=c99} (or the equivalent 4333@option{-std=iso9899:1999}) is used. 4334 4335The way to solve these problems is to put @samp{__} at the beginning and 4336end of each problematical keyword. For example, use @code{__asm__} 4337instead of @code{asm}, and @code{__inline__} instead of @code{inline}. 4338 4339Other C compilers won't accept these alternative keywords; if you want to 4340compile with another compiler, you can define the alternate keywords as 4341macros to replace them with the customary keywords. It looks like this: 4342 4343@example 4344#ifndef __GNUC__ 4345#define __asm__ asm 4346#endif 4347@end example 4348 4349@findex __extension__ 4350@opindex pedantic 4351@option{-pedantic} and other options cause warnings for many GNU C extensions. 4352You can 4353prevent such warnings within one expression by writing 4354@code{__extension__} before the expression. @code{__extension__} has no 4355effect aside from this. 4356 4357@node Incomplete Enums 4358@section Incomplete @code{enum} Types 4359 4360You can define an @code{enum} tag without specifying its possible values. 4361This results in an incomplete type, much like what you get if you write 4362@code{struct foo} without describing the elements. A later declaration 4363which does specify the possible values completes the type. 4364 4365You can't allocate variables or storage using the type while it is 4366incomplete. However, you can work with pointers to that type. 4367 4368This extension may not be very useful, but it makes the handling of 4369@code{enum} more consistent with the way @code{struct} and @code{union} 4370are handled. 4371 4372This extension is not supported by GNU C++. 4373 4374@node Function Names 4375@section Function Names as Strings 4376@cindex @code{__FUNCTION__} identifier 4377@cindex @code{__PRETTY_FUNCTION__} identifier 4378@cindex @code{__func__} identifier 4379 4380GCC predefines two magic identifiers to hold the name of the current 4381function. The identifier @code{__FUNCTION__} holds the name of the function 4382as it appears in the source. The identifier @code{__PRETTY_FUNCTION__} 4383holds the name of the function pretty printed in a language specific 4384fashion. 4385 4386These names are always the same in a C function, but in a C++ function 4387they may be different. For example, this program: 4388 4389@smallexample 4390extern "C" @{ 4391extern int printf (char *, ...); 4392@} 4393 4394class a @{ 4395 public: 4396 sub (int i) 4397 @{ 4398 printf ("__FUNCTION__ = %s\n", __FUNCTION__); 4399 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); 4400 @} 4401@}; 4402 4403int 4404main (void) 4405@{ 4406 a ax; 4407 ax.sub (0); 4408 return 0; 4409@} 4410@end smallexample 4411 4412@noindent 4413gives this output: 4414 4415@smallexample 4416__FUNCTION__ = sub 4417__PRETTY_FUNCTION__ = int a::sub (int) 4418@end smallexample 4419 4420The compiler automagically replaces the identifiers with a string 4421literal containing the appropriate name. Thus, they are neither 4422preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor 4423variables. This means that they catenate with other string literals, and 4424that they can be used to initialize char arrays. For example 4425 4426@smallexample 4427char here[] = "Function " __FUNCTION__ " in " __FILE__; 4428@end smallexample 4429 4430On the other hand, @samp{#ifdef __FUNCTION__} does not have any special 4431meaning inside a function, since the preprocessor does not do anything 4432special with the identifier @code{__FUNCTION__}. 4433 4434Note that these semantics are deprecated, and that GCC 3.2 will handle 4435@code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} the same way as 4436@code{__func__}. @code{__func__} is defined by the ISO standard C99: 4437 4438@display 4439The identifier @code{__func__} is implicitly declared by the translator 4440as if, immediately following the opening brace of each function 4441definition, the declaration 4442 4443@smallexample 4444static const char __func__[] = "function-name"; 4445@end smallexample 4446 4447appeared, where function-name is the name of the lexically-enclosing 4448function. This name is the unadorned name of the function. 4449@end display 4450 4451By this definition, @code{__func__} is a variable, not a string literal. 4452In particular, @code{__func__} does not catenate with other string 4453literals. 4454 4455In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are 4456variables, declared in the same way as @code{__func__}. 4457 4458@node Return Address 4459@section Getting the Return or Frame Address of a Function 4460 4461These functions may be used to get information about the callers of a 4462function. 4463 4464@deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level}) 4465This function returns the return address of the current function, or of 4466one of its callers. The @var{level} argument is number of frames to 4467scan up the call stack. A value of @code{0} yields the return address 4468of the current function, a value of @code{1} yields the return address 4469of the caller of the current function, and so forth. When inlining 4470the expected behavior is that the function will return the address of 4471the function that will be returned to. To work around this behavior use 4472the @code{noinline} function attribute. 4473 4474The @var{level} argument must be a constant integer. 4475 4476On some machines it may be impossible to determine the return address of 4477any function other than the current one; in such cases, or when the top 4478of the stack has been reached, this function will return @code{0} or a 4479random value. In addition, @code{__builtin_frame_address} may be used 4480to determine if the top of the stack has been reached. 4481 4482This function should only be used with a nonzero argument for debugging 4483purposes. 4484@end deftypefn 4485 4486@deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level}) 4487This function is similar to @code{__builtin_return_address}, but it 4488returns the address of the function frame rather than the return address 4489of the function. Calling @code{__builtin_frame_address} with a value of 4490@code{0} yields the frame address of the current function, a value of 4491@code{1} yields the frame address of the caller of the current function, 4492and so forth. 4493 4494The frame is the area on the stack which holds local variables and saved 4495registers. The frame address is normally the address of the first word 4496pushed on to the stack by the function. However, the exact definition 4497depends upon the processor and the calling convention. If the processor 4498has a dedicated frame pointer register, and the function has a frame, 4499then @code{__builtin_frame_address} will return the value of the frame 4500pointer register. 4501 4502On some machines it may be impossible to determine the frame address of 4503any function other than the current one; in such cases, or when the top 4504of the stack has been reached, this function will return @code{0} if 4505the first frame pointer is properly initialized by the startup code. 4506 4507This function should only be used with a nonzero argument for debugging 4508purposes. 4509@end deftypefn 4510 4511@node Vector Extensions 4512@section Using vector instructions through built-in functions 4513 4514On some targets, the instruction set contains SIMD vector instructions that 4515operate on multiple values contained in one large register at the same time. 4516For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used 4517this way. 4518 4519The first step in using these extensions is to provide the necessary data 4520types. This should be done using an appropriate @code{typedef}: 4521 4522@example 4523typedef int v4si __attribute__ ((mode(V4SI))); 4524@end example 4525 4526The base type @code{int} is effectively ignored by the compiler, the 4527actual properties of the new type @code{v4si} are defined by the 4528@code{__attribute__}. It defines the machine mode to be used; for vector 4529types these have the form @code{V@var{n}@var{B}}; @var{n} should be the 4530number of elements in the vector, and @var{B} should be the base mode of the 4531individual elements. The following can be used as base modes: 4532 4533@table @code 4534@item QI 4535An integer that is as wide as the smallest addressable unit, usually 8 bits. 4536@item HI 4537An integer, twice as wide as a QI mode integer, usually 16 bits. 4538@item SI 4539An integer, four times as wide as a QI mode integer, usually 32 bits. 4540@item DI 4541An integer, eight times as wide as a QI mode integer, usually 64 bits. 4542@item SF 4543A floating point value, as wide as a SI mode integer, usually 32 bits. 4544@item DF 4545A floating point value, as wide as a DI mode integer, usually 64 bits. 4546@end table 4547 4548Specifying a combination that is not valid for the current architecture 4549will cause gcc to synthesize the instructions using a narrower mode. 4550For example, if you specify a variable of type @code{V4SI} and your 4551architecture does not allow for this specific SIMD type, gcc will 4552produce code that uses 4 @code{SIs}. 4553 4554The types defined in this manner can be used with a subset of normal C 4555operations. Currently, gcc will allow using the following operators on 4556these types: @code{+, -, *, /, unary minus}@. 4557 4558The operations behave like C++ @code{valarrays}. Addition is defined as 4559the addition of the corresponding elements of the operands. For 4560example, in the code below, each of the 4 elements in @var{a} will be 4561added to the corresponding 4 elements in @var{b} and the resulting 4562vector will be stored in @var{c}. 4563 4564@example 4565typedef int v4si __attribute__ ((mode(V4SI))); 4566 4567v4si a, b, c; 4568 4569c = a + b; 4570@end example 4571 4572Subtraction, multiplication, and division operate in a similar manner. 4573Likewise, the result of using the unary minus operator on a vector type 4574is a vector whose elements are the negative value of the corresponding 4575elements in the operand. 4576 4577You can declare variables and use them in function calls and returns, as 4578well as in assignments and some casts. You can specify a vector type as 4579a return type for a function. Vector types can also be used as function 4580arguments. It is possible to cast from one vector type to another, 4581provided they are of the same size (in fact, you can also cast vectors 4582to and from other datatypes of the same size). 4583 4584You cannot operate between vectors of different lengths or different 4585signedness without a cast. 4586 4587A port that supports hardware vector operations, usually provides a set 4588of built-in functions that can be used to operate on vectors. For 4589example, a function to add two vectors and multiply the result by a 4590third could look like this: 4591 4592@example 4593v4si f (v4si a, v4si b, v4si c) 4594@{ 4595 v4si tmp = __builtin_addv4si (a, b); 4596 return __builtin_mulv4si (tmp, c); 4597@} 4598 4599@end example 4600 4601@node Other Builtins 4602@section Other built-in functions provided by GCC 4603@cindex built-in functions 4604@findex __builtin_isgreater 4605@findex __builtin_isgreaterequal 4606@findex __builtin_isless 4607@findex __builtin_islessequal 4608@findex __builtin_islessgreater 4609@findex __builtin_isunordered 4610@findex abort 4611@findex abs 4612@findex alloca 4613@findex bcmp 4614@findex bzero 4615@findex cimag 4616@findex cimagf 4617@findex cimagl 4618@findex conj 4619@findex conjf 4620@findex conjl 4621@findex cos 4622@findex cosf 4623@findex cosl 4624@findex creal 4625@findex crealf 4626@findex creall 4627@findex exit 4628@findex _exit 4629@findex _Exit 4630@findex exp 4631@findex expf 4632@findex expl 4633@findex fabs 4634@findex fabsf 4635@findex fabsl 4636@findex ffs 4637@findex fprintf 4638@findex fprintf_unlocked 4639@findex fputs 4640@findex fputs_unlocked 4641@findex imaxabs 4642@findex index 4643@findex labs 4644@findex llabs 4645@findex log 4646@findex logf 4647@findex logl 4648@findex memcmp 4649@findex memcpy 4650@findex memset 4651@findex printf 4652@findex printf_unlocked 4653@findex putchar 4654@findex puts 4655@findex rindex 4656@findex scanf 4657@findex sin 4658@findex sinf 4659@findex sinl 4660@findex snprintf 4661@findex sprintf 4662@findex sqrt 4663@findex sqrtf 4664@findex sqrtl 4665@findex sscanf 4666@findex strcat 4667@findex strchr 4668@findex strcmp 4669@findex strcpy 4670@findex strcspn 4671@findex strlen 4672@findex strncat 4673@findex strncmp 4674@findex strncpy 4675@findex strpbrk 4676@findex strrchr 4677@findex strspn 4678@findex strstr 4679@findex vprintf 4680@findex vscanf 4681@findex vsnprintf 4682@findex vsprintf 4683@findex vsscanf 4684 4685GCC provides a large number of built-in functions other than the ones 4686mentioned above. Some of these are for internal use in the processing 4687of exceptions or variable-length argument lists and will not be 4688documented here because they may change from time to time; we do not 4689recommend general use of these functions. 4690 4691The remaining functions are provided for optimization purposes. 4692 4693@opindex fno-builtin 4694GCC includes built-in versions of many of the functions in the standard 4695C library. The versions prefixed with @code{__builtin_} will always be 4696treated as having the same meaning as the C library function even if you 4697specify the @option{-fno-builtin} option. (@pxref{C Dialect Options}) 4698Many of these functions are only optimized in certain cases; if they are 4699not optimized in a particular case, a call to the library function will 4700be emitted. 4701 4702@opindex ansi 4703@opindex std 4704The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit} 4705are recognized and presumed not to return, but otherwise are not built 4706in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi}, 4707@option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in 4708strict C89 mode (@option{-ansi} or @option{-std=c89}). All these functions 4709have corresponding versions prefixed with @code{__builtin_}, which may be 4710used even in strict C89 mode. 4711 4712Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp}, 4713@code{bzero}, @code{index}, @code{rindex}, @code{ffs}, @code{fputs_unlocked}, 4714@code{printf_unlocked} and @code{fprintf_unlocked} may be handled as 4715built-in functions. All these functions have corresponding versions 4716prefixed with @code{__builtin_}, which may be used even in strict C89 4717mode. 4718 4719The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl}, 4720@code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf}, 4721@code{cimagl}, @code{imaxabs}, @code{llabs}, @code{snprintf}, 4722@code{vscanf}, @code{vsnprintf} and @code{vsscanf} are handled as built-in 4723functions except in strict ISO C90 mode. There are also built-in 4724versions of the ISO C99 functions @code{cosf}, @code{cosl}, 4725@code{expf}, @code{expl}, @code{fabsf}, @code{fabsl}, 4726@code{logf}, @code{logl}, @code{sinf}, @code{sinl}, @code{sqrtf}, and 4727@code{sqrtl}, that are recognized in any mode since ISO C90 reserves 4728these names for the purpose to which ISO C99 puts them. All these 4729functions have corresponding versions prefixed with @code{__builtin_}. 4730 4731The ISO C90 functions @code{abs}, @code{cos}, @code{exp}, @code{fabs}, 4732@code{fprintf}, @code{fputs}, @code{labs}, @code{log}, 4733@code{memcmp}, @code{memcpy}, 4734@code{memset}, @code{printf}, @code{putchar}, @code{puts}, @code{scanf}, 4735@code{sin}, @code{snprintf}, @code{sprintf}, @code{sqrt}, @code{sscanf}, 4736@code{strcat}, 4737@code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn}, 4738@code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy}, 4739@code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr}, 4740@code{vprintf} and @code{vsprintf} are all 4741recognized as built-in functions unless @option{-fno-builtin} is 4742specified (or @option{-fno-builtin-@var{function}} is specified for an 4743individual function). All of these functions have corresponding 4744versions prefixed with @code{__builtin_}. 4745 4746GCC provides built-in versions of the ISO C99 floating point comparison 4747macros that avoid raising exceptions for unordered operands. They have 4748the same names as the standard macros ( @code{isgreater}, 4749@code{isgreaterequal}, @code{isless}, @code{islessequal}, 4750@code{islessgreater}, and @code{isunordered}) , with @code{__builtin_} 4751prefixed. We intend for a library implementor to be able to simply 4752@code{#define} each standard macro to its built-in equivalent. 4753 4754@deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2}) 4755 4756You can use the built-in function @code{__builtin_types_compatible_p} to 4757determine whether two types are the same. 4758 4759This built-in function returns 1 if the unqualified versions of the 4760types @var{type1} and @var{type2} (which are types, not expressions) are 4761compatible, 0 otherwise. The result of this built-in function can be 4762used in integer constant expressions. 4763 4764This built-in function ignores top level qualifiers (e.g., @code{const}, 4765@code{volatile}). For example, @code{int} is equivalent to @code{const 4766int}. 4767 4768The type @code{int[]} and @code{int[5]} are compatible. On the other 4769hand, @code{int} and @code{char *} are not compatible, even if the size 4770of their types, on the particular architecture are the same. Also, the 4771amount of pointer indirection is taken into account when determining 4772similarity. Consequently, @code{short *} is not similar to 4773@code{short **}. Furthermore, two types that are typedefed are 4774considered compatible if their underlying types are compatible. 4775 4776An @code{enum} type is considered to be compatible with another 4777@code{enum} type. For example, @code{enum @{foo, bar@}} is similar to 4778@code{enum @{hot, dog@}}. 4779 4780You would typically use this function in code whose execution varies 4781depending on the arguments' types. For example: 4782 4783@smallexample 4784#define foo(x) \ 4785 (@{ \ 4786 typeof (x) tmp; \ 4787 if (__builtin_types_compatible_p (typeof (x), long double)) \ 4788 tmp = foo_long_double (tmp); \ 4789 else if (__builtin_types_compatible_p (typeof (x), double)) \ 4790 tmp = foo_double (tmp); \ 4791 else if (__builtin_types_compatible_p (typeof (x), float)) \ 4792 tmp = foo_float (tmp); \ 4793 else \ 4794 abort (); \ 4795 tmp; \ 4796 @}) 4797@end smallexample 4798 4799@emph{Note:} This construct is only available for C. 4800 4801@end deftypefn 4802 4803@deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2}) 4804 4805You can use the built-in function @code{__builtin_choose_expr} to 4806evaluate code depending on the value of a constant expression. This 4807built-in function returns @var{exp1} if @var{const_exp}, which is a 4808constant expression that must be able to be determined at compile time, 4809is nonzero. Otherwise it returns 0. 4810 4811This built-in function is analogous to the @samp{? :} operator in C, 4812except that the expression returned has its type unaltered by promotion 4813rules. Also, the built-in function does not evaluate the expression 4814that was not chosen. For example, if @var{const_exp} evaluates to true, 4815@var{exp2} is not evaluated even if it has side-effects. 4816 4817This built-in function can return an lvalue if the chosen argument is an 4818lvalue. 4819 4820If @var{exp1} is returned, the return type is the same as @var{exp1}'s 4821type. Similarly, if @var{exp2} is returned, its return type is the same 4822as @var{exp2}. 4823 4824Example: 4825 4826@smallexample 4827#define foo(x) \ 4828 __builtin_choose_expr ( \ 4829 __builtin_types_compatible_p (typeof (x), double), \ 4830 foo_double (x), \ 4831 __builtin_choose_expr ( \ 4832 __builtin_types_compatible_p (typeof (x), float), \ 4833 foo_float (x), \ 4834 /* @r{The void expression results in a compile-time error} \ 4835 @r{when assigning the result to something.} */ \ 4836 (void)0)) 4837@end smallexample 4838 4839@emph{Note:} This construct is only available for C. Furthermore, the 4840unused expression (@var{exp1} or @var{exp2} depending on the value of 4841@var{const_exp}) may still generate syntax errors. This may change in 4842future revisions. 4843 4844@end deftypefn 4845 4846@deftypefn {Built-in Function} int __builtin_constant_p (@var{exp}) 4847You can use the built-in function @code{__builtin_constant_p} to 4848determine if a value is known to be constant at compile-time and hence 4849that GCC can perform constant-folding on expressions involving that 4850value. The argument of the function is the value to test. The function 4851returns the integer 1 if the argument is known to be a compile-time 4852constant and 0 if it is not known to be a compile-time constant. A 4853return of 0 does not indicate that the value is @emph{not} a constant, 4854but merely that GCC cannot prove it is a constant with the specified 4855value of the @option{-O} option. 4856 4857You would typically use this function in an embedded application where 4858memory was a critical resource. If you have some complex calculation, 4859you may want it to be folded if it involves constants, but need to call 4860a function if it does not. For example: 4861 4862@smallexample 4863#define Scale_Value(X) \ 4864 (__builtin_constant_p (X) \ 4865 ? ((X) * SCALE + OFFSET) : Scale (X)) 4866@end smallexample 4867 4868You may use this built-in function in either a macro or an inline 4869function. However, if you use it in an inlined function and pass an 4870argument of the function as the argument to the built-in, GCC will 4871never return 1 when you call the inline function with a string constant 4872or compound literal (@pxref{Compound Literals}) and will not return 1 4873when you pass a constant numeric value to the inline function unless you 4874specify the @option{-O} option. 4875 4876You may also use @code{__builtin_constant_p} in initializers for static 4877data. For instance, you can write 4878 4879@smallexample 4880static const int table[] = @{ 4881 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1, 4882 /* @r{@dots{}} */ 4883@}; 4884@end smallexample 4885 4886@noindent 4887This is an acceptable initializer even if @var{EXPRESSION} is not a 4888constant expression. GCC must be more conservative about evaluating the 4889built-in in this case, because it has no opportunity to perform 4890optimization. 4891 4892Previous versions of GCC did not accept this built-in in data 4893initializers. The earliest version where it is completely safe is 48943.0.1. 4895@end deftypefn 4896 4897@deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c}) 4898@opindex fprofile-arcs 4899You may use @code{__builtin_expect} to provide the compiler with 4900branch prediction information. In general, you should prefer to 4901use actual profile feedback for this (@option{-fprofile-arcs}), as 4902programmers are notoriously bad at predicting how their programs 4903actually perform. However, there are applications in which this 4904data is hard to collect. 4905 4906The return value is the value of @var{exp}, which should be an 4907integral expression. The value of @var{c} must be a compile-time 4908constant. The semantics of the built-in are that it is expected 4909that @var{exp} == @var{c}. For example: 4910 4911@smallexample 4912if (__builtin_expect (x, 0)) 4913 foo (); 4914@end smallexample 4915 4916@noindent 4917would indicate that we do not expect to call @code{foo}, since 4918we expect @code{x} to be zero. Since you are limited to integral 4919expressions for @var{exp}, you should use constructions such as 4920 4921@smallexample 4922if (__builtin_expect (ptr != NULL, 1)) 4923 error (); 4924@end smallexample 4925 4926@noindent 4927when testing pointer or floating-point values. 4928@end deftypefn 4929 4930@deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...) 4931This function is used to minimize cache-miss latency by moving data into 4932a cache before it is accessed. 4933You can insert calls to @code{__builtin_prefetch} into code for which 4934you know addresses of data in memory that is likely to be accessed soon. 4935If the target supports them, data prefetch instructions will be generated. 4936If the prefetch is done early enough before the access then the data will 4937be in the cache by the time it is accessed. 4938 4939The value of @var{addr} is the address of the memory to prefetch. 4940There are two optional arguments, @var{rw} and @var{locality}. 4941The value of @var{rw} is a compile-time constant one or zero; one 4942means that the prefetch is preparing for a write to the memory address 4943and zero, the default, means that the prefetch is preparing for a read. 4944The value @var{locality} must be a compile-time constant integer between 4945zero and three. A value of zero means that the data has no temporal 4946locality, so it need not be left in the cache after the access. A value 4947of three means that the data has a high degree of temporal locality and 4948should be left in all levels of cache possible. Values of one and two 4949mean, respectively, a low or moderate degree of temporal locality. The 4950default is three. 4951 4952@smallexample 4953for (i = 0; i < n; i++) 4954 @{ 4955 a[i] = a[i] + b[i]; 4956 __builtin_prefetch (&a[i+j], 1, 1); 4957 __builtin_prefetch (&b[i+j], 0, 1); 4958 /* @r{@dots{}} */ 4959 @} 4960@end smallexample 4961 4962Data prefetch does not generate faults if @var{addr} is invalid, but 4963the address expression itself must be valid. For example, a prefetch 4964of @code{p->next} will not fault if @code{p->next} is not a valid 4965address, but evaluation will fault if @code{p} is not a valid address. 4966 4967If the target does not support data prefetch, the address expression 4968is evaluated if it includes side effects but no other code is generated 4969and GCC does not issue a warning. 4970@end deftypefn 4971 4972@deftypefn {Built-in Function} double __builtin_huge_val (void) 4973Returns a positive infinity, if supported by the floating-point format, 4974else @code{DBL_MAX}. This function is suitable for implementing the 4975ISO C macro @code{HUGE_VAL}. 4976@end deftypefn 4977 4978@deftypefn {Built-in Function} float __builtin_huge_valf (void) 4979Similar to @code{__builtin_huge_val}, except the return type is @code{float}. 4980@end deftypefn 4981 4982@deftypefn {Built-in Function} {long double} __builtin_huge_vall (void) 4983Similar to @code{__builtin_huge_val}, except the return 4984type is @code{long double}. 4985@end deftypefn 4986 4987@deftypefn {Built-in Function} double __builtin_inf (void) 4988Similar to @code{__builtin_huge_val}, except a warning is generated 4989if the target floating-point format does not support infinities. 4990This function is suitable for implementing the ISO C99 macro @code{INFINITY}. 4991@end deftypefn 4992 4993@deftypefn {Built-in Function} float __builtin_inff (void) 4994Similar to @code{__builtin_inf}, except the return type is @code{float}. 4995@end deftypefn 4996 4997@deftypefn {Built-in Function} {long double} __builtin_infl (void) 4998Similar to @code{__builtin_inf}, except the return 4999type is @code{long double}. 5000@end deftypefn 5001 5002@deftypefn {Built-in Function} double __builtin_nan (const char *str) 5003This is an implementation of the ISO C99 function @code{nan}. 5004 5005Since ISO C99 defines this function in terms of @code{strtod}, which we 5006do not implement, a description of the parsing is in order. The string 5007is parsed as by @code{strtol}; that is, the base is recognized by 5008leading @samp{0} or @samp{0x} prefixes. The number parsed is placed 5009in the significand such that the least significant bit of the number 5010is at the least significant bit of the significand. The number is 5011truncated to fit the significand field provided. The significand is 5012forced to be a quiet NaN. 5013 5014This function, if given a string literal, is evaluated early enough 5015that it is considered a compile-time constant. 5016@end deftypefn 5017 5018@deftypefn {Built-in Function} float __builtin_nanf (const char *str) 5019Similar to @code{__builtin_nan}, except the return type is @code{float}. 5020@end deftypefn 5021 5022@deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str) 5023Similar to @code{__builtin_nan}, except the return type is @code{long double}. 5024@end deftypefn 5025 5026@deftypefn {Built-in Function} double __builtin_nans (const char *str) 5027Similar to @code{__builtin_nan}, except the significand is forced 5028to be a signaling NaN. The @code{nans} function is proposed by 5029@uref{http://std.dkuug.dk/JTC1/SC22/WG14/www/docs/n965.htm,,WG14 N965}. 5030@end deftypefn 5031 5032@deftypefn {Built-in Function} float __builtin_nansf (const char *str) 5033Similar to @code{__builtin_nans}, except the return type is @code{float}. 5034@end deftypefn 5035 5036@deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str) 5037Similar to @code{__builtin_nans}, except the return type is @code{long double}. 5038@end deftypefn 5039 5040@node Target Builtins 5041@section Built-in Functions Specific to Particular Target Machines 5042 5043On some target machines, GCC supports many built-in functions specific 5044to those machines. Generally these generate calls to specific machine 5045instructions, but allow the compiler to schedule those calls. 5046 5047@menu 5048* Alpha Built-in Functions:: 5049* X86 Built-in Functions:: 5050* PowerPC AltiVec Built-in Functions:: 5051@end menu 5052 5053@node Alpha Built-in Functions 5054@subsection Alpha Built-in Functions 5055 5056These built-in functions are available for the Alpha family of 5057processors, depending on the command-line switches used. 5058 5059The following built-in functions are always available. They 5060all generate the machine instruction that is part of the name. 5061 5062@example 5063long __builtin_alpha_implver (void) 5064long __builtin_alpha_rpcc (void) 5065long __builtin_alpha_amask (long) 5066long __builtin_alpha_cmpbge (long, long) 5067long __builtin_alpha_extbl (long, long) 5068long __builtin_alpha_extwl (long, long) 5069long __builtin_alpha_extll (long, long) 5070long __builtin_alpha_extql (long, long) 5071long __builtin_alpha_extwh (long, long) 5072long __builtin_alpha_extlh (long, long) 5073long __builtin_alpha_extqh (long, long) 5074long __builtin_alpha_insbl (long, long) 5075long __builtin_alpha_inswl (long, long) 5076long __builtin_alpha_insll (long, long) 5077long __builtin_alpha_insql (long, long) 5078long __builtin_alpha_inswh (long, long) 5079long __builtin_alpha_inslh (long, long) 5080long __builtin_alpha_insqh (long, long) 5081long __builtin_alpha_mskbl (long, long) 5082long __builtin_alpha_mskwl (long, long) 5083long __builtin_alpha_mskll (long, long) 5084long __builtin_alpha_mskql (long, long) 5085long __builtin_alpha_mskwh (long, long) 5086long __builtin_alpha_msklh (long, long) 5087long __builtin_alpha_mskqh (long, long) 5088long __builtin_alpha_umulh (long, long) 5089long __builtin_alpha_zap (long, long) 5090long __builtin_alpha_zapnot (long, long) 5091@end example 5092 5093The following built-in functions are always with @option{-mmax} 5094or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or 5095later. They all generate the machine instruction that is part 5096of the name. 5097 5098@example 5099long __builtin_alpha_pklb (long) 5100long __builtin_alpha_pkwb (long) 5101long __builtin_alpha_unpkbl (long) 5102long __builtin_alpha_unpkbw (long) 5103long __builtin_alpha_minub8 (long, long) 5104long __builtin_alpha_minsb8 (long, long) 5105long __builtin_alpha_minuw4 (long, long) 5106long __builtin_alpha_minsw4 (long, long) 5107long __builtin_alpha_maxub8 (long, long) 5108long __builtin_alpha_maxsb8 (long, long) 5109long __builtin_alpha_maxuw4 (long, long) 5110long __builtin_alpha_maxsw4 (long, long) 5111long __builtin_alpha_perr (long, long) 5112@end example 5113 5114The following built-in functions are always with @option{-mcix} 5115or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or 5116later. They all generate the machine instruction that is part 5117of the name. 5118 5119@example 5120long __builtin_alpha_cttz (long) 5121long __builtin_alpha_ctlz (long) 5122long __builtin_alpha_ctpop (long) 5123@end example 5124 5125The following builtins are available on systems that use the OSF/1 5126PALcode. Normally they invoke the @code{rduniq} and @code{wruniq} 5127PAL calls, but when invoked with @option{-mtls-kernel}, they invoke 5128@code{rdval} and @code{wrval}. 5129 5130@example 5131void *__builtin_thread_pointer (void) 5132void __builtin_set_thread_pointer (void *) 5133@end example 5134 5135@node X86 Built-in Functions 5136@subsection X86 Built-in Functions 5137 5138These built-in functions are available for the i386 and x86-64 family 5139of computers, depending on the command-line switches used. 5140 5141The following machine modes are available for use with MMX built-in functions 5142(@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers, 5143@code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a 5144vector of eight 8-bit integers. Some of the built-in functions operate on 5145MMX registers as a whole 64-bit entity, these use @code{DI} as their mode. 5146 5147If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector 5148of two 32-bit floating point values. 5149 5150If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit 5151floating point values. Some instructions use a vector of four 32-bit 5152integers, these use @code{V4SI}. Finally, some instructions operate on an 5153entire vector register, interpreting it as a 128-bit integer, these use mode 5154@code{TI}. 5155 5156The following built-in functions are made available by @option{-mmmx}. 5157All of them generate the machine instruction that is part of the name. 5158 5159@example 5160v8qi __builtin_ia32_paddb (v8qi, v8qi) 5161v4hi __builtin_ia32_paddw (v4hi, v4hi) 5162v2si __builtin_ia32_paddd (v2si, v2si) 5163v8qi __builtin_ia32_psubb (v8qi, v8qi) 5164v4hi __builtin_ia32_psubw (v4hi, v4hi) 5165v2si __builtin_ia32_psubd (v2si, v2si) 5166v8qi __builtin_ia32_paddsb (v8qi, v8qi) 5167v4hi __builtin_ia32_paddsw (v4hi, v4hi) 5168v8qi __builtin_ia32_psubsb (v8qi, v8qi) 5169v4hi __builtin_ia32_psubsw (v4hi, v4hi) 5170v8qi __builtin_ia32_paddusb (v8qi, v8qi) 5171v4hi __builtin_ia32_paddusw (v4hi, v4hi) 5172v8qi __builtin_ia32_psubusb (v8qi, v8qi) 5173v4hi __builtin_ia32_psubusw (v4hi, v4hi) 5174v4hi __builtin_ia32_pmullw (v4hi, v4hi) 5175v4hi __builtin_ia32_pmulhw (v4hi, v4hi) 5176di __builtin_ia32_pand (di, di) 5177di __builtin_ia32_pandn (di,di) 5178di __builtin_ia32_por (di, di) 5179di __builtin_ia32_pxor (di, di) 5180v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi) 5181v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi) 5182v2si __builtin_ia32_pcmpeqd (v2si, v2si) 5183v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi) 5184v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi) 5185v2si __builtin_ia32_pcmpgtd (v2si, v2si) 5186v8qi __builtin_ia32_punpckhbw (v8qi, v8qi) 5187v4hi __builtin_ia32_punpckhwd (v4hi, v4hi) 5188v2si __builtin_ia32_punpckhdq (v2si, v2si) 5189v8qi __builtin_ia32_punpcklbw (v8qi, v8qi) 5190v4hi __builtin_ia32_punpcklwd (v4hi, v4hi) 5191v2si __builtin_ia32_punpckldq (v2si, v2si) 5192v8qi __builtin_ia32_packsswb (v4hi, v4hi) 5193v4hi __builtin_ia32_packssdw (v2si, v2si) 5194v8qi __builtin_ia32_packuswb (v4hi, v4hi) 5195@end example 5196 5197The following built-in functions are made available either with 5198@option{-msse}, or with a combination of @option{-m3dnow} and 5199@option{-march=athlon}. All of them generate the machine 5200instruction that is part of the name. 5201 5202@example 5203v4hi __builtin_ia32_pmulhuw (v4hi, v4hi) 5204v8qi __builtin_ia32_pavgb (v8qi, v8qi) 5205v4hi __builtin_ia32_pavgw (v4hi, v4hi) 5206v4hi __builtin_ia32_psadbw (v8qi, v8qi) 5207v8qi __builtin_ia32_pmaxub (v8qi, v8qi) 5208v4hi __builtin_ia32_pmaxsw (v4hi, v4hi) 5209v8qi __builtin_ia32_pminub (v8qi, v8qi) 5210v4hi __builtin_ia32_pminsw (v4hi, v4hi) 5211int __builtin_ia32_pextrw (v4hi, int) 5212v4hi __builtin_ia32_pinsrw (v4hi, int, int) 5213int __builtin_ia32_pmovmskb (v8qi) 5214void __builtin_ia32_maskmovq (v8qi, v8qi, char *) 5215void __builtin_ia32_movntq (di *, di) 5216void __builtin_ia32_sfence (void) 5217@end example 5218 5219The following built-in functions are available when @option{-msse} is used. 5220All of them generate the machine instruction that is part of the name. 5221 5222@example 5223int __builtin_ia32_comieq (v4sf, v4sf) 5224int __builtin_ia32_comineq (v4sf, v4sf) 5225int __builtin_ia32_comilt (v4sf, v4sf) 5226int __builtin_ia32_comile (v4sf, v4sf) 5227int __builtin_ia32_comigt (v4sf, v4sf) 5228int __builtin_ia32_comige (v4sf, v4sf) 5229int __builtin_ia32_ucomieq (v4sf, v4sf) 5230int __builtin_ia32_ucomineq (v4sf, v4sf) 5231int __builtin_ia32_ucomilt (v4sf, v4sf) 5232int __builtin_ia32_ucomile (v4sf, v4sf) 5233int __builtin_ia32_ucomigt (v4sf, v4sf) 5234int __builtin_ia32_ucomige (v4sf, v4sf) 5235v4sf __builtin_ia32_addps (v4sf, v4sf) 5236v4sf __builtin_ia32_subps (v4sf, v4sf) 5237v4sf __builtin_ia32_mulps (v4sf, v4sf) 5238v4sf __builtin_ia32_divps (v4sf, v4sf) 5239v4sf __builtin_ia32_addss (v4sf, v4sf) 5240v4sf __builtin_ia32_subss (v4sf, v4sf) 5241v4sf __builtin_ia32_mulss (v4sf, v4sf) 5242v4sf __builtin_ia32_divss (v4sf, v4sf) 5243v4si __builtin_ia32_cmpeqps (v4sf, v4sf) 5244v4si __builtin_ia32_cmpltps (v4sf, v4sf) 5245v4si __builtin_ia32_cmpleps (v4sf, v4sf) 5246v4si __builtin_ia32_cmpgtps (v4sf, v4sf) 5247v4si __builtin_ia32_cmpgeps (v4sf, v4sf) 5248v4si __builtin_ia32_cmpunordps (v4sf, v4sf) 5249v4si __builtin_ia32_cmpneqps (v4sf, v4sf) 5250v4si __builtin_ia32_cmpnltps (v4sf, v4sf) 5251v4si __builtin_ia32_cmpnleps (v4sf, v4sf) 5252v4si __builtin_ia32_cmpngtps (v4sf, v4sf) 5253v4si __builtin_ia32_cmpngeps (v4sf, v4sf) 5254v4si __builtin_ia32_cmpordps (v4sf, v4sf) 5255v4si __builtin_ia32_cmpeqss (v4sf, v4sf) 5256v4si __builtin_ia32_cmpltss (v4sf, v4sf) 5257v4si __builtin_ia32_cmpless (v4sf, v4sf) 5258v4si __builtin_ia32_cmpunordss (v4sf, v4sf) 5259v4si __builtin_ia32_cmpneqss (v4sf, v4sf) 5260v4si __builtin_ia32_cmpnlts (v4sf, v4sf) 5261v4si __builtin_ia32_cmpnless (v4sf, v4sf) 5262v4si __builtin_ia32_cmpordss (v4sf, v4sf) 5263v4sf __builtin_ia32_maxps (v4sf, v4sf) 5264v4sf __builtin_ia32_maxss (v4sf, v4sf) 5265v4sf __builtin_ia32_minps (v4sf, v4sf) 5266v4sf __builtin_ia32_minss (v4sf, v4sf) 5267v4sf __builtin_ia32_andps (v4sf, v4sf) 5268v4sf __builtin_ia32_andnps (v4sf, v4sf) 5269v4sf __builtin_ia32_orps (v4sf, v4sf) 5270v4sf __builtin_ia32_xorps (v4sf, v4sf) 5271v4sf __builtin_ia32_movss (v4sf, v4sf) 5272v4sf __builtin_ia32_movhlps (v4sf, v4sf) 5273v4sf __builtin_ia32_movlhps (v4sf, v4sf) 5274v4sf __builtin_ia32_unpckhps (v4sf, v4sf) 5275v4sf __builtin_ia32_unpcklps (v4sf, v4sf) 5276v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si) 5277v4sf __builtin_ia32_cvtsi2ss (v4sf, int) 5278v2si __builtin_ia32_cvtps2pi (v4sf) 5279int __builtin_ia32_cvtss2si (v4sf) 5280v2si __builtin_ia32_cvttps2pi (v4sf) 5281int __builtin_ia32_cvttss2si (v4sf) 5282v4sf __builtin_ia32_rcpps (v4sf) 5283v4sf __builtin_ia32_rsqrtps (v4sf) 5284v4sf __builtin_ia32_sqrtps (v4sf) 5285v4sf __builtin_ia32_rcpss (v4sf) 5286v4sf __builtin_ia32_rsqrtss (v4sf) 5287v4sf __builtin_ia32_sqrtss (v4sf) 5288v4sf __builtin_ia32_shufps (v4sf, v4sf, int) 5289void __builtin_ia32_movntps (float *, v4sf) 5290int __builtin_ia32_movmskps (v4sf) 5291@end example 5292 5293The following built-in functions are available when @option{-msse} is used. 5294 5295@table @code 5296@item v4sf __builtin_ia32_loadaps (float *) 5297Generates the @code{movaps} machine instruction as a load from memory. 5298@item void __builtin_ia32_storeaps (float *, v4sf) 5299Generates the @code{movaps} machine instruction as a store to memory. 5300@item v4sf __builtin_ia32_loadups (float *) 5301Generates the @code{movups} machine instruction as a load from memory. 5302@item void __builtin_ia32_storeups (float *, v4sf) 5303Generates the @code{movups} machine instruction as a store to memory. 5304@item v4sf __builtin_ia32_loadsss (float *) 5305Generates the @code{movss} machine instruction as a load from memory. 5306@item void __builtin_ia32_storess (float *, v4sf) 5307Generates the @code{movss} machine instruction as a store to memory. 5308@item v4sf __builtin_ia32_loadhps (v4sf, v2si *) 5309Generates the @code{movhps} machine instruction as a load from memory. 5310@item v4sf __builtin_ia32_loadlps (v4sf, v2si *) 5311Generates the @code{movlps} machine instruction as a load from memory 5312@item void __builtin_ia32_storehps (v4sf, v2si *) 5313Generates the @code{movhps} machine instruction as a store to memory. 5314@item void __builtin_ia32_storelps (v4sf, v2si *) 5315Generates the @code{movlps} machine instruction as a store to memory. 5316@end table 5317 5318The following built-in functions are available when @option{-mpni} is used. 5319All of them generate the machine instruction that is part of the name. 5320 5321@example 5322v2df __builtin_ia32_addsubpd (v2df, v2df) 5323v2df __builtin_ia32_addsubps (v2df, v2df) 5324v2df __builtin_ia32_haddpd (v2df, v2df) 5325v2df __builtin_ia32_haddps (v2df, v2df) 5326v2df __builtin_ia32_hsubpd (v2df, v2df) 5327v2df __builtin_ia32_hsubps (v2df, v2df) 5328v16qi __builtin_ia32_lddqu (char const *) 5329void __builtin_ia32_monitor (void *, unsigned int, unsigned int) 5330v2df __builtin_ia32_movddup (v2df) 5331v4sf __builtin_ia32_movshdup (v4sf) 5332v4sf __builtin_ia32_movsldup (v4sf) 5333void __builtin_ia32_mwait (unsigned int, unsigned int) 5334@end example 5335 5336The following built-in functions are available when @option{-mpni} is used. 5337 5338@table @code 5339@item v2df __builtin_ia32_loadddup (double const *) 5340Generates the @code{movddup} machine instruction as a load from memory. 5341@end table 5342 5343The following built-in functions are available when @option{-m3dnow} is used. 5344All of them generate the machine instruction that is part of the name. 5345 5346@example 5347void __builtin_ia32_femms (void) 5348v8qi __builtin_ia32_pavgusb (v8qi, v8qi) 5349v2si __builtin_ia32_pf2id (v2sf) 5350v2sf __builtin_ia32_pfacc (v2sf, v2sf) 5351v2sf __builtin_ia32_pfadd (v2sf, v2sf) 5352v2si __builtin_ia32_pfcmpeq (v2sf, v2sf) 5353v2si __builtin_ia32_pfcmpge (v2sf, v2sf) 5354v2si __builtin_ia32_pfcmpgt (v2sf, v2sf) 5355v2sf __builtin_ia32_pfmax (v2sf, v2sf) 5356v2sf __builtin_ia32_pfmin (v2sf, v2sf) 5357v2sf __builtin_ia32_pfmul (v2sf, v2sf) 5358v2sf __builtin_ia32_pfrcp (v2sf) 5359v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf) 5360v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf) 5361v2sf __builtin_ia32_pfrsqrt (v2sf) 5362v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf) 5363v2sf __builtin_ia32_pfsub (v2sf, v2sf) 5364v2sf __builtin_ia32_pfsubr (v2sf, v2sf) 5365v2sf __builtin_ia32_pi2fd (v2si) 5366v4hi __builtin_ia32_pmulhrw (v4hi, v4hi) 5367@end example 5368 5369The following built-in functions are available when both @option{-m3dnow} 5370and @option{-march=athlon} are used. All of them generate the machine 5371instruction that is part of the name. 5372 5373@example 5374v2si __builtin_ia32_pf2iw (v2sf) 5375v2sf __builtin_ia32_pfnacc (v2sf, v2sf) 5376v2sf __builtin_ia32_pfpnacc (v2sf, v2sf) 5377v2sf __builtin_ia32_pi2fw (v2si) 5378v2sf __builtin_ia32_pswapdsf (v2sf) 5379v2si __builtin_ia32_pswapdsi (v2si) 5380@end example 5381 5382@node PowerPC AltiVec Built-in Functions 5383@subsection PowerPC AltiVec Built-in Functions 5384 5385These built-in functions are available for the PowerPC family 5386of computers, depending on the command-line switches used. 5387 5388The following machine modes are available for use with AltiVec built-in 5389functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four 539032-bit integers, @code{V4SF} for a vector of four 32-bit floating point 5391numbers, @code{V8HI} for a vector of eight 16-bit integers, and 5392@code{V16QI} for a vector of sixteen 8-bit integers. 5393 5394The following functions are made available by including 5395@code{<altivec.h>} and using @option{-maltivec} and 5396@option{-mabi=altivec}. The functions implement the functionality 5397described in Motorola's AltiVec Programming Interface Manual. 5398 5399There are a few differences from Motorola's documentation and GCC's 5400implementation. Vector constants are done with curly braces (not 5401parentheses). Vector initializers require no casts if the vector 5402constant is of the same type as the variable it is initializing. The 5403@code{vector bool} type is deprecated and will be discontinued in 5404further revisions. Use @code{vector signed} instead. If @code{signed} 5405or @code{unsigned} is omitted, the vector type will default to 5406@code{signed}. Lastly, all overloaded functions are implemented with macros 5407for the C implementation. So code the following example will not work: 5408 5409@smallexample 5410 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo); 5411@end smallexample 5412 5413Since vec_add is a macro, the vector constant in the above example will 5414be treated as four different arguments. Wrap the entire argument in 5415parentheses for this to work. The C++ implementation does not use 5416macros. 5417 5418@emph{Note:} Only the @code{<altivec.h>} interface is supported. 5419Internally, GCC uses built-in functions to achieve the functionality in 5420the aforementioned header file, but they are not supported and are 5421subject to change without notice. 5422 5423@smallexample 5424vector signed char vec_abs (vector signed char, vector signed char); 5425vector signed short vec_abs (vector signed short, vector signed short); 5426vector signed int vec_abs (vector signed int, vector signed int); 5427vector signed float vec_abs (vector signed float, vector signed float); 5428 5429vector signed char vec_abss (vector signed char, vector signed char); 5430vector signed short vec_abss (vector signed short, vector signed short); 5431 5432vector signed char vec_add (vector signed char, vector signed char); 5433vector unsigned char vec_add (vector signed char, vector unsigned char); 5434 5435vector unsigned char vec_add (vector unsigned char, vector signed char); 5436 5437vector unsigned char vec_add (vector unsigned char, 5438 vector unsigned char); 5439vector signed short vec_add (vector signed short, vector signed short); 5440vector unsigned short vec_add (vector signed short, 5441 vector unsigned short); 5442vector unsigned short vec_add (vector unsigned short, 5443 vector signed short); 5444vector unsigned short vec_add (vector unsigned short, 5445 vector unsigned short); 5446vector signed int vec_add (vector signed int, vector signed int); 5447vector unsigned int vec_add (vector signed int, vector unsigned int); 5448vector unsigned int vec_add (vector unsigned int, vector signed int); 5449vector unsigned int vec_add (vector unsigned int, vector unsigned int); 5450vector float vec_add (vector float, vector float); 5451 5452vector unsigned int vec_addc (vector unsigned int, vector unsigned int); 5453 5454vector unsigned char vec_adds (vector signed char, 5455 vector unsigned char); 5456vector unsigned char vec_adds (vector unsigned char, 5457 vector signed char); 5458vector unsigned char vec_adds (vector unsigned char, 5459 vector unsigned char); 5460vector signed char vec_adds (vector signed char, vector signed char); 5461vector unsigned short vec_adds (vector signed short, 5462 vector unsigned short); 5463vector unsigned short vec_adds (vector unsigned short, 5464 vector signed short); 5465vector unsigned short vec_adds (vector unsigned short, 5466 vector unsigned short); 5467vector signed short vec_adds (vector signed short, vector signed short); 5468 5469vector unsigned int vec_adds (vector signed int, vector unsigned int); 5470vector unsigned int vec_adds (vector unsigned int, vector signed int); 5471vector unsigned int vec_adds (vector unsigned int, vector unsigned int); 5472 5473vector signed int vec_adds (vector signed int, vector signed int); 5474 5475vector float vec_and (vector float, vector float); 5476vector float vec_and (vector float, vector signed int); 5477vector float vec_and (vector signed int, vector float); 5478vector signed int vec_and (vector signed int, vector signed int); 5479vector unsigned int vec_and (vector signed int, vector unsigned int); 5480vector unsigned int vec_and (vector unsigned int, vector signed int); 5481vector unsigned int vec_and (vector unsigned int, vector unsigned int); 5482vector signed short vec_and (vector signed short, vector signed short); 5483vector unsigned short vec_and (vector signed short, 5484 vector unsigned short); 5485vector unsigned short vec_and (vector unsigned short, 5486 vector signed short); 5487vector unsigned short vec_and (vector unsigned short, 5488 vector unsigned short); 5489vector signed char vec_and (vector signed char, vector signed char); 5490vector unsigned char vec_and (vector signed char, vector unsigned char); 5491 5492vector unsigned char vec_and (vector unsigned char, vector signed char); 5493 5494vector unsigned char vec_and (vector unsigned char, 5495 vector unsigned char); 5496 5497vector float vec_andc (vector float, vector float); 5498vector float vec_andc (vector float, vector signed int); 5499vector float vec_andc (vector signed int, vector float); 5500vector signed int vec_andc (vector signed int, vector signed int); 5501vector unsigned int vec_andc (vector signed int, vector unsigned int); 5502vector unsigned int vec_andc (vector unsigned int, vector signed int); 5503vector unsigned int vec_andc (vector unsigned int, vector unsigned int); 5504 5505vector signed short vec_andc (vector signed short, vector signed short); 5506 5507vector unsigned short vec_andc (vector signed short, 5508 vector unsigned short); 5509vector unsigned short vec_andc (vector unsigned short, 5510 vector signed short); 5511vector unsigned short vec_andc (vector unsigned short, 5512 vector unsigned short); 5513vector signed char vec_andc (vector signed char, vector signed char); 5514vector unsigned char vec_andc (vector signed char, 5515 vector unsigned char); 5516vector unsigned char vec_andc (vector unsigned char, 5517 vector signed char); 5518vector unsigned char vec_andc (vector unsigned char, 5519 vector unsigned char); 5520 5521vector unsigned char vec_avg (vector unsigned char, 5522 vector unsigned char); 5523vector signed char vec_avg (vector signed char, vector signed char); 5524vector unsigned short vec_avg (vector unsigned short, 5525 vector unsigned short); 5526vector signed short vec_avg (vector signed short, vector signed short); 5527vector unsigned int vec_avg (vector unsigned int, vector unsigned int); 5528vector signed int vec_avg (vector signed int, vector signed int); 5529 5530vector float vec_ceil (vector float); 5531 5532vector signed int vec_cmpb (vector float, vector float); 5533 5534vector signed char vec_cmpeq (vector signed char, vector signed char); 5535vector signed char vec_cmpeq (vector unsigned char, 5536 vector unsigned char); 5537vector signed short vec_cmpeq (vector signed short, 5538 vector signed short); 5539vector signed short vec_cmpeq (vector unsigned short, 5540 vector unsigned short); 5541vector signed int vec_cmpeq (vector signed int, vector signed int); 5542vector signed int vec_cmpeq (vector unsigned int, vector unsigned int); 5543vector signed int vec_cmpeq (vector float, vector float); 5544 5545vector signed int vec_cmpge (vector float, vector float); 5546 5547vector signed char vec_cmpgt (vector unsigned char, 5548 vector unsigned char); 5549vector signed char vec_cmpgt (vector signed char, vector signed char); 5550vector signed short vec_cmpgt (vector unsigned short, 5551 vector unsigned short); 5552vector signed short vec_cmpgt (vector signed short, 5553 vector signed short); 5554vector signed int vec_cmpgt (vector unsigned int, vector unsigned int); 5555vector signed int vec_cmpgt (vector signed int, vector signed int); 5556vector signed int vec_cmpgt (vector float, vector float); 5557 5558vector signed int vec_cmple (vector float, vector float); 5559 5560vector signed char vec_cmplt (vector unsigned char, 5561 vector unsigned char); 5562vector signed char vec_cmplt (vector signed char, vector signed char); 5563vector signed short vec_cmplt (vector unsigned short, 5564 vector unsigned short); 5565vector signed short vec_cmplt (vector signed short, 5566 vector signed short); 5567vector signed int vec_cmplt (vector unsigned int, vector unsigned int); 5568vector signed int vec_cmplt (vector signed int, vector signed int); 5569vector signed int vec_cmplt (vector float, vector float); 5570 5571vector float vec_ctf (vector unsigned int, const char); 5572vector float vec_ctf (vector signed int, const char); 5573 5574vector signed int vec_cts (vector float, const char); 5575 5576vector unsigned int vec_ctu (vector float, const char); 5577 5578void vec_dss (const char); 5579 5580void vec_dssall (void); 5581 5582void vec_dst (void *, int, const char); 5583 5584void vec_dstst (void *, int, const char); 5585 5586void vec_dststt (void *, int, const char); 5587 5588void vec_dstt (void *, int, const char); 5589 5590vector float vec_expte (vector float, vector float); 5591 5592vector float vec_floor (vector float, vector float); 5593 5594vector float vec_ld (int, vector float *); 5595vector float vec_ld (int, float *): 5596vector signed int vec_ld (int, int *); 5597vector signed int vec_ld (int, vector signed int *); 5598vector unsigned int vec_ld (int, vector unsigned int *); 5599vector unsigned int vec_ld (int, unsigned int *); 5600vector signed short vec_ld (int, short *, vector signed short *); 5601vector unsigned short vec_ld (int, unsigned short *, 5602 vector unsigned short *); 5603vector signed char vec_ld (int, signed char *); 5604vector signed char vec_ld (int, vector signed char *); 5605vector unsigned char vec_ld (int, unsigned char *); 5606vector unsigned char vec_ld (int, vector unsigned char *); 5607 5608vector signed char vec_lde (int, signed char *); 5609vector unsigned char vec_lde (int, unsigned char *); 5610vector signed short vec_lde (int, short *); 5611vector unsigned short vec_lde (int, unsigned short *); 5612vector float vec_lde (int, float *); 5613vector signed int vec_lde (int, int *); 5614vector unsigned int vec_lde (int, unsigned int *); 5615 5616void float vec_ldl (int, float *); 5617void float vec_ldl (int, vector float *); 5618void signed int vec_ldl (int, vector signed int *); 5619void signed int vec_ldl (int, int *); 5620void unsigned int vec_ldl (int, unsigned int *); 5621void unsigned int vec_ldl (int, vector unsigned int *); 5622void signed short vec_ldl (int, vector signed short *); 5623void signed short vec_ldl (int, short *); 5624void unsigned short vec_ldl (int, vector unsigned short *); 5625void unsigned short vec_ldl (int, unsigned short *); 5626void signed char vec_ldl (int, vector signed char *); 5627void signed char vec_ldl (int, signed char *); 5628void unsigned char vec_ldl (int, vector unsigned char *); 5629void unsigned char vec_ldl (int, unsigned char *); 5630 5631vector float vec_loge (vector float); 5632 5633vector unsigned char vec_lvsl (int, void *, int *); 5634 5635vector unsigned char vec_lvsr (int, void *, int *); 5636 5637vector float vec_madd (vector float, vector float, vector float); 5638 5639vector signed short vec_madds (vector signed short, vector signed short, 5640 vector signed short); 5641 5642vector unsigned char vec_max (vector signed char, vector unsigned char); 5643 5644vector unsigned char vec_max (vector unsigned char, vector signed char); 5645 5646vector unsigned char vec_max (vector unsigned char, 5647 vector unsigned char); 5648vector signed char vec_max (vector signed char, vector signed char); 5649vector unsigned short vec_max (vector signed short, 5650 vector unsigned short); 5651vector unsigned short vec_max (vector unsigned short, 5652 vector signed short); 5653vector unsigned short vec_max (vector unsigned short, 5654 vector unsigned short); 5655vector signed short vec_max (vector signed short, vector signed short); 5656vector unsigned int vec_max (vector signed int, vector unsigned int); 5657vector unsigned int vec_max (vector unsigned int, vector signed int); 5658vector unsigned int vec_max (vector unsigned int, vector unsigned int); 5659vector signed int vec_max (vector signed int, vector signed int); 5660vector float vec_max (vector float, vector float); 5661 5662vector signed char vec_mergeh (vector signed char, vector signed char); 5663vector unsigned char vec_mergeh (vector unsigned char, 5664 vector unsigned char); 5665vector signed short vec_mergeh (vector signed short, 5666 vector signed short); 5667vector unsigned short vec_mergeh (vector unsigned short, 5668 vector unsigned short); 5669vector float vec_mergeh (vector float, vector float); 5670vector signed int vec_mergeh (vector signed int, vector signed int); 5671vector unsigned int vec_mergeh (vector unsigned int, 5672 vector unsigned int); 5673 5674vector signed char vec_mergel (vector signed char, vector signed char); 5675vector unsigned char vec_mergel (vector unsigned char, 5676 vector unsigned char); 5677vector signed short vec_mergel (vector signed short, 5678 vector signed short); 5679vector unsigned short vec_mergel (vector unsigned short, 5680 vector unsigned short); 5681vector float vec_mergel (vector float, vector float); 5682vector signed int vec_mergel (vector signed int, vector signed int); 5683vector unsigned int vec_mergel (vector unsigned int, 5684 vector unsigned int); 5685 5686vector unsigned short vec_mfvscr (void); 5687 5688vector unsigned char vec_min (vector signed char, vector unsigned char); 5689 5690vector unsigned char vec_min (vector unsigned char, vector signed char); 5691 5692vector unsigned char vec_min (vector unsigned char, 5693 vector unsigned char); 5694vector signed char vec_min (vector signed char, vector signed char); 5695vector unsigned short vec_min (vector signed short, 5696 vector unsigned short); 5697vector unsigned short vec_min (vector unsigned short, 5698 vector signed short); 5699vector unsigned short vec_min (vector unsigned short, 5700 vector unsigned short); 5701vector signed short vec_min (vector signed short, vector signed short); 5702vector unsigned int vec_min (vector signed int, vector unsigned int); 5703vector unsigned int vec_min (vector unsigned int, vector signed int); 5704vector unsigned int vec_min (vector unsigned int, vector unsigned int); 5705vector signed int vec_min (vector signed int, vector signed int); 5706vector float vec_min (vector float, vector float); 5707 5708vector signed short vec_mladd (vector signed short, vector signed short, 5709 vector signed short); 5710vector signed short vec_mladd (vector signed short, 5711 vector unsigned short, 5712 vector unsigned short); 5713vector signed short vec_mladd (vector unsigned short, 5714 vector signed short, 5715 vector signed short); 5716vector unsigned short vec_mladd (vector unsigned short, 5717 vector unsigned short, 5718 vector unsigned short); 5719 5720vector signed short vec_mradds (vector signed short, 5721 vector signed short, 5722 vector signed short); 5723 5724vector unsigned int vec_msum (vector unsigned char, 5725 vector unsigned char, 5726 vector unsigned int); 5727vector signed int vec_msum (vector signed char, vector unsigned char, 5728 vector signed int); 5729vector unsigned int vec_msum (vector unsigned short, 5730 vector unsigned short, 5731 vector unsigned int); 5732vector signed int vec_msum (vector signed short, vector signed short, 5733 vector signed int); 5734 5735vector unsigned int vec_msums (vector unsigned short, 5736 vector unsigned short, 5737 vector unsigned int); 5738vector signed int vec_msums (vector signed short, vector signed short, 5739 vector signed int); 5740 5741void vec_mtvscr (vector signed int); 5742void vec_mtvscr (vector unsigned int); 5743void vec_mtvscr (vector signed short); 5744void vec_mtvscr (vector unsigned short); 5745void vec_mtvscr (vector signed char); 5746void vec_mtvscr (vector unsigned char); 5747 5748vector unsigned short vec_mule (vector unsigned char, 5749 vector unsigned char); 5750vector signed short vec_mule (vector signed char, vector signed char); 5751vector unsigned int vec_mule (vector unsigned short, 5752 vector unsigned short); 5753vector signed int vec_mule (vector signed short, vector signed short); 5754 5755vector unsigned short vec_mulo (vector unsigned char, 5756 vector unsigned char); 5757vector signed short vec_mulo (vector signed char, vector signed char); 5758vector unsigned int vec_mulo (vector unsigned short, 5759 vector unsigned short); 5760vector signed int vec_mulo (vector signed short, vector signed short); 5761 5762vector float vec_nmsub (vector float, vector float, vector float); 5763 5764vector float vec_nor (vector float, vector float); 5765vector signed int vec_nor (vector signed int, vector signed int); 5766vector unsigned int vec_nor (vector unsigned int, vector unsigned int); 5767vector signed short vec_nor (vector signed short, vector signed short); 5768vector unsigned short vec_nor (vector unsigned short, 5769 vector unsigned short); 5770vector signed char vec_nor (vector signed char, vector signed char); 5771vector unsigned char vec_nor (vector unsigned char, 5772 vector unsigned char); 5773 5774vector float vec_or (vector float, vector float); 5775vector float vec_or (vector float, vector signed int); 5776vector float vec_or (vector signed int, vector float); 5777vector signed int vec_or (vector signed int, vector signed int); 5778vector unsigned int vec_or (vector signed int, vector unsigned int); 5779vector unsigned int vec_or (vector unsigned int, vector signed int); 5780vector unsigned int vec_or (vector unsigned int, vector unsigned int); 5781vector signed short vec_or (vector signed short, vector signed short); 5782vector unsigned short vec_or (vector signed short, 5783 vector unsigned short); 5784vector unsigned short vec_or (vector unsigned short, 5785 vector signed short); 5786vector unsigned short vec_or (vector unsigned short, 5787 vector unsigned short); 5788vector signed char vec_or (vector signed char, vector signed char); 5789vector unsigned char vec_or (vector signed char, vector unsigned char); 5790vector unsigned char vec_or (vector unsigned char, vector signed char); 5791vector unsigned char vec_or (vector unsigned char, 5792 vector unsigned char); 5793 5794vector signed char vec_pack (vector signed short, vector signed short); 5795vector unsigned char vec_pack (vector unsigned short, 5796 vector unsigned short); 5797vector signed short vec_pack (vector signed int, vector signed int); 5798vector unsigned short vec_pack (vector unsigned int, 5799 vector unsigned int); 5800 5801vector signed short vec_packpx (vector unsigned int, 5802 vector unsigned int); 5803 5804vector unsigned char vec_packs (vector unsigned short, 5805 vector unsigned short); 5806vector signed char vec_packs (vector signed short, vector signed short); 5807 5808vector unsigned short vec_packs (vector unsigned int, 5809 vector unsigned int); 5810vector signed short vec_packs (vector signed int, vector signed int); 5811 5812vector unsigned char vec_packsu (vector unsigned short, 5813 vector unsigned short); 5814vector unsigned char vec_packsu (vector signed short, 5815 vector signed short); 5816vector unsigned short vec_packsu (vector unsigned int, 5817 vector unsigned int); 5818vector unsigned short vec_packsu (vector signed int, vector signed int); 5819 5820vector float vec_perm (vector float, vector float, 5821 vector unsigned char); 5822vector signed int vec_perm (vector signed int, vector signed int, 5823 vector unsigned char); 5824vector unsigned int vec_perm (vector unsigned int, vector unsigned int, 5825 vector unsigned char); 5826vector signed short vec_perm (vector signed short, vector signed short, 5827 vector unsigned char); 5828vector unsigned short vec_perm (vector unsigned short, 5829 vector unsigned short, 5830 vector unsigned char); 5831vector signed char vec_perm (vector signed char, vector signed char, 5832 vector unsigned char); 5833vector unsigned char vec_perm (vector unsigned char, 5834 vector unsigned char, 5835 vector unsigned char); 5836 5837vector float vec_re (vector float); 5838 5839vector signed char vec_rl (vector signed char, vector unsigned char); 5840vector unsigned char vec_rl (vector unsigned char, 5841 vector unsigned char); 5842vector signed short vec_rl (vector signed short, vector unsigned short); 5843 5844vector unsigned short vec_rl (vector unsigned short, 5845 vector unsigned short); 5846vector signed int vec_rl (vector signed int, vector unsigned int); 5847vector unsigned int vec_rl (vector unsigned int, vector unsigned int); 5848 5849vector float vec_round (vector float); 5850 5851vector float vec_rsqrte (vector float); 5852 5853vector float vec_sel (vector float, vector float, vector signed int); 5854vector float vec_sel (vector float, vector float, vector unsigned int); 5855vector signed int vec_sel (vector signed int, vector signed int, 5856 vector signed int); 5857vector signed int vec_sel (vector signed int, vector signed int, 5858 vector unsigned int); 5859vector unsigned int vec_sel (vector unsigned int, vector unsigned int, 5860 vector signed int); 5861vector unsigned int vec_sel (vector unsigned int, vector unsigned int, 5862 vector unsigned int); 5863vector signed short vec_sel (vector signed short, vector signed short, 5864 vector signed short); 5865vector signed short vec_sel (vector signed short, vector signed short, 5866 vector unsigned short); 5867vector unsigned short vec_sel (vector unsigned short, 5868 vector unsigned short, 5869 vector signed short); 5870vector unsigned short vec_sel (vector unsigned short, 5871 vector unsigned short, 5872 vector unsigned short); 5873vector signed char vec_sel (vector signed char, vector signed char, 5874 vector signed char); 5875vector signed char vec_sel (vector signed char, vector signed char, 5876 vector unsigned char); 5877vector unsigned char vec_sel (vector unsigned char, 5878 vector unsigned char, 5879 vector signed char); 5880vector unsigned char vec_sel (vector unsigned char, 5881 vector unsigned char, 5882 vector unsigned char); 5883 5884vector signed char vec_sl (vector signed char, vector unsigned char); 5885vector unsigned char vec_sl (vector unsigned char, 5886 vector unsigned char); 5887vector signed short vec_sl (vector signed short, vector unsigned short); 5888 5889vector unsigned short vec_sl (vector unsigned short, 5890 vector unsigned short); 5891vector signed int vec_sl (vector signed int, vector unsigned int); 5892vector unsigned int vec_sl (vector unsigned int, vector unsigned int); 5893 5894vector float vec_sld (vector float, vector float, const char); 5895vector signed int vec_sld (vector signed int, vector signed int, 5896 const char); 5897vector unsigned int vec_sld (vector unsigned int, vector unsigned int, 5898 const char); 5899vector signed short vec_sld (vector signed short, vector signed short, 5900 const char); 5901vector unsigned short vec_sld (vector unsigned short, 5902 vector unsigned short, const char); 5903vector signed char vec_sld (vector signed char, vector signed char, 5904 const char); 5905vector unsigned char vec_sld (vector unsigned char, 5906 vector unsigned char, 5907 const char); 5908 5909vector signed int vec_sll (vector signed int, vector unsigned int); 5910vector signed int vec_sll (vector signed int, vector unsigned short); 5911vector signed int vec_sll (vector signed int, vector unsigned char); 5912vector unsigned int vec_sll (vector unsigned int, vector unsigned int); 5913vector unsigned int vec_sll (vector unsigned int, 5914 vector unsigned short); 5915vector unsigned int vec_sll (vector unsigned int, vector unsigned char); 5916 5917vector signed short vec_sll (vector signed short, vector unsigned int); 5918vector signed short vec_sll (vector signed short, 5919 vector unsigned short); 5920vector signed short vec_sll (vector signed short, vector unsigned char); 5921 5922vector unsigned short vec_sll (vector unsigned short, 5923 vector unsigned int); 5924vector unsigned short vec_sll (vector unsigned short, 5925 vector unsigned short); 5926vector unsigned short vec_sll (vector unsigned short, 5927 vector unsigned char); 5928vector signed char vec_sll (vector signed char, vector unsigned int); 5929vector signed char vec_sll (vector signed char, vector unsigned short); 5930vector signed char vec_sll (vector signed char, vector unsigned char); 5931vector unsigned char vec_sll (vector unsigned char, 5932 vector unsigned int); 5933vector unsigned char vec_sll (vector unsigned char, 5934 vector unsigned short); 5935vector unsigned char vec_sll (vector unsigned char, 5936 vector unsigned char); 5937 5938vector float vec_slo (vector float, vector signed char); 5939vector float vec_slo (vector float, vector unsigned char); 5940vector signed int vec_slo (vector signed int, vector signed char); 5941vector signed int vec_slo (vector signed int, vector unsigned char); 5942vector unsigned int vec_slo (vector unsigned int, vector signed char); 5943vector unsigned int vec_slo (vector unsigned int, vector unsigned char); 5944 5945vector signed short vec_slo (vector signed short, vector signed char); 5946vector signed short vec_slo (vector signed short, vector unsigned char); 5947 5948vector unsigned short vec_slo (vector unsigned short, 5949 vector signed char); 5950vector unsigned short vec_slo (vector unsigned short, 5951 vector unsigned char); 5952vector signed char vec_slo (vector signed char, vector signed char); 5953vector signed char vec_slo (vector signed char, vector unsigned char); 5954vector unsigned char vec_slo (vector unsigned char, vector signed char); 5955 5956vector unsigned char vec_slo (vector unsigned char, 5957 vector unsigned char); 5958 5959vector signed char vec_splat (vector signed char, const char); 5960vector unsigned char vec_splat (vector unsigned char, const char); 5961vector signed short vec_splat (vector signed short, const char); 5962vector unsigned short vec_splat (vector unsigned short, const char); 5963vector float vec_splat (vector float, const char); 5964vector signed int vec_splat (vector signed int, const char); 5965vector unsigned int vec_splat (vector unsigned int, const char); 5966 5967vector signed char vec_splat_s8 (const char); 5968 5969vector signed short vec_splat_s16 (const char); 5970 5971vector signed int vec_splat_s32 (const char); 5972 5973vector unsigned char vec_splat_u8 (const char); 5974 5975vector unsigned short vec_splat_u16 (const char); 5976 5977vector unsigned int vec_splat_u32 (const char); 5978 5979vector signed char vec_sr (vector signed char, vector unsigned char); 5980vector unsigned char vec_sr (vector unsigned char, 5981 vector unsigned char); 5982vector signed short vec_sr (vector signed short, vector unsigned short); 5983 5984vector unsigned short vec_sr (vector unsigned short, 5985 vector unsigned short); 5986vector signed int vec_sr (vector signed int, vector unsigned int); 5987vector unsigned int vec_sr (vector unsigned int, vector unsigned int); 5988 5989vector signed char vec_sra (vector signed char, vector unsigned char); 5990vector unsigned char vec_sra (vector unsigned char, 5991 vector unsigned char); 5992vector signed short vec_sra (vector signed short, 5993 vector unsigned short); 5994vector unsigned short vec_sra (vector unsigned short, 5995 vector unsigned short); 5996vector signed int vec_sra (vector signed int, vector unsigned int); 5997vector unsigned int vec_sra (vector unsigned int, vector unsigned int); 5998 5999vector signed int vec_srl (vector signed int, vector unsigned int); 6000vector signed int vec_srl (vector signed int, vector unsigned short); 6001vector signed int vec_srl (vector signed int, vector unsigned char); 6002vector unsigned int vec_srl (vector unsigned int, vector unsigned int); 6003vector unsigned int vec_srl (vector unsigned int, 6004 vector unsigned short); 6005vector unsigned int vec_srl (vector unsigned int, vector unsigned char); 6006 6007vector signed short vec_srl (vector signed short, vector unsigned int); 6008vector signed short vec_srl (vector signed short, 6009 vector unsigned short); 6010vector signed short vec_srl (vector signed short, vector unsigned char); 6011 6012vector unsigned short vec_srl (vector unsigned short, 6013 vector unsigned int); 6014vector unsigned short vec_srl (vector unsigned short, 6015 vector unsigned short); 6016vector unsigned short vec_srl (vector unsigned short, 6017 vector unsigned char); 6018vector signed char vec_srl (vector signed char, vector unsigned int); 6019vector signed char vec_srl (vector signed char, vector unsigned short); 6020vector signed char vec_srl (vector signed char, vector unsigned char); 6021vector unsigned char vec_srl (vector unsigned char, 6022 vector unsigned int); 6023vector unsigned char vec_srl (vector unsigned char, 6024 vector unsigned short); 6025vector unsigned char vec_srl (vector unsigned char, 6026 vector unsigned char); 6027 6028vector float vec_sro (vector float, vector signed char); 6029vector float vec_sro (vector float, vector unsigned char); 6030vector signed int vec_sro (vector signed int, vector signed char); 6031vector signed int vec_sro (vector signed int, vector unsigned char); 6032vector unsigned int vec_sro (vector unsigned int, vector signed char); 6033vector unsigned int vec_sro (vector unsigned int, vector unsigned char); 6034 6035vector signed short vec_sro (vector signed short, vector signed char); 6036vector signed short vec_sro (vector signed short, vector unsigned char); 6037 6038vector unsigned short vec_sro (vector unsigned short, 6039 vector signed char); 6040vector unsigned short vec_sro (vector unsigned short, 6041 vector unsigned char); 6042vector signed char vec_sro (vector signed char, vector signed char); 6043vector signed char vec_sro (vector signed char, vector unsigned char); 6044vector unsigned char vec_sro (vector unsigned char, vector signed char); 6045 6046vector unsigned char vec_sro (vector unsigned char, 6047 vector unsigned char); 6048 6049void vec_st (vector float, int, float *); 6050void vec_st (vector float, int, vector float *); 6051void vec_st (vector signed int, int, int *); 6052void vec_st (vector signed int, int, unsigned int *); 6053void vec_st (vector unsigned int, int, unsigned int *); 6054void vec_st (vector unsigned int, int, vector unsigned int *); 6055void vec_st (vector signed short, int, short *); 6056void vec_st (vector signed short, int, vector unsigned short *); 6057void vec_st (vector signed short, int, vector signed short *); 6058void vec_st (vector unsigned short, int, unsigned short *); 6059void vec_st (vector unsigned short, int, vector unsigned short *); 6060void vec_st (vector signed char, int, signed char *); 6061void vec_st (vector signed char, int, unsigned char *); 6062void vec_st (vector signed char, int, vector signed char *); 6063void vec_st (vector unsigned char, int, unsigned char *); 6064void vec_st (vector unsigned char, int, vector unsigned char *); 6065 6066void vec_ste (vector signed char, int, unsigned char *); 6067void vec_ste (vector signed char, int, signed char *); 6068void vec_ste (vector unsigned char, int, unsigned char *); 6069void vec_ste (vector signed short, int, short *); 6070void vec_ste (vector signed short, int, unsigned short *); 6071void vec_ste (vector unsigned short, int, void *); 6072void vec_ste (vector signed int, int, unsigned int *); 6073void vec_ste (vector signed int, int, int *); 6074void vec_ste (vector unsigned int, int, unsigned int *); 6075void vec_ste (vector float, int, float *); 6076 6077void vec_stl (vector float, int, vector float *); 6078void vec_stl (vector float, int, float *); 6079void vec_stl (vector signed int, int, vector signed int *); 6080void vec_stl (vector signed int, int, int *); 6081void vec_stl (vector signed int, int, unsigned int *); 6082void vec_stl (vector unsigned int, int, vector unsigned int *); 6083void vec_stl (vector unsigned int, int, unsigned int *); 6084void vec_stl (vector signed short, int, short *); 6085void vec_stl (vector signed short, int, unsigned short *); 6086void vec_stl (vector signed short, int, vector signed short *); 6087void vec_stl (vector unsigned short, int, unsigned short *); 6088void vec_stl (vector unsigned short, int, vector signed short *); 6089void vec_stl (vector signed char, int, signed char *); 6090void vec_stl (vector signed char, int, unsigned char *); 6091void vec_stl (vector signed char, int, vector signed char *); 6092void vec_stl (vector unsigned char, int, unsigned char *); 6093void vec_stl (vector unsigned char, int, vector unsigned char *); 6094 6095vector signed char vec_sub (vector signed char, vector signed char); 6096vector unsigned char vec_sub (vector signed char, vector unsigned char); 6097 6098vector unsigned char vec_sub (vector unsigned char, vector signed char); 6099 6100vector unsigned char vec_sub (vector unsigned char, 6101 vector unsigned char); 6102vector signed short vec_sub (vector signed short, vector signed short); 6103vector unsigned short vec_sub (vector signed short, 6104 vector unsigned short); 6105vector unsigned short vec_sub (vector unsigned short, 6106 vector signed short); 6107vector unsigned short vec_sub (vector unsigned short, 6108 vector unsigned short); 6109vector signed int vec_sub (vector signed int, vector signed int); 6110vector unsigned int vec_sub (vector signed int, vector unsigned int); 6111vector unsigned int vec_sub (vector unsigned int, vector signed int); 6112vector unsigned int vec_sub (vector unsigned int, vector unsigned int); 6113vector float vec_sub (vector float, vector float); 6114 6115vector unsigned int vec_subc (vector unsigned int, vector unsigned int); 6116 6117vector unsigned char vec_subs (vector signed char, 6118 vector unsigned char); 6119vector unsigned char vec_subs (vector unsigned char, 6120 vector signed char); 6121vector unsigned char vec_subs (vector unsigned char, 6122 vector unsigned char); 6123vector signed char vec_subs (vector signed char, vector signed char); 6124vector unsigned short vec_subs (vector signed short, 6125 vector unsigned short); 6126vector unsigned short vec_subs (vector unsigned short, 6127 vector signed short); 6128vector unsigned short vec_subs (vector unsigned short, 6129 vector unsigned short); 6130vector signed short vec_subs (vector signed short, vector signed short); 6131 6132vector unsigned int vec_subs (vector signed int, vector unsigned int); 6133vector unsigned int vec_subs (vector unsigned int, vector signed int); 6134vector unsigned int vec_subs (vector unsigned int, vector unsigned int); 6135 6136vector signed int vec_subs (vector signed int, vector signed int); 6137 6138vector unsigned int vec_sum4s (vector unsigned char, 6139 vector unsigned int); 6140vector signed int vec_sum4s (vector signed char, vector signed int); 6141vector signed int vec_sum4s (vector signed short, vector signed int); 6142 6143vector signed int vec_sum2s (vector signed int, vector signed int); 6144 6145vector signed int vec_sums (vector signed int, vector signed int); 6146 6147vector float vec_trunc (vector float); 6148 6149vector signed short vec_unpackh (vector signed char); 6150vector unsigned int vec_unpackh (vector signed short); 6151vector signed int vec_unpackh (vector signed short); 6152 6153vector signed short vec_unpackl (vector signed char); 6154vector unsigned int vec_unpackl (vector signed short); 6155vector signed int vec_unpackl (vector signed short); 6156 6157vector float vec_xor (vector float, vector float); 6158vector float vec_xor (vector float, vector signed int); 6159vector float vec_xor (vector signed int, vector float); 6160vector signed int vec_xor (vector signed int, vector signed int); 6161vector unsigned int vec_xor (vector signed int, vector unsigned int); 6162vector unsigned int vec_xor (vector unsigned int, vector signed int); 6163vector unsigned int vec_xor (vector unsigned int, vector unsigned int); 6164vector signed short vec_xor (vector signed short, vector signed short); 6165vector unsigned short vec_xor (vector signed short, 6166 vector unsigned short); 6167vector unsigned short vec_xor (vector unsigned short, 6168 vector signed short); 6169vector unsigned short vec_xor (vector unsigned short, 6170 vector unsigned short); 6171vector signed char vec_xor (vector signed char, vector signed char); 6172vector unsigned char vec_xor (vector signed char, vector unsigned char); 6173 6174vector unsigned char vec_xor (vector unsigned char, vector signed char); 6175 6176vector unsigned char vec_xor (vector unsigned char, 6177 vector unsigned char); 6178 6179vector signed int vec_all_eq (vector signed char, vector unsigned char); 6180 6181vector signed int vec_all_eq (vector signed char, vector signed char); 6182vector signed int vec_all_eq (vector unsigned char, vector signed char); 6183 6184vector signed int vec_all_eq (vector unsigned char, 6185 vector unsigned char); 6186vector signed int vec_all_eq (vector signed short, 6187 vector unsigned short); 6188vector signed int vec_all_eq (vector signed short, vector signed short); 6189 6190vector signed int vec_all_eq (vector unsigned short, 6191 vector signed short); 6192vector signed int vec_all_eq (vector unsigned short, 6193 vector unsigned short); 6194vector signed int vec_all_eq (vector signed int, vector unsigned int); 6195vector signed int vec_all_eq (vector signed int, vector signed int); 6196vector signed int vec_all_eq (vector unsigned int, vector signed int); 6197vector signed int vec_all_eq (vector unsigned int, vector unsigned int); 6198 6199vector signed int vec_all_eq (vector float, vector float); 6200 6201vector signed int vec_all_ge (vector signed char, vector unsigned char); 6202 6203vector signed int vec_all_ge (vector unsigned char, vector signed char); 6204 6205vector signed int vec_all_ge (vector unsigned char, 6206 vector unsigned char); 6207vector signed int vec_all_ge (vector signed char, vector signed char); 6208vector signed int vec_all_ge (vector signed short, 6209 vector unsigned short); 6210vector signed int vec_all_ge (vector unsigned short, 6211 vector signed short); 6212vector signed int vec_all_ge (vector unsigned short, 6213 vector unsigned short); 6214vector signed int vec_all_ge (vector signed short, vector signed short); 6215 6216vector signed int vec_all_ge (vector signed int, vector unsigned int); 6217vector signed int vec_all_ge (vector unsigned int, vector signed int); 6218vector signed int vec_all_ge (vector unsigned int, vector unsigned int); 6219 6220vector signed int vec_all_ge (vector signed int, vector signed int); 6221vector signed int vec_all_ge (vector float, vector float); 6222 6223vector signed int vec_all_gt (vector signed char, vector unsigned char); 6224 6225vector signed int vec_all_gt (vector unsigned char, vector signed char); 6226 6227vector signed int vec_all_gt (vector unsigned char, 6228 vector unsigned char); 6229vector signed int vec_all_gt (vector signed char, vector signed char); 6230vector signed int vec_all_gt (vector signed short, 6231 vector unsigned short); 6232vector signed int vec_all_gt (vector unsigned short, 6233 vector signed short); 6234vector signed int vec_all_gt (vector unsigned short, 6235 vector unsigned short); 6236vector signed int vec_all_gt (vector signed short, vector signed short); 6237 6238vector signed int vec_all_gt (vector signed int, vector unsigned int); 6239vector signed int vec_all_gt (vector unsigned int, vector signed int); 6240vector signed int vec_all_gt (vector unsigned int, vector unsigned int); 6241 6242vector signed int vec_all_gt (vector signed int, vector signed int); 6243vector signed int vec_all_gt (vector float, vector float); 6244 6245vector signed int vec_all_in (vector float, vector float); 6246 6247vector signed int vec_all_le (vector signed char, vector unsigned char); 6248 6249vector signed int vec_all_le (vector unsigned char, vector signed char); 6250 6251vector signed int vec_all_le (vector unsigned char, 6252 vector unsigned char); 6253vector signed int vec_all_le (vector signed char, vector signed char); 6254vector signed int vec_all_le (vector signed short, 6255 vector unsigned short); 6256vector signed int vec_all_le (vector unsigned short, 6257 vector signed short); 6258vector signed int vec_all_le (vector unsigned short, 6259 vector unsigned short); 6260vector signed int vec_all_le (vector signed short, vector signed short); 6261 6262vector signed int vec_all_le (vector signed int, vector unsigned int); 6263vector signed int vec_all_le (vector unsigned int, vector signed int); 6264vector signed int vec_all_le (vector unsigned int, vector unsigned int); 6265 6266vector signed int vec_all_le (vector signed int, vector signed int); 6267vector signed int vec_all_le (vector float, vector float); 6268 6269vector signed int vec_all_lt (vector signed char, vector unsigned char); 6270 6271vector signed int vec_all_lt (vector unsigned char, vector signed char); 6272 6273vector signed int vec_all_lt (vector unsigned char, 6274 vector unsigned char); 6275vector signed int vec_all_lt (vector signed char, vector signed char); 6276vector signed int vec_all_lt (vector signed short, 6277 vector unsigned short); 6278vector signed int vec_all_lt (vector unsigned short, 6279 vector signed short); 6280vector signed int vec_all_lt (vector unsigned short, 6281 vector unsigned short); 6282vector signed int vec_all_lt (vector signed short, vector signed short); 6283 6284vector signed int vec_all_lt (vector signed int, vector unsigned int); 6285vector signed int vec_all_lt (vector unsigned int, vector signed int); 6286vector signed int vec_all_lt (vector unsigned int, vector unsigned int); 6287 6288vector signed int vec_all_lt (vector signed int, vector signed int); 6289vector signed int vec_all_lt (vector float, vector float); 6290 6291vector signed int vec_all_nan (vector float); 6292 6293vector signed int vec_all_ne (vector signed char, vector unsigned char); 6294 6295vector signed int vec_all_ne (vector signed char, vector signed char); 6296vector signed int vec_all_ne (vector unsigned char, vector signed char); 6297 6298vector signed int vec_all_ne (vector unsigned char, 6299 vector unsigned char); 6300vector signed int vec_all_ne (vector signed short, 6301 vector unsigned short); 6302vector signed int vec_all_ne (vector signed short, vector signed short); 6303 6304vector signed int vec_all_ne (vector unsigned short, 6305 vector signed short); 6306vector signed int vec_all_ne (vector unsigned short, 6307 vector unsigned short); 6308vector signed int vec_all_ne (vector signed int, vector unsigned int); 6309vector signed int vec_all_ne (vector signed int, vector signed int); 6310vector signed int vec_all_ne (vector unsigned int, vector signed int); 6311vector signed int vec_all_ne (vector unsigned int, vector unsigned int); 6312 6313vector signed int vec_all_ne (vector float, vector float); 6314 6315vector signed int vec_all_nge (vector float, vector float); 6316 6317vector signed int vec_all_ngt (vector float, vector float); 6318 6319vector signed int vec_all_nle (vector float, vector float); 6320 6321vector signed int vec_all_nlt (vector float, vector float); 6322 6323vector signed int vec_all_numeric (vector float); 6324 6325vector signed int vec_any_eq (vector signed char, vector unsigned char); 6326 6327vector signed int vec_any_eq (vector signed char, vector signed char); 6328vector signed int vec_any_eq (vector unsigned char, vector signed char); 6329 6330vector signed int vec_any_eq (vector unsigned char, 6331 vector unsigned char); 6332vector signed int vec_any_eq (vector signed short, 6333 vector unsigned short); 6334vector signed int vec_any_eq (vector signed short, vector signed short); 6335 6336vector signed int vec_any_eq (vector unsigned short, 6337 vector signed short); 6338vector signed int vec_any_eq (vector unsigned short, 6339 vector unsigned short); 6340vector signed int vec_any_eq (vector signed int, vector unsigned int); 6341vector signed int vec_any_eq (vector signed int, vector signed int); 6342vector signed int vec_any_eq (vector unsigned int, vector signed int); 6343vector signed int vec_any_eq (vector unsigned int, vector unsigned int); 6344 6345vector signed int vec_any_eq (vector float, vector float); 6346 6347vector signed int vec_any_ge (vector signed char, vector unsigned char); 6348 6349vector signed int vec_any_ge (vector unsigned char, vector signed char); 6350 6351vector signed int vec_any_ge (vector unsigned char, 6352 vector unsigned char); 6353vector signed int vec_any_ge (vector signed char, vector signed char); 6354vector signed int vec_any_ge (vector signed short, 6355 vector unsigned short); 6356vector signed int vec_any_ge (vector unsigned short, 6357 vector signed short); 6358vector signed int vec_any_ge (vector unsigned short, 6359 vector unsigned short); 6360vector signed int vec_any_ge (vector signed short, vector signed short); 6361 6362vector signed int vec_any_ge (vector signed int, vector unsigned int); 6363vector signed int vec_any_ge (vector unsigned int, vector signed int); 6364vector signed int vec_any_ge (vector unsigned int, vector unsigned int); 6365 6366vector signed int vec_any_ge (vector signed int, vector signed int); 6367vector signed int vec_any_ge (vector float, vector float); 6368 6369vector signed int vec_any_gt (vector signed char, vector unsigned char); 6370 6371vector signed int vec_any_gt (vector unsigned char, vector signed char); 6372 6373vector signed int vec_any_gt (vector unsigned char, 6374 vector unsigned char); 6375vector signed int vec_any_gt (vector signed char, vector signed char); 6376vector signed int vec_any_gt (vector signed short, 6377 vector unsigned short); 6378vector signed int vec_any_gt (vector unsigned short, 6379 vector signed short); 6380vector signed int vec_any_gt (vector unsigned short, 6381 vector unsigned short); 6382vector signed int vec_any_gt (vector signed short, vector signed short); 6383 6384vector signed int vec_any_gt (vector signed int, vector unsigned int); 6385vector signed int vec_any_gt (vector unsigned int, vector signed int); 6386vector signed int vec_any_gt (vector unsigned int, vector unsigned int); 6387 6388vector signed int vec_any_gt (vector signed int, vector signed int); 6389vector signed int vec_any_gt (vector float, vector float); 6390 6391vector signed int vec_any_le (vector signed char, vector unsigned char); 6392 6393vector signed int vec_any_le (vector unsigned char, vector signed char); 6394 6395vector signed int vec_any_le (vector unsigned char, 6396 vector unsigned char); 6397vector signed int vec_any_le (vector signed char, vector signed char); 6398vector signed int vec_any_le (vector signed short, 6399 vector unsigned short); 6400vector signed int vec_any_le (vector unsigned short, 6401 vector signed short); 6402vector signed int vec_any_le (vector unsigned short, 6403 vector unsigned short); 6404vector signed int vec_any_le (vector signed short, vector signed short); 6405 6406vector signed int vec_any_le (vector signed int, vector unsigned int); 6407vector signed int vec_any_le (vector unsigned int, vector signed int); 6408vector signed int vec_any_le (vector unsigned int, vector unsigned int); 6409 6410vector signed int vec_any_le (vector signed int, vector signed int); 6411vector signed int vec_any_le (vector float, vector float); 6412 6413vector signed int vec_any_lt (vector signed char, vector unsigned char); 6414 6415vector signed int vec_any_lt (vector unsigned char, vector signed char); 6416 6417vector signed int vec_any_lt (vector unsigned char, 6418 vector unsigned char); 6419vector signed int vec_any_lt (vector signed char, vector signed char); 6420vector signed int vec_any_lt (vector signed short, 6421 vector unsigned short); 6422vector signed int vec_any_lt (vector unsigned short, 6423 vector signed short); 6424vector signed int vec_any_lt (vector unsigned short, 6425 vector unsigned short); 6426vector signed int vec_any_lt (vector signed short, vector signed short); 6427 6428vector signed int vec_any_lt (vector signed int, vector unsigned int); 6429vector signed int vec_any_lt (vector unsigned int, vector signed int); 6430vector signed int vec_any_lt (vector unsigned int, vector unsigned int); 6431 6432vector signed int vec_any_lt (vector signed int, vector signed int); 6433vector signed int vec_any_lt (vector float, vector float); 6434 6435vector signed int vec_any_nan (vector float); 6436 6437vector signed int vec_any_ne (vector signed char, vector unsigned char); 6438 6439vector signed int vec_any_ne (vector signed char, vector signed char); 6440vector signed int vec_any_ne (vector unsigned char, vector signed char); 6441 6442vector signed int vec_any_ne (vector unsigned char, 6443 vector unsigned char); 6444vector signed int vec_any_ne (vector signed short, 6445 vector unsigned short); 6446vector signed int vec_any_ne (vector signed short, vector signed short); 6447 6448vector signed int vec_any_ne (vector unsigned short, 6449 vector signed short); 6450vector signed int vec_any_ne (vector unsigned short, 6451 vector unsigned short); 6452vector signed int vec_any_ne (vector signed int, vector unsigned int); 6453vector signed int vec_any_ne (vector signed int, vector signed int); 6454vector signed int vec_any_ne (vector unsigned int, vector signed int); 6455vector signed int vec_any_ne (vector unsigned int, vector unsigned int); 6456 6457vector signed int vec_any_ne (vector float, vector float); 6458 6459vector signed int vec_any_nge (vector float, vector float); 6460 6461vector signed int vec_any_ngt (vector float, vector float); 6462 6463vector signed int vec_any_nle (vector float, vector float); 6464 6465vector signed int vec_any_nlt (vector float, vector float); 6466 6467vector signed int vec_any_numeric (vector float); 6468 6469vector signed int vec_any_out (vector float, vector float); 6470@end smallexample 6471 6472@node Pragmas 6473@section Pragmas Accepted by GCC 6474@cindex pragmas 6475@cindex #pragma 6476 6477GCC supports several types of pragmas, primarily in order to compile 6478code originally written for other compilers. Note that in general 6479we do not recommend the use of pragmas; @xref{Function Attributes}, 6480for further explanation. 6481 6482@menu 6483* ARM Pragmas:: 6484* RS/6000 and PowerPC Pragmas:: 6485* Darwin Pragmas:: 6486* Solaris Pragmas:: 6487* Tru64 Pragmas:: 6488@end menu 6489 6490@node ARM Pragmas 6491@subsection ARM Pragmas 6492 6493The ARM target defines pragmas for controlling the default addition of 6494@code{long_call} and @code{short_call} attributes to functions. 6495@xref{Function Attributes}, for information about the effects of these 6496attributes. 6497 6498@table @code 6499@item long_calls 6500@cindex pragma, long_calls 6501Set all subsequent functions to have the @code{long_call} attribute. 6502 6503@item no_long_calls 6504@cindex pragma, no_long_calls 6505Set all subsequent functions to have the @code{short_call} attribute. 6506 6507@item long_calls_off 6508@cindex pragma, long_calls_off 6509Do not affect the @code{long_call} or @code{short_call} attributes of 6510subsequent functions. 6511@end table 6512 6513@node RS/6000 and PowerPC Pragmas 6514@subsection RS/6000 and PowerPC Pragmas 6515 6516The RS/6000 and PowerPC targets define one pragma for controlling 6517whether or not the @code{longcall} attribute is added to function 6518declarations by default. This pragma overrides the @option{-mlongcall} 6519option, but not the @code{longcall} and @code{shortcall} attributes. 6520@xref{RS/6000 and PowerPC Options}, for more information about when long 6521calls are and are not necessary. 6522 6523@table @code 6524@item longcall (1) 6525@cindex pragma, longcall 6526Apply the @code{longcall} attribute to all subsequent function 6527declarations. 6528 6529@item longcall (0) 6530Do not apply the @code{longcall} attribute to subsequent function 6531declarations. 6532@end table 6533 6534@c Describe c4x pragmas here. 6535@c Describe h8300 pragmas here. 6536@c Describe i370 pragmas here. 6537@c Describe i960 pragmas here. 6538@c Describe sh pragmas here. 6539@c Describe v850 pragmas here. 6540 6541@node Darwin Pragmas 6542@subsection Darwin Pragmas 6543 6544The following pragmas are available for all architectures running the 6545Darwin operating system. These are useful for compatibility with other 6546Mac OS compilers. 6547 6548@table @code 6549@item mark @var{tokens}@dots{} 6550@cindex pragma, mark 6551This pragma is accepted, but has no effect. 6552 6553@item options align=@var{alignment} 6554@cindex pragma, options align 6555This pragma sets the alignment of fields in structures. The values of 6556@var{alignment} may be @code{mac68k}, to emulate m68k alignment, or 6557@code{power}, to emulate PowerPC alignment. Uses of this pragma nest 6558properly; to restore the previous setting, use @code{reset} for the 6559@var{alignment}. 6560 6561@item segment @var{tokens}@dots{} 6562@cindex pragma, segment 6563This pragma is accepted, but has no effect. 6564 6565@item unused (@var{var} [, @var{var}]@dots{}) 6566@cindex pragma, unused 6567This pragma declares variables to be possibly unused. GCC will not 6568produce warnings for the listed variables. The effect is similar to 6569that of the @code{unused} attribute, except that this pragma may appear 6570anywhere within the variables' scopes. 6571@end table 6572 6573@node Solaris Pragmas 6574@subsection Solaris Pragmas 6575 6576For compatibility with the SunPRO compiler, the following pragma 6577is supported. 6578 6579@table @code 6580@item redefine_extname @var{oldname} @var{newname} 6581@cindex pragma, redefine_extname 6582 6583This pragma gives the C function @var{oldname} the assembler label 6584@var{newname}. The pragma must appear before the function declaration. 6585This pragma is equivalent to the asm labels extension (@pxref{Asm 6586Labels}). The preprocessor defines @code{__PRAGMA_REDEFINE_EXTNAME} 6587if the pragma is available. 6588@end table 6589 6590@node Tru64 Pragmas 6591@subsection Tru64 Pragmas 6592 6593For compatibility with the Compaq C compiler, the following pragma 6594is supported. 6595 6596@table @code 6597@item extern_prefix @var{string} 6598@cindex pragma, extern_prefix 6599 6600This pragma renames all subsequent function and variable declarations 6601such that @var{string} is prepended to the name. This effect may be 6602terminated by using another @code{extern_prefix} pragma with the 6603empty string. 6604 6605This pragma is similar in intent to to the asm labels extension 6606(@pxref{Asm Labels}) in that the system programmer wants to change 6607the assembly-level ABI without changing the source-level API. The 6608preprocessor defines @code{__PRAGMA_EXTERN_PREFIX} if the pragma is 6609available. 6610@end table 6611 6612@node Unnamed Fields 6613@section Unnamed struct/union fields within structs/unions. 6614@cindex struct 6615@cindex union 6616 6617For compatibility with other compilers, GCC allows you to define 6618a structure or union that contains, as fields, structures and unions 6619without names. For example: 6620 6621@example 6622struct @{ 6623 int a; 6624 union @{ 6625 int b; 6626 float c; 6627 @}; 6628 int d; 6629@} foo; 6630@end example 6631 6632In this example, the user would be able to access members of the unnamed 6633union with code like @samp{foo.b}. Note that only unnamed structs and 6634unions are allowed, you may not have, for example, an unnamed 6635@code{int}. 6636 6637You must never create such structures that cause ambiguous field definitions. 6638For example, this structure: 6639 6640@example 6641struct @{ 6642 int a; 6643 struct @{ 6644 int a; 6645 @}; 6646@} foo; 6647@end example 6648 6649It is ambiguous which @code{a} is being referred to with @samp{foo.a}. 6650Such constructs are not supported and must be avoided. In the future, 6651such constructs may be detected and treated as compilation errors. 6652 6653@node Thread-Local 6654@section Thread-Local Storage 6655@cindex Thread-Local Storage 6656@cindex @acronym{TLS} 6657@cindex __thread 6658 6659Thread-local storage (@acronym{TLS}) is a mechanism by which variables 6660are allocated such that there is one instance of the variable per extant 6661thread. The run-time model GCC uses to implement this originates 6662in the IA-64 processor-specific ABI, but has since been migrated 6663to other processors as well. It requires significant support from 6664the linker (@command{ld}), dynamic linker (@command{ld.so}), and 6665system libraries (@file{libc.so} and @file{libpthread.so}), so it 6666is not available everywhere. 6667 6668At the user level, the extension is visible with a new storage 6669class keyword: @code{__thread}. For example: 6670 6671@example 6672__thread int i; 6673extern __thread struct state s; 6674static __thread char *p; 6675@end example 6676 6677The @code{__thread} specifier may be used alone, with the @code{extern} 6678or @code{static} specifiers, but with no other storage class specifier. 6679When used with @code{extern} or @code{static}, @code{__thread} must appear 6680immediately after the other storage class specifier. 6681 6682The @code{__thread} specifier may be applied to any global, file-scoped 6683static, function-scoped static, or static data member of a class. It may 6684not be applied to block-scoped automatic or non-static data member. 6685 6686When the address-of operator is applied to a thread-local variable, it is 6687evaluated at run-time and returns the address of the current thread's 6688instance of that variable. An address so obtained may be used by any 6689thread. When a thread terminates, any pointers to thread-local variables 6690in that thread become invalid. 6691 6692No static initialization may refer to the address of a thread-local variable. 6693 6694In C++, if an initializer is present for a thread-local variable, it must 6695be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++ 6696standard. 6697 6698See @uref{http://people.redhat.com/drepper/tls.pdf, 6699ELF Handling For Thread-Local Storage} for a detailed explanation of 6700the four thread-local storage addressing models, and how the run-time 6701is expected to function. 6702 6703@menu 6704* C99 Thread-Local Edits:: 6705* C++98 Thread-Local Edits:: 6706@end menu 6707 6708@node C99 Thread-Local Edits 6709@subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage 6710 6711The following are a set of changes to ISO/IEC 9899:1999 (aka C99) 6712that document the exact semantics of the language extension. 6713 6714@itemize @bullet 6715@item 6716@cite{5.1.2 Execution environments} 6717 6718Add new text after paragraph 1 6719 6720@quotation 6721Within either execution environment, a @dfn{thread} is a flow of 6722control within a program. It is implementation defined whether 6723or not there may be more than one thread associated with a program. 6724It is implementation defined how threads beyond the first are 6725created, the name and type of the function called at thread 6726startup, and how threads may be terminated. However, objects 6727with thread storage duration shall be initialized before thread 6728startup. 6729@end quotation 6730 6731@item 6732@cite{6.2.4 Storage durations of objects} 6733 6734Add new text before paragraph 3 6735 6736@quotation 6737An object whose identifier is declared with the storage-class 6738specifier @w{@code{__thread}} has @dfn{thread storage duration}. 6739Its lifetime is the entire execution of the thread, and its 6740stored value is initialized only once, prior to thread startup. 6741@end quotation 6742 6743@item 6744@cite{6.4.1 Keywords} 6745 6746Add @code{__thread}. 6747 6748@item 6749@cite{6.7.1 Storage-class specifiers} 6750 6751Add @code{__thread} to the list of storage class specifiers in 6752paragraph 1. 6753 6754Change paragraph 2 to 6755 6756@quotation 6757With the exception of @code{__thread}, at most one storage-class 6758specifier may be given [@dots{}]. The @code{__thread} specifier may 6759be used alone, or immediately following @code{extern} or 6760@code{static}. 6761@end quotation 6762 6763Add new text after paragraph 6 6764 6765@quotation 6766The declaration of an identifier for a variable that has 6767block scope that specifies @code{__thread} shall also 6768specify either @code{extern} or @code{static}. 6769 6770The @code{__thread} specifier shall be used only with 6771variables. 6772@end quotation 6773@end itemize 6774 6775@node C++98 Thread-Local Edits 6776@subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage 6777 6778The following are a set of changes to ISO/IEC 14882:1998 (aka C++98) 6779that document the exact semantics of the language extension. 6780 6781@itemize @bullet 6782@item 6783@b{[intro.execution]} 6784 6785New text after paragraph 4 6786 6787@quotation 6788A @dfn{thread} is a flow of control within the abstract machine. 6789It is implementation defined whether or not there may be more than 6790one thread. 6791@end quotation 6792 6793New text after paragraph 7 6794 6795@quotation 6796It is unspecified whether additional action must be taken to 6797ensure when and whether side effects are visible to other threads. 6798@end quotation 6799 6800@item 6801@b{[lex.key]} 6802 6803Add @code{__thread}. 6804 6805@item 6806@b{[basic.start.main]} 6807 6808Add after paragraph 5 6809 6810@quotation 6811The thread that begins execution at the @code{main} function is called 6812the @dfn{main thread}. It is implementation defined how functions 6813beginning threads other than the main thread are designated or typed. 6814A function so designated, as well as the @code{main} function, is called 6815a @dfn{thread startup function}. It is implementation defined what 6816happens if a thread startup function returns. It is implementation 6817defined what happens to other threads when any thread calls @code{exit}. 6818@end quotation 6819 6820@item 6821@b{[basic.start.init]} 6822 6823Add after paragraph 4 6824 6825@quotation 6826The storage for an object of thread storage duration shall be 6827statically initialized before the first statement of the thread startup 6828function. An object of thread storage duration shall not require 6829dynamic initialization. 6830@end quotation 6831 6832@item 6833@b{[basic.start.term]} 6834 6835Add after paragraph 3 6836 6837@quotation 6838The type of an object with thread storage duration shall not have a 6839non-trivial destructor, nor shall it be an array type whose elements 6840(directly or indirectly) have non-trivial destructors. 6841@end quotation 6842 6843@item 6844@b{[basic.stc]} 6845 6846Add ``thread storage duration'' to the list in paragraph 1. 6847 6848Change paragraph 2 6849 6850@quotation 6851Thread, static, and automatic storage durations are associated with 6852objects introduced by declarations [@dots{}]. 6853@end quotation 6854 6855Add @code{__thread} to the list of specifiers in paragraph 3. 6856 6857@item 6858@b{[basic.stc.thread]} 6859 6860New section before @b{[basic.stc.static]} 6861 6862@quotation 6863The keyword @code{__thread} applied to a non-local object gives the 6864object thread storage duration. 6865 6866A local variable or class data member declared both @code{static} 6867and @code{__thread} gives the variable or member thread storage 6868duration. 6869@end quotation 6870 6871@item 6872@b{[basic.stc.static]} 6873 6874Change paragraph 1 6875 6876@quotation 6877All objects which have neither thread storage duration, dynamic 6878storage duration nor are local [@dots{}]. 6879@end quotation 6880 6881@item 6882@b{[dcl.stc]} 6883 6884Add @code{__thread} to the list in paragraph 1. 6885 6886Change paragraph 1 6887 6888@quotation 6889With the exception of @code{__thread}, at most one 6890@var{storage-class-specifier} shall appear in a given 6891@var{decl-specifier-seq}. The @code{__thread} specifier may 6892be used alone, or immediately following the @code{extern} or 6893@code{static} specifiers. [@dots{}] 6894@end quotation 6895 6896Add after paragraph 5 6897 6898@quotation 6899The @code{__thread} specifier can be applied only to the names of objects 6900and to anonymous unions. 6901@end quotation 6902 6903@item 6904@b{[class.mem]} 6905 6906Add after paragraph 6 6907 6908@quotation 6909Non-@code{static} members shall not be @code{__thread}. 6910@end quotation 6911@end itemize 6912 6913@node C++ Extensions 6914@chapter Extensions to the C++ Language 6915@cindex extensions, C++ language 6916@cindex C++ language extensions 6917 6918The GNU compiler provides these extensions to the C++ language (and you 6919can also use most of the C language extensions in your C++ programs). If you 6920want to write code that checks whether these features are available, you can 6921test for the GNU compiler the same way as for C programs: check for a 6922predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to 6923test specifically for GNU C++ (@pxref{Common Predefined Macros,, 6924Predefined Macros,cpp,The GNU C Preprocessor}). 6925 6926@menu 6927* Min and Max:: C++ Minimum and maximum operators. 6928* Volatiles:: What constitutes an access to a volatile object. 6929* Restricted Pointers:: C99 restricted pointers and references. 6930* Vague Linkage:: Where G++ puts inlines, vtables and such. 6931* C++ Interface:: You can use a single C++ header file for both 6932 declarations and definitions. 6933* Template Instantiation:: Methods for ensuring that exactly one copy of 6934 each needed template instantiation is emitted. 6935* Bound member functions:: You can extract a function pointer to the 6936 method denoted by a @samp{->*} or @samp{.*} expression. 6937* C++ Attributes:: Variable, function, and type attributes for C++ only. 6938* Java Exceptions:: Tweaking exception handling to work with Java. 6939* Deprecated Features:: Things might disappear from g++. 6940* Backwards Compatibility:: Compatibilities with earlier definitions of C++. 6941@end menu 6942 6943@node Min and Max 6944@section Minimum and Maximum Operators in C++ 6945 6946It is very convenient to have operators which return the ``minimum'' or the 6947``maximum'' of two arguments. In GNU C++ (but not in GNU C), 6948 6949@table @code 6950@item @var{a} <? @var{b} 6951@findex <? 6952@cindex minimum operator 6953is the @dfn{minimum}, returning the smaller of the numeric values 6954@var{a} and @var{b}; 6955 6956@item @var{a} >? @var{b} 6957@findex >? 6958@cindex maximum operator 6959is the @dfn{maximum}, returning the larger of the numeric values @var{a} 6960and @var{b}. 6961@end table 6962 6963These operations are not primitive in ordinary C++, since you can 6964use a macro to return the minimum of two things in C++, as in the 6965following example. 6966 6967@example 6968#define MIN(X,Y) ((X) < (Y) ? : (X) : (Y)) 6969@end example 6970 6971@noindent 6972You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to 6973the minimum value of variables @var{i} and @var{j}. 6974 6975However, side effects in @code{X} or @code{Y} may cause unintended 6976behavior. For example, @code{MIN (i++, j++)} will fail, incrementing 6977the smaller counter twice. The GNU C @code{typeof} extension allows you 6978to write safe macros that avoid this kind of problem (@pxref{Typeof}). 6979However, writing @code{MIN} and @code{MAX} as macros also forces you to 6980use function-call notation for a fundamental arithmetic operation. 6981Using GNU C++ extensions, you can write @w{@samp{int min = i <? j;}} 6982instead. 6983 6984Since @code{<?} and @code{>?} are built into the compiler, they properly 6985handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}} 6986works correctly. 6987 6988@node Volatiles 6989@section When is a Volatile Object Accessed? 6990@cindex accessing volatiles 6991@cindex volatile read 6992@cindex volatile write 6993@cindex volatile access 6994 6995Both the C and C++ standard have the concept of volatile objects. These 6996are normally accessed by pointers and used for accessing hardware. The 6997standards encourage compilers to refrain from optimizations 6998concerning accesses to volatile objects that it might perform on 6999non-volatile objects. The C standard leaves it implementation defined 7000as to what constitutes a volatile access. The C++ standard omits to 7001specify this, except to say that C++ should behave in a similar manner 7002to C with respect to volatiles, where possible. The minimum either 7003standard specifies is that at a sequence point all previous accesses to 7004volatile objects have stabilized and no subsequent accesses have 7005occurred. Thus an implementation is free to reorder and combine 7006volatile accesses which occur between sequence points, but cannot do so 7007for accesses across a sequence point. The use of volatiles does not 7008allow you to violate the restriction on updating objects multiple times 7009within a sequence point. 7010 7011In most expressions, it is intuitively obvious what is a read and what is 7012a write. For instance 7013 7014@example 7015volatile int *dst = @var{somevalue}; 7016volatile int *src = @var{someothervalue}; 7017*dst = *src; 7018@end example 7019 7020@noindent 7021will cause a read of the volatile object pointed to by @var{src} and stores the 7022value into the volatile object pointed to by @var{dst}. There is no 7023guarantee that these reads and writes are atomic, especially for objects 7024larger than @code{int}. 7025 7026Less obvious expressions are where something which looks like an access 7027is used in a void context. An example would be, 7028 7029@example 7030volatile int *src = @var{somevalue}; 7031*src; 7032@end example 7033 7034With C, such expressions are rvalues, and as rvalues cause a read of 7035the object, GCC interprets this as a read of the volatile being pointed 7036to. The C++ standard specifies that such expressions do not undergo 7037lvalue to rvalue conversion, and that the type of the dereferenced 7038object may be incomplete. The C++ standard does not specify explicitly 7039that it is this lvalue to rvalue conversion which is responsible for 7040causing an access. However, there is reason to believe that it is, 7041because otherwise certain simple expressions become undefined. However, 7042because it would surprise most programmers, G++ treats dereferencing a 7043pointer to volatile object of complete type in a void context as a read 7044of the object. When the object has incomplete type, G++ issues a 7045warning. 7046 7047@example 7048struct S; 7049struct T @{int m;@}; 7050volatile S *ptr1 = @var{somevalue}; 7051volatile T *ptr2 = @var{somevalue}; 7052*ptr1; 7053*ptr2; 7054@end example 7055 7056In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2} 7057causes a read of the object pointed to. If you wish to force an error on 7058the first case, you must force a conversion to rvalue with, for instance 7059a static cast, @code{static_cast<S>(*ptr1)}. 7060 7061When using a reference to volatile, G++ does not treat equivalent 7062expressions as accesses to volatiles, but instead issues a warning that 7063no volatile is accessed. The rationale for this is that otherwise it 7064becomes difficult to determine where volatile access occur, and not 7065possible to ignore the return value from functions returning volatile 7066references. Again, if you wish to force a read, cast the reference to 7067an rvalue. 7068 7069@node Restricted Pointers 7070@section Restricting Pointer Aliasing 7071@cindex restricted pointers 7072@cindex restricted references 7073@cindex restricted this pointer 7074 7075As with gcc, g++ understands the C99 feature of restricted pointers, 7076specified with the @code{__restrict__}, or @code{__restrict} type 7077qualifier. Because you cannot compile C++ by specifying the @option{-std=c99} 7078language flag, @code{restrict} is not a keyword in C++. 7079 7080In addition to allowing restricted pointers, you can specify restricted 7081references, which indicate that the reference is not aliased in the local 7082context. 7083 7084@example 7085void fn (int *__restrict__ rptr, int &__restrict__ rref) 7086@{ 7087 /* @r{@dots{}} */ 7088@} 7089@end example 7090 7091@noindent 7092In the body of @code{fn}, @var{rptr} points to an unaliased integer and 7093@var{rref} refers to a (different) unaliased integer. 7094 7095You may also specify whether a member function's @var{this} pointer is 7096unaliased by using @code{__restrict__} as a member function qualifier. 7097 7098@example 7099void T::fn () __restrict__ 7100@{ 7101 /* @r{@dots{}} */ 7102@} 7103@end example 7104 7105@noindent 7106Within the body of @code{T::fn}, @var{this} will have the effective 7107definition @code{T *__restrict__ const this}. Notice that the 7108interpretation of a @code{__restrict__} member function qualifier is 7109different to that of @code{const} or @code{volatile} qualifier, in that it 7110is applied to the pointer rather than the object. This is consistent with 7111other compilers which implement restricted pointers. 7112 7113As with all outermost parameter qualifiers, @code{__restrict__} is 7114ignored in function definition matching. This means you only need to 7115specify @code{__restrict__} in a function definition, rather than 7116in a function prototype as well. 7117 7118@node Vague Linkage 7119@section Vague Linkage 7120@cindex vague linkage 7121 7122There are several constructs in C++ which require space in the object 7123file but are not clearly tied to a single translation unit. We say that 7124these constructs have ``vague linkage''. Typically such constructs are 7125emitted wherever they are needed, though sometimes we can be more 7126clever. 7127 7128@table @asis 7129@item Inline Functions 7130Inline functions are typically defined in a header file which can be 7131included in many different compilations. Hopefully they can usually be 7132inlined, but sometimes an out-of-line copy is necessary, if the address 7133of the function is taken or if inlining fails. In general, we emit an 7134out-of-line copy in all translation units where one is needed. As an 7135exception, we only emit inline virtual functions with the vtable, since 7136it will always require a copy. 7137 7138Local static variables and string constants used in an inline function 7139are also considered to have vague linkage, since they must be shared 7140between all inlined and out-of-line instances of the function. 7141 7142@item VTables 7143@cindex vtable 7144C++ virtual functions are implemented in most compilers using a lookup 7145table, known as a vtable. The vtable contains pointers to the virtual 7146functions provided by a class, and each object of the class contains a 7147pointer to its vtable (or vtables, in some multiple-inheritance 7148situations). If the class declares any non-inline, non-pure virtual 7149functions, the first one is chosen as the ``key method'' for the class, 7150and the vtable is only emitted in the translation unit where the key 7151method is defined. 7152 7153@emph{Note:} If the chosen key method is later defined as inline, the 7154vtable will still be emitted in every translation unit which defines it. 7155Make sure that any inline virtuals are declared inline in the class 7156body, even if they are not defined there. 7157 7158@item type_info objects 7159@cindex type_info 7160@cindex RTTI 7161C++ requires information about types to be written out in order to 7162implement @samp{dynamic_cast}, @samp{typeid} and exception handling. 7163For polymorphic classes (classes with virtual functions), the type_info 7164object is written out along with the vtable so that @samp{dynamic_cast} 7165can determine the dynamic type of a class object at runtime. For all 7166other types, we write out the type_info object when it is used: when 7167applying @samp{typeid} to an expression, throwing an object, or 7168referring to a type in a catch clause or exception specification. 7169 7170@item Template Instantiations 7171Most everything in this section also applies to template instantiations, 7172but there are other options as well. 7173@xref{Template Instantiation,,Where's the Template?}. 7174 7175@end table 7176 7177When used with GNU ld version 2.8 or later on an ELF system such as 7178Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of 7179these constructs will be discarded at link time. This is known as 7180COMDAT support. 7181 7182On targets that don't support COMDAT, but do support weak symbols, GCC 7183will use them. This way one copy will override all the others, but 7184the unused copies will still take up space in the executable. 7185 7186For targets which do not support either COMDAT or weak symbols, 7187most entities with vague linkage will be emitted as local symbols to 7188avoid duplicate definition errors from the linker. This will not happen 7189for local statics in inlines, however, as having multiple copies will 7190almost certainly break things. 7191 7192@xref{C++ Interface,,Declarations and Definitions in One Header}, for 7193another way to control placement of these constructs. 7194 7195@node C++ Interface 7196@section Declarations and Definitions in One Header 7197 7198@cindex interface and implementation headers, C++ 7199@cindex C++ interface and implementation headers 7200C++ object definitions can be quite complex. In principle, your source 7201code will need two kinds of things for each object that you use across 7202more than one source file. First, you need an @dfn{interface} 7203specification, describing its structure with type declarations and 7204function prototypes. Second, you need the @dfn{implementation} itself. 7205It can be tedious to maintain a separate interface description in a 7206header file, in parallel to the actual implementation. It is also 7207dangerous, since separate interface and implementation definitions may 7208not remain parallel. 7209 7210@cindex pragmas, interface and implementation 7211With GNU C++, you can use a single header file for both purposes. 7212 7213@quotation 7214@emph{Warning:} The mechanism to specify this is in transition. For the 7215nonce, you must use one of two @code{#pragma} commands; in a future 7216release of GNU C++, an alternative mechanism will make these 7217@code{#pragma} commands unnecessary. 7218@end quotation 7219 7220The header file contains the full definitions, but is marked with 7221@samp{#pragma interface} in the source code. This allows the compiler 7222to use the header file only as an interface specification when ordinary 7223source files incorporate it with @code{#include}. In the single source 7224file where the full implementation belongs, you can use either a naming 7225convention or @samp{#pragma implementation} to indicate this alternate 7226use of the header file. 7227 7228@table @code 7229@item #pragma interface 7230@itemx #pragma interface "@var{subdir}/@var{objects}.h" 7231@kindex #pragma interface 7232Use this directive in @emph{header files} that define object classes, to save 7233space in most of the object files that use those classes. Normally, 7234local copies of certain information (backup copies of inline member 7235functions, debugging information, and the internal tables that implement 7236virtual functions) must be kept in each object file that includes class 7237definitions. You can use this pragma to avoid such duplication. When a 7238header file containing @samp{#pragma interface} is included in a 7239compilation, this auxiliary information will not be generated (unless 7240the main input source file itself uses @samp{#pragma implementation}). 7241Instead, the object files will contain references to be resolved at link 7242time. 7243 7244The second form of this directive is useful for the case where you have 7245multiple headers with the same name in different directories. If you 7246use this form, you must specify the same string to @samp{#pragma 7247implementation}. 7248 7249@item #pragma implementation 7250@itemx #pragma implementation "@var{objects}.h" 7251@kindex #pragma implementation 7252Use this pragma in a @emph{main input file}, when you want full output from 7253included header files to be generated (and made globally visible). The 7254included header file, in turn, should use @samp{#pragma interface}. 7255Backup copies of inline member functions, debugging information, and the 7256internal tables used to implement virtual functions are all generated in 7257implementation files. 7258 7259@cindex implied @code{#pragma implementation} 7260@cindex @code{#pragma implementation}, implied 7261@cindex naming convention, implementation headers 7262If you use @samp{#pragma implementation} with no argument, it applies to 7263an include file with the same basename@footnote{A file's @dfn{basename} 7264was the name stripped of all leading path information and of trailing 7265suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source 7266file. For example, in @file{allclass.cc}, giving just 7267@samp{#pragma implementation} 7268by itself is equivalent to @samp{#pragma implementation "allclass.h"}. 7269 7270In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as 7271an implementation file whenever you would include it from 7272@file{allclass.cc} even if you never specified @samp{#pragma 7273implementation}. This was deemed to be more trouble than it was worth, 7274however, and disabled. 7275 7276If you use an explicit @samp{#pragma implementation}, it must appear in 7277your source file @emph{before} you include the affected header files. 7278 7279Use the string argument if you want a single implementation file to 7280include code from multiple header files. (You must also use 7281@samp{#include} to include the header file; @samp{#pragma 7282implementation} only specifies how to use the file---it doesn't actually 7283include it.) 7284 7285There is no way to split up the contents of a single header file into 7286multiple implementation files. 7287@end table 7288 7289@cindex inlining and C++ pragmas 7290@cindex C++ pragmas, effect on inlining 7291@cindex pragmas in C++, effect on inlining 7292@samp{#pragma implementation} and @samp{#pragma interface} also have an 7293effect on function inlining. 7294 7295If you define a class in a header file marked with @samp{#pragma 7296interface}, the effect on a function defined in that class is similar to 7297an explicit @code{extern} declaration---the compiler emits no code at 7298all to define an independent version of the function. Its definition 7299is used only for inlining with its callers. 7300 7301@opindex fno-implement-inlines 7302Conversely, when you include the same header file in a main source file 7303that declares it as @samp{#pragma implementation}, the compiler emits 7304code for the function itself; this defines a version of the function 7305that can be found via pointers (or by callers compiled without 7306inlining). If all calls to the function can be inlined, you can avoid 7307emitting the function by compiling with @option{-fno-implement-inlines}. 7308If any calls were not inlined, you will get linker errors. 7309 7310@node Template Instantiation 7311@section Where's the Template? 7312@cindex template instantiation 7313 7314C++ templates are the first language feature to require more 7315intelligence from the environment than one usually finds on a UNIX 7316system. Somehow the compiler and linker have to make sure that each 7317template instance occurs exactly once in the executable if it is needed, 7318and not at all otherwise. There are two basic approaches to this 7319problem, which I will refer to as the Borland model and the Cfront model. 7320 7321@table @asis 7322@item Borland model 7323Borland C++ solved the template instantiation problem by adding the code 7324equivalent of common blocks to their linker; the compiler emits template 7325instances in each translation unit that uses them, and the linker 7326collapses them together. The advantage of this model is that the linker 7327only has to consider the object files themselves; there is no external 7328complexity to worry about. This disadvantage is that compilation time 7329is increased because the template code is being compiled repeatedly. 7330Code written for this model tends to include definitions of all 7331templates in the header file, since they must be seen to be 7332instantiated. 7333 7334@item Cfront model 7335The AT&T C++ translator, Cfront, solved the template instantiation 7336problem by creating the notion of a template repository, an 7337automatically maintained place where template instances are stored. A 7338more modern version of the repository works as follows: As individual 7339object files are built, the compiler places any template definitions and 7340instantiations encountered in the repository. At link time, the link 7341wrapper adds in the objects in the repository and compiles any needed 7342instances that were not previously emitted. The advantages of this 7343model are more optimal compilation speed and the ability to use the 7344system linker; to implement the Borland model a compiler vendor also 7345needs to replace the linker. The disadvantages are vastly increased 7346complexity, and thus potential for error; for some code this can be 7347just as transparent, but in practice it can been very difficult to build 7348multiple programs in one directory and one program in multiple 7349directories. Code written for this model tends to separate definitions 7350of non-inline member templates into a separate file, which should be 7351compiled separately. 7352@end table 7353 7354When used with GNU ld version 2.8 or later on an ELF system such as 7355Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the 7356Borland model. On other systems, g++ implements neither automatic 7357model. 7358 7359A future version of g++ will support a hybrid model whereby the compiler 7360will emit any instantiations for which the template definition is 7361included in the compile, and store template definitions and 7362instantiation context information into the object file for the rest. 7363The link wrapper will extract that information as necessary and invoke 7364the compiler to produce the remaining instantiations. The linker will 7365then combine duplicate instantiations. 7366 7367In the mean time, you have the following options for dealing with 7368template instantiations: 7369 7370@enumerate 7371@item 7372@opindex frepo 7373Compile your template-using code with @option{-frepo}. The compiler will 7374generate files with the extension @samp{.rpo} listing all of the 7375template instantiations used in the corresponding object files which 7376could be instantiated there; the link wrapper, @samp{collect2}, will 7377then update the @samp{.rpo} files to tell the compiler where to place 7378those instantiations and rebuild any affected object files. The 7379link-time overhead is negligible after the first pass, as the compiler 7380will continue to place the instantiations in the same files. 7381 7382This is your best option for application code written for the Borland 7383model, as it will just work. Code written for the Cfront model will 7384need to be modified so that the template definitions are available at 7385one or more points of instantiation; usually this is as simple as adding 7386@code{#include <tmethods.cc>} to the end of each template header. 7387 7388For library code, if you want the library to provide all of the template 7389instantiations it needs, just try to link all of its object files 7390together; the link will fail, but cause the instantiations to be 7391generated as a side effect. Be warned, however, that this may cause 7392conflicts if multiple libraries try to provide the same instantiations. 7393For greater control, use explicit instantiation as described in the next 7394option. 7395 7396@item 7397@opindex fno-implicit-templates 7398Compile your code with @option{-fno-implicit-templates} to disable the 7399implicit generation of template instances, and explicitly instantiate 7400all the ones you use. This approach requires more knowledge of exactly 7401which instances you need than do the others, but it's less 7402mysterious and allows greater control. You can scatter the explicit 7403instantiations throughout your program, perhaps putting them in the 7404translation units where the instances are used or the translation units 7405that define the templates themselves; you can put all of the explicit 7406instantiations you need into one big file; or you can create small files 7407like 7408 7409@example 7410#include "Foo.h" 7411#include "Foo.cc" 7412 7413template class Foo<int>; 7414template ostream& operator << 7415 (ostream&, const Foo<int>&); 7416@end example 7417 7418for each of the instances you need, and create a template instantiation 7419library from those. 7420 7421If you are using Cfront-model code, you can probably get away with not 7422using @option{-fno-implicit-templates} when compiling files that don't 7423@samp{#include} the member template definitions. 7424 7425If you use one big file to do the instantiations, you may want to 7426compile it without @option{-fno-implicit-templates} so you get all of the 7427instances required by your explicit instantiations (but not by any 7428other files) without having to specify them as well. 7429 7430g++ has extended the template instantiation syntax given in the ISO 7431standard to allow forward declaration of explicit instantiations 7432(with @code{extern}), instantiation of the compiler support data for a 7433template class (i.e.@: the vtable) without instantiating any of its 7434members (with @code{inline}), and instantiation of only the static data 7435members of a template class, without the support data or member 7436functions (with (@code{static}): 7437 7438@example 7439extern template int max (int, int); 7440inline template class Foo<int>; 7441static template class Foo<int>; 7442@end example 7443 7444@item 7445Do nothing. Pretend g++ does implement automatic instantiation 7446management. Code written for the Borland model will work fine, but 7447each translation unit will contain instances of each of the templates it 7448uses. In a large program, this can lead to an unacceptable amount of code 7449duplication. 7450 7451@xref{C++ Interface,,Declarations and Definitions in One Header}, for 7452more discussion of these pragmas. 7453@end enumerate 7454 7455@node Bound member functions 7456@section Extracting the function pointer from a bound pointer to member function 7457@cindex pmf 7458@cindex pointer to member function 7459@cindex bound pointer to member function 7460 7461In C++, pointer to member functions (PMFs) are implemented using a wide 7462pointer of sorts to handle all the possible call mechanisms; the PMF 7463needs to store information about how to adjust the @samp{this} pointer, 7464and if the function pointed to is virtual, where to find the vtable, and 7465where in the vtable to look for the member function. If you are using 7466PMFs in an inner loop, you should really reconsider that decision. If 7467that is not an option, you can extract the pointer to the function that 7468would be called for a given object/PMF pair and call it directly inside 7469the inner loop, to save a bit of time. 7470 7471Note that you will still be paying the penalty for the call through a 7472function pointer; on most modern architectures, such a call defeats the 7473branch prediction features of the CPU@. This is also true of normal 7474virtual function calls. 7475 7476The syntax for this extension is 7477 7478@example 7479extern A a; 7480extern int (A::*fp)(); 7481typedef int (*fptr)(A *); 7482 7483fptr p = (fptr)(a.*fp); 7484@end example 7485 7486For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}), 7487no object is needed to obtain the address of the function. They can be 7488converted to function pointers directly: 7489 7490@example 7491fptr p1 = (fptr)(&A::foo); 7492@end example 7493 7494@opindex Wno-pmf-conversions 7495You must specify @option{-Wno-pmf-conversions} to use this extension. 7496 7497@node C++ Attributes 7498@section C++-Specific Variable, Function, and Type Attributes 7499 7500Some attributes only make sense for C++ programs. 7501 7502@table @code 7503@item init_priority (@var{priority}) 7504@cindex init_priority attribute 7505 7506 7507In Standard C++, objects defined at namespace scope are guaranteed to be 7508initialized in an order in strict accordance with that of their definitions 7509@emph{in a given translation unit}. No guarantee is made for initializations 7510across translation units. However, GNU C++ allows users to control the 7511order of initialization of objects defined at namespace scope with the 7512@code{init_priority} attribute by specifying a relative @var{priority}, 7513a constant integral expression currently bounded between 101 and 65535 7514inclusive. Lower numbers indicate a higher priority. 7515 7516In the following example, @code{A} would normally be created before 7517@code{B}, but the @code{init_priority} attribute has reversed that order: 7518 7519@smallexample 7520Some_Class A __attribute__ ((init_priority (2000))); 7521Some_Class B __attribute__ ((init_priority (543))); 7522@end smallexample 7523 7524@noindent 7525Note that the particular values of @var{priority} do not matter; only their 7526relative ordering. 7527 7528@item java_interface 7529@cindex java_interface attribute 7530 7531This type attribute informs C++ that the class is a Java interface. It may 7532only be applied to classes declared within an @code{extern "Java"} block. 7533Calls to methods declared in this interface will be dispatched using GCJ's 7534interface table mechanism, instead of regular virtual table dispatch. 7535 7536@end table 7537 7538@node Java Exceptions 7539@section Java Exceptions 7540 7541The Java language uses a slightly different exception handling model 7542from C++. Normally, GNU C++ will automatically detect when you are 7543writing C++ code that uses Java exceptions, and handle them 7544appropriately. However, if C++ code only needs to execute destructors 7545when Java exceptions are thrown through it, GCC will guess incorrectly. 7546Sample problematic code is: 7547 7548@smallexample 7549 struct S @{ ~S(); @}; 7550 extern void bar(); // is written in Java, and may throw exceptions 7551 void foo() 7552 @{ 7553 S s; 7554 bar(); 7555 @} 7556@end smallexample 7557 7558@noindent 7559The usual effect of an incorrect guess is a link failure, complaining of 7560a missing routine called @samp{__gxx_personality_v0}. 7561 7562You can inform the compiler that Java exceptions are to be used in a 7563translation unit, irrespective of what it might think, by writing 7564@samp{@w{#pragma GCC java_exceptions}} at the head of the file. This 7565@samp{#pragma} must appear before any functions that throw or catch 7566exceptions, or run destructors when exceptions are thrown through them. 7567 7568You cannot mix Java and C++ exceptions in the same translation unit. It 7569is believed to be safe to throw a C++ exception from one file through 7570another file compiled for the Java exception model, or vice versa, but 7571there may be bugs in this area. 7572 7573@node Deprecated Features 7574@section Deprecated Features 7575 7576In the past, the GNU C++ compiler was extended to experiment with new 7577features, at a time when the C++ language was still evolving. Now that 7578the C++ standard is complete, some of those features are superseded by 7579superior alternatives. Using the old features might cause a warning in 7580some cases that the feature will be dropped in the future. In other 7581cases, the feature might be gone already. 7582 7583While the list below is not exhaustive, it documents some of the options 7584that are now deprecated: 7585 7586@table @code 7587@item -fexternal-templates 7588@itemx -falt-external-templates 7589These are two of the many ways for g++ to implement template 7590instantiation. @xref{Template Instantiation}. The C++ standard clearly 7591defines how template definitions have to be organized across 7592implementation units. g++ has an implicit instantiation mechanism that 7593should work just fine for standard-conforming code. 7594 7595@item -fstrict-prototype 7596@itemx -fno-strict-prototype 7597Previously it was possible to use an empty prototype parameter list to 7598indicate an unspecified number of parameters (like C), rather than no 7599parameters, as C++ demands. This feature has been removed, except where 7600it is required for backwards compatibility @xref{Backwards Compatibility}. 7601@end table 7602 7603The named return value extension has been deprecated, and is now 7604removed from g++. 7605 7606The use of initializer lists with new expressions has been deprecated, 7607and is now removed from g++. 7608 7609Floating and complex non-type template parameters have been deprecated, 7610and are now removed from g++. 7611 7612The implicit typename extension has been deprecated and will be removed 7613from g++ at some point. In some cases g++ determines that a dependent 7614type such as @code{TPL<T>::X} is a type without needing a 7615@code{typename} keyword, contrary to the standard. 7616 7617@node Backwards Compatibility 7618@section Backwards Compatibility 7619@cindex Backwards Compatibility 7620@cindex ARM [Annotated C++ Reference Manual] 7621 7622Now that there is a definitive ISO standard C++, G++ has a specification 7623to adhere to. The C++ language evolved over time, and features that 7624used to be acceptable in previous drafts of the standard, such as the ARM 7625[Annotated C++ Reference Manual], are no longer accepted. In order to allow 7626compilation of C++ written to such drafts, G++ contains some backwards 7627compatibilities. @emph{All such backwards compatibility features are 7628liable to disappear in future versions of G++.} They should be considered 7629deprecated @xref{Deprecated Features}. 7630 7631@table @code 7632@item For scope 7633If a variable is declared at for scope, it used to remain in scope until 7634the end of the scope which contained the for statement (rather than just 7635within the for scope). G++ retains this, but issues a warning, if such a 7636variable is accessed outside the for scope. 7637 7638@item Implicit C language 7639Old C system header files did not contain an @code{extern "C" @{@dots{}@}} 7640scope to set the language. On such systems, all header files are 7641implicitly scoped inside a C language scope. Also, an empty prototype 7642@code{()} will be treated as an unspecified number of arguments, rather 7643than no arguments, as C++ demands. 7644@end table 7645