extend.texi revision 169689
1224133Sdim@c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1996, 1998, 1999, 2000, 2224133Sdim@c 2001, 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc. 3224133Sdim 4224133Sdim@c This is part of the GCC manual. 5224133Sdim@c For copying conditions, see the file gcc.texi. 6224133Sdim 7224133Sdim@node C Extensions 8224133Sdim@chapter Extensions to the C Language Family 9224133Sdim@cindex extensions, C language 10224133Sdim@cindex C language extensions 11224133Sdim 12224133Sdim@opindex pedantic 13224133SdimGNU C provides several language features not found in ISO standard C@. 14224133Sdim(The @option{-pedantic} option directs GCC to print a warning message if 15224133Sdimany of these features is used.) To test for the availability of these 16224133Sdimfeatures in conditional compilation, check for a predefined macro 17224133Sdim@code{__GNUC__}, which is always defined under GCC@. 18224133Sdim 19226633SdimThese extensions are available in C and Objective-C@. Most of them are 20234353Sdimalso available in C++. @xref{C++ Extensions,,Extensions to the 21224133SdimC++ Language}, for extensions that apply @emph{only} to C++. 22224133Sdim 23224133SdimSome features that are in ISO C99 but not C89 or C++ are also, as 24224133Sdimextensions, accepted by GCC in C89 mode and in C++. 25226633Sdim 26226633Sdim@menu 27226633Sdim* Statement Exprs:: Putting statements and declarations inside expressions. 28234353Sdim* Local Labels:: Labels local to a block. 29234353Sdim* Labels as Values:: Getting pointers to labels, and computed gotos. 30234353Sdim* Nested Functions:: As in Algol and Pascal, lexical scoping of functions. 31226633Sdim* Constructing Calls:: Dispatching a call to another function. 32234353Sdim* Typeof:: @code{typeof}: referring to the type of an expression. 33234353Sdim* Conditionals:: Omitting the middle operand of a @samp{?:} expression. 34234353Sdim* Long Long:: Double-word integers---@code{long long int}. 35234353Sdim* Complex:: Data types for complex numbers. 36234353Sdim* Decimal Float:: Decimal Floating Types. 37234353Sdim* Hex Floats:: Hexadecimal floating-point constants. 38234353Sdim* Zero Length:: Zero-length arrays. 39226633Sdim* Variable Length:: Arrays whose length is computed at run time. 40226633Sdim* Empty Structures:: Structures with no members. 41226633Sdim* Variadic Macros:: Macros with a variable number of arguments. 42226633Sdim* Escaped Newlines:: Slightly looser rules for escaped newlines. 43226633Sdim* Subscripting:: Any array can be subscripted, even if not an lvalue. 44226633Sdim* Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. 45226633Sdim* Initializers:: Non-constant initializers. 46226633Sdim* Compound Literals:: Compound literals give structures, unions 47226633Sdim or arrays as values. 48226633Sdim* Designated Inits:: Labeling elements of initializers. 49226633Sdim* Cast to Union:: Casting to union type from any member of the union. 50226633Sdim* Case Ranges:: `case 1 ... 9' and such. 51226633Sdim* Mixed Declarations:: Mixing declarations and code. 52234353Sdim* Function Attributes:: Declaring that functions have no side effects, 53226633Sdim or that they can never return. 54226633Sdim* Attribute Syntax:: Formal syntax for attributes. 55226633Sdim* Function Prototypes:: Prototype declarations and old-style definitions. 56234353Sdim* C++ Comments:: C++ comments are recognized. 57226633Sdim* Dollar Signs:: Dollar sign is allowed in identifiers. 58226633Sdim* Character Escapes:: @samp{\e} stands for the character @key{ESC}. 59226633Sdim* Variable Attributes:: Specifying attributes of variables. 60226633Sdim* Type Attributes:: Specifying attributes of types. 61226633Sdim* Alignment:: Inquiring about the alignment of a type or variable. 62226633Sdim* Inline:: Defining inline functions (as fast as macros). 63226633Sdim* Extended Asm:: Assembler instructions with C expressions as operands. 64226633Sdim (With them you can define ``built-in'' functions.) 65226633Sdim* Constraints:: Constraints for asm operands 66226633Sdim* Asm Labels:: Specifying the assembler name to use for a C symbol. 67226633Sdim* Explicit Reg Vars:: Defining variables residing in specified registers. 68226633Sdim* Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. 69226633Sdim* Incomplete Enums:: @code{enum foo;}, with details to follow. 70226633Sdim* Function Names:: Printable strings which are the name of the current 71226633Sdim function. 72226633Sdim* Return Address:: Getting the return or frame address of a function. 73226633Sdim* Vector Extensions:: Using vector instructions through built-in functions. 74226633Sdim* Offsetof:: Special syntax for implementing @code{offsetof}. 75226633Sdim* Atomic Builtins:: Built-in functions for atomic memory access. 76226633Sdim* Object Size Checking:: Built-in functions for limited buffer overflow 77226633Sdim checking. 78226633Sdim* Other Builtins:: Other built-in functions. 79226633Sdim* Target Builtins:: Built-in functions specific to particular targets. 80226633Sdim* Target Format Checks:: Format checks specific to particular targets. 81226633Sdim* Pragmas:: Pragmas accepted by GCC. 82226633Sdim* Unnamed Fields:: Unnamed struct/union fields within structs/unions. 83226633Sdim* Thread-Local:: Per-thread variables. 84226633Sdim@end menu 85226633Sdim 86226633Sdim@node Statement Exprs 87226633Sdim@section Statements and Declarations in Expressions 88226633Sdim@cindex statements inside expressions 89226633Sdim@cindex declarations inside expressions 90226633Sdim@cindex expressions containing statements 91226633Sdim@cindex macros, statements in expressions 92226633Sdim 93226633Sdim@c the above section title wrapped and causes an underfull hbox.. i 94226633Sdim@c changed it from "within" to "in". --mew 4feb93 95226633SdimA compound statement enclosed in parentheses may appear as an expression 96226633Sdimin GNU C@. This allows you to use loops, switches, and local variables 97226633Sdimwithin an expression. 98224133Sdim 99224133SdimRecall that a compound statement is a sequence of statements surrounded 100224133Sdimby braces; in this construct, parentheses go around the braces. For 101224133Sdimexample: 102224133Sdim 103224133Sdim@smallexample 104224133Sdim(@{ int y = foo (); int z; 105224133Sdim if (y > 0) z = y; 106224133Sdim else z = - y; 107224133Sdim z; @}) 108224133Sdim@end smallexample 109239462Sdim 110239462Sdim@noindent 111239462Sdimis a valid (though slightly more complex than necessary) expression 112239462Sdimfor the absolute value of @code{foo ()}. 113239462Sdim 114239462SdimThe last thing in the compound statement should be an expression 115239462Sdimfollowed by a semicolon; the value of this subexpression serves as the 116239462Sdimvalue of the entire construct. (If you use some other kind of statement 117239462Sdimlast within the braces, the construct has type @code{void}, and thus 118239462Sdimeffectively no value.) 119239462Sdim 120239462SdimThis feature is especially useful in making macro definitions ``safe'' (so 121224133Sdimthat they evaluate each operand exactly once). For example, the 122224133Sdim``maximum'' function is commonly defined as a macro in standard C as 123224133Sdimfollows: 124224133Sdim 125224133Sdim@smallexample 126224133Sdim#define max(a,b) ((a) > (b) ? (a) : (b)) 127224133Sdim@end smallexample 128224133Sdim 129224133Sdim@noindent 130224133Sdim@cindex side effects, macro argument 131224133SdimBut this definition computes either @var{a} or @var{b} twice, with bad 132224133Sdimresults if the operand has side effects. In GNU C, if you know the 133224133Sdimtype of the operands (here taken as @code{int}), you can define 134224133Sdimthe macro safely as follows: 135224133Sdim 136226633Sdim@smallexample 137226633Sdim#define maxint(a,b) \ 138234353Sdim (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @}) 139234353Sdim@end smallexample 140234353Sdim 141234353SdimEmbedded statements are not allowed in constant expressions, such as 142234353Sdimthe value of an enumeration constant, the width of a bit-field, or 143234353Sdimthe initial value of a static variable. 144234353Sdim 145234353SdimIf you don't know the type of the operand, you can still do this, but you 146234353Sdimmust use @code{typeof} (@pxref{Typeof}). 147224133Sdim 148226633SdimIn G++, the result value of a statement expression undergoes array and 149226633Sdimfunction pointer decay, and is returned by value to the enclosing 150226633Sdimexpression. For instance, if @code{A} is a class, then 151226633Sdim 152226633Sdim@smallexample 153239462Sdim A a; 154239462Sdim 155239462Sdim (@{a;@}).Foo () 156239462Sdim@end smallexample 157234353Sdim 158234353Sdim@noindent 159234353Sdimwill construct a temporary @code{A} object to hold the result of the 160239462Sdimstatement expression, and that will be used to invoke @code{Foo}. 161239462SdimTherefore the @code{this} pointer observed by @code{Foo} will not be the 162234353Sdimaddress of @code{a}. 163234353Sdim 164234353SdimAny temporaries created within a statement within a statement expression 165234353Sdimwill be destroyed at the statement's end. This makes statement 166234353Sdimexpressions inside macros slightly different from function calls. In 167234353Sdimthe latter case temporaries introduced during argument evaluation will 168234353Sdimbe destroyed at the end of the statement that includes the function 169234353Sdimcall. In the statement expression case they will be destroyed during 170234353Sdimthe statement expression. For instance, 171226633Sdim 172224133Sdim@smallexample 173224133Sdim#define macro(a) (@{__typeof__(a) b = (a); b + 3; @}) 174239462Sdimtemplate<typename T> T function(T a) @{ T b = a; return b + 3; @} 175239462Sdim 176239462Sdimvoid foo () 177239462Sdim@{ 178239462Sdim macro (X ()); 179239462Sdim function (X ()); 180239462Sdim@} 181239462Sdim@end smallexample 182239462Sdim 183239462Sdim@noindent 184239462Sdimwill have different places where temporaries are destroyed. For the 185239462Sdim@code{macro} case, the temporary @code{X} will be destroyed just after 186239462Sdimthe initialization of @code{b}. In the @code{function} case that 187239462Sdimtemporary will be destroyed when the function returns. 188239462Sdim 189239462SdimThese considerations mean that it is probably a bad idea to use 190239462Sdimstatement-expressions of this form in header files that are designed to 191239462Sdimwork with C++. (Note that some versions of the GNU C Library contained 192239462Sdimheader files using statement-expression that lead to precisely this 193239462Sdimbug.) 194239462Sdim 195239462SdimJumping into a statement expression with @code{goto} or using a 196239462Sdim@code{switch} statement outside the statement expression with a 197239462Sdim@code{case} or @code{default} label inside the statement expression is 198239462Sdimnot permitted. Jumping into a statement expression with a computed 199239462Sdim@code{goto} (@pxref{Labels as Values}) yields undefined behavior. 200239462SdimJumping out of a statement expression is permitted, but if the 201239462Sdimstatement expression is part of a larger expression then it is 202239462Sdimunspecified which other subexpressions of that expression have been 203239462Sdimevaluated except where the language definition requires certain 204239462Sdimsubexpressions to be evaluated before or after the statement 205239462Sdimexpression. In any case, as with a function call the evaluation of a 206239462Sdimstatement expression is not interleaved with the evaluation of other 207239462Sdimparts of the containing expression. For example, 208239462Sdim 209239462Sdim@smallexample 210239462Sdim foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz(); 211239462Sdim@end smallexample 212239462Sdim 213239462Sdim@noindent 214239462Sdimwill call @code{foo} and @code{bar1} and will not call @code{baz} but 215239462Sdimmay or may not call @code{bar2}. If @code{bar2} is called, it will be 216239462Sdimcalled after @code{foo} and before @code{bar1} 217239462Sdim 218239462Sdim@node Local Labels 219239462Sdim@section Locally Declared Labels 220239462Sdim@cindex local labels 221239462Sdim@cindex macros, local labels 222239462Sdim 223239462SdimGCC allows you to declare @dfn{local labels} in any nested block 224239462Sdimscope. A local label is just like an ordinary label, but you can 225239462Sdimonly reference it (with a @code{goto} statement, or by taking its 226239462Sdimaddress) within the block in which it was declared. 227239462Sdim 228224133SdimA local label declaration looks like this: 229224133Sdim 230226633Sdim@smallexample 231234353Sdim__label__ @var{label}; 232239462Sdim@end smallexample 233239462Sdim 234239462Sdim@noindent 235239462Sdimor 236234353Sdim 237239462Sdim@smallexample 238239462Sdim__label__ @var{label1}, @var{label2}, /* @r{@dots{}} */; 239224133Sdim@end smallexample 240224133Sdim 241226633SdimLocal label declarations must come at the beginning of the block, 242226633Sdimbefore any ordinary declarations or statements. 243239462Sdim 244239462SdimThe label declaration defines the label @emph{name}, but does not define 245226633Sdimthe label itself. You must do this in the usual way, with 246239462Sdim@code{@var{label}:}, within the statements of the statement expression. 247239462Sdim 248234353SdimThe local label feature is useful for complex macros. If a macro 249234353Sdimcontains nested loops, a @code{goto} can be useful for breaking out of 250239462Sdimthem. However, an ordinary label whose scope is the whole function 251224133Sdimcannot be used: if the macro can be expanded several times in one 252226633Sdimfunction, the label will be multiply defined in that function. A 253234353Sdimlocal label avoids this problem. For example: 254226633Sdim 255226633Sdim@smallexample 256234353Sdim#define SEARCH(value, array, target) \ 257234353Sdimdo @{ \ 258234353Sdim __label__ found; \ 259234353Sdim typeof (target) _SEARCH_target = (target); \ 260234353Sdim typeof (*(array)) *_SEARCH_array = (array); \ 261234353Sdim int i, j; \ 262234353Sdim int value; \ 263234353Sdim for (i = 0; i < max; i++) \ 264234353Sdim for (j = 0; j < max; j++) \ 265226633Sdim if (_SEARCH_array[i][j] == _SEARCH_target) \ 266234353Sdim @{ (value) = i; goto found; @} \ 267234353Sdim (value) = -1; \ 268226633Sdim found:; \ 269226633Sdim@} while (0) 270234353Sdim@end smallexample 271234353Sdim 272234353SdimThis could also be written using a statement-expression: 273234353Sdim 274234353Sdim@smallexample 275234353Sdim#define SEARCH(array, target) \ 276234353Sdim(@{ \ 277234353Sdim __label__ found; \ 278234353Sdim typeof (target) _SEARCH_target = (target); \ 279226633Sdim typeof (*(array)) *_SEARCH_array = (array); \ 280234353Sdim int i, j; \ 281226633Sdim int value; \ 282226633Sdim for (i = 0; i < max; i++) \ 283226633Sdim for (j = 0; j < max; j++) \ 284226633Sdim if (_SEARCH_array[i][j] == _SEARCH_target) \ 285226633Sdim @{ value = i; goto found; @} \ 286226633Sdim value = -1; \ 287226633Sdim found: \ 288226633Sdim value; \ 289226633Sdim@}) 290226633Sdim@end smallexample 291226633Sdim 292226633SdimLocal label declarations also make the labels they declare visible to 293226633Sdimnested functions, if there are any. @xref{Nested Functions}, for details. 294226633Sdim 295226633Sdim@node Labels as Values 296224133Sdim@section Labels as Values 297224133Sdim@cindex labels as values 298224133Sdim@cindex computed gotos 299224133Sdim@cindex goto with computed label 300224133Sdim@cindex address of a label 301224133Sdim 302224133SdimYou can get the address of a label defined in the current function 303224133Sdim(or a containing function) with the unary operator @samp{&&}. The 304224133Sdimvalue has type @code{void *}. This value is a constant and can be used 305224133Sdimwherever a constant of that type is valid. For example: 306224133Sdim 307224133Sdim@smallexample 308224133Sdimvoid *ptr; 309234353Sdim/* @r{@dots{}} */ 310234353Sdimptr = &&foo; 311234353Sdim@end smallexample 312239462Sdim 313234353SdimTo use these values, you need to be able to jump to one. This is done 314234353Sdimwith the computed goto statement@footnote{The analogous feature in 315234353SdimFortran is called an assigned goto, but that name seems inappropriate in 316234353SdimC, where one can do more than simply store label addresses in label 317239462Sdimvariables.}, @code{goto *@var{exp};}. For example, 318234353Sdim 319234353Sdim@smallexample 320234353Sdimgoto *ptr; 321234353Sdim@end smallexample 322239462Sdim 323234353Sdim@noindent 324224133SdimAny expression of type @code{void *} is allowed. 325224133Sdim 326224133SdimOne way of using these constants is in initializing a static array that 327239462Sdimwill serve as a jump table: 328224133Sdim 329224133Sdim@smallexample 330224133Sdimstatic void *array[] = @{ &&foo, &&bar, &&hack @}; 331224133Sdim@end smallexample 332224133Sdim 333224133SdimThen you can select a label with indexing, like this: 334224133Sdim 335226633Sdim@smallexample 336239462Sdimgoto *array[i]; 337239462Sdim@end smallexample 338239462Sdim 339239462Sdim@noindent 340239462SdimNote that this does not check whether the subscript is in bounds---array 341239462Sdimindexing in C never does that. 342239462Sdim 343226633SdimSuch an array of label values serves a purpose much like that of the 344226633Sdim@code{switch} statement. The @code{switch} statement is cleaner, so 345226633Sdimuse that rather than an array unless the problem does not fit a 346226633Sdim@code{switch} statement very well. 347239462Sdim 348234353SdimAnother use of label values is in an interpreter for threaded code. 349226633SdimThe labels within the interpreter function can be stored in the 350226633Sdimthreaded code for super-fast dispatching. 351239462Sdim 352234353SdimYou may not use this mechanism to jump to code in a different function. 353226633SdimIf you do that, totally unpredictable things will happen. The best way to 354226633Sdimavoid this is to store the label address only in automatic variables and 355239462Sdimnever pass it as an argument. 356226633Sdim 357226633SdimAn alternate way to write the above example is 358226633Sdim 359226633Sdim@smallexample 360226633Sdimstatic const int array[] = @{ &&foo - &&foo, &&bar - &&foo, 361226633Sdim &&hack - &&foo @}; 362226633Sdimgoto *(&&foo + array[i]); 363226633Sdim@end smallexample 364226633Sdim 365226633Sdim@noindent 366226633SdimThis is more friendly to code living in shared libraries, as it reduces 367226633Sdimthe number of dynamic relocations that are needed, and by consequence, 368226633Sdimallows the data to be read-only. 369226633Sdim 370239462Sdim@node Nested Functions 371239462Sdim@section Nested Functions 372239462Sdim@cindex nested functions 373239462Sdim@cindex downward funargs 374239462Sdim@cindex thunks 375239462Sdim 376239462SdimA @dfn{nested function} is a function defined inside another function. 377239462Sdim(Nested functions are not supported for GNU C++.) The nested function's 378224133Sdimname is local to the block where it is defined. For example, here we 379234353Sdimdefine a nested function named @code{square}, and call it twice: 380239462Sdim 381239462Sdim@smallexample 382239462Sdim@group 383239462Sdimfoo (double a, double b) 384239462Sdim@{ 385239462Sdim double square (double z) @{ return z * z; @} 386239462Sdim 387239462Sdim return square (a) + square (b); 388239462Sdim@} 389239462Sdim@end group 390239462Sdim@end smallexample 391239462Sdim 392239462SdimThe nested function can access all the variables of the containing 393239462Sdimfunction that are visible at the point of its definition. This is 394239462Sdimcalled @dfn{lexical scoping}. For example, here we show a nested 395239462Sdimfunction which uses an inherited variable named @code{offset}: 396239462Sdim 397239462Sdim@smallexample 398239462Sdim@group 399239462Sdimbar (int *array, int offset, int size) 400239462Sdim@{ 401239462Sdim int access (int *array, int index) 402239462Sdim @{ return array[index + offset]; @} 403239462Sdim int i; 404239462Sdim /* @r{@dots{}} */ 405239462Sdim for (i = 0; i < size; i++) 406239462Sdim /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */ 407239462Sdim@} 408239462Sdim@end group 409239462Sdim@end smallexample 410239462Sdim 411239462SdimNested function definitions are permitted within functions in the places 412239462Sdimwhere variable definitions are allowed; that is, in any block, mixed 413239462Sdimwith the other declarations and statements in the block. 414239462Sdim 415239462SdimIt is possible to call the nested function from outside the scope of its 416239462Sdimname by storing its address or passing the address to another function: 417239462Sdim 418239462Sdim@smallexample 419239462Sdimhack (int *array, int size) 420239462Sdim@{ 421239462Sdim void store (int index, int value) 422239462Sdim @{ array[index] = value; @} 423239462Sdim 424239462Sdim intermediate (store, size); 425239462Sdim@} 426239462Sdim@end smallexample 427239462Sdim 428239462SdimHere, the function @code{intermediate} receives the address of 429239462Sdim@code{store} as an argument. If @code{intermediate} calls @code{store}, 430239462Sdimthe arguments given to @code{store} are used to store into @code{array}. 431239462SdimBut this technique works only so long as the containing function 432239462Sdim(@code{hack}, in this example) does not exit. 433239462Sdim 434239462SdimIf you try to call the nested function through its address after the 435239462Sdimcontaining function has exited, all hell will break loose. If you try 436239462Sdimto call it after a containing scope level has exited, and if it refers 437239462Sdimto some of the variables that are no longer in scope, you may be lucky, 438239462Sdimbut it's not wise to take the risk. If, however, the nested function 439239462Sdimdoes not refer to anything that has gone out of scope, you should be 440239462Sdimsafe. 441239462Sdim 442239462SdimGCC implements taking the address of a nested function using a technique 443239462Sdimcalled @dfn{trampolines}. A paper describing them is available as 444239462Sdim 445239462Sdim@noindent 446239462Sdim@uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}. 447239462Sdim 448239462SdimA nested function can jump to a label inherited from a containing 449239462Sdimfunction, provided the label was explicitly declared in the containing 450239462Sdimfunction (@pxref{Local Labels}). Such a jump returns instantly to the 451239462Sdimcontaining function, exiting the nested function which did the 452239462Sdim@code{goto} and any intermediate functions as well. Here is an example: 453239462Sdim 454239462Sdim@smallexample 455239462Sdim@group 456239462Sdimbar (int *array, int offset, int size) 457239462Sdim@{ 458239462Sdim __label__ failure; 459239462Sdim int access (int *array, int index) 460239462Sdim @{ 461239462Sdim if (index > size) 462239462Sdim goto failure; 463239462Sdim return array[index + offset]; 464239462Sdim @} 465239462Sdim int i; 466239462Sdim /* @r{@dots{}} */ 467239462Sdim for (i = 0; i < size; i++) 468239462Sdim /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */ 469239462Sdim /* @r{@dots{}} */ 470239462Sdim return 0; 471239462Sdim 472239462Sdim /* @r{Control comes here from @code{access} 473239462Sdim if it detects an error.} */ 474239462Sdim failure: 475239462Sdim return -1; 476239462Sdim@} 477239462Sdim@end group 478239462Sdim@end smallexample 479239462Sdim 480239462SdimA nested function always has no linkage. Declaring one with 481239462Sdim@code{extern} or @code{static} is erroneous. If you need to declare the nested function 482239462Sdimbefore its definition, use @code{auto} (which is otherwise meaningless 483239462Sdimfor function declarations). 484239462Sdim 485239462Sdim@smallexample 486239462Sdimbar (int *array, int offset, int size) 487239462Sdim@{ 488239462Sdim __label__ failure; 489239462Sdim auto int access (int *, int); 490239462Sdim /* @r{@dots{}} */ 491239462Sdim int access (int *array, int index) 492239462Sdim @{ 493239462Sdim if (index > size) 494224133Sdim goto failure; 495224133Sdim return array[index + offset]; 496224133Sdim @} 497 /* @r{@dots{}} */ 498@} 499@end smallexample 500 501@node Constructing Calls 502@section Constructing Function Calls 503@cindex constructing calls 504@cindex forwarding calls 505 506Using the built-in functions described below, you can record 507the arguments a function received, and call another function 508with the same arguments, without knowing the number or types 509of the arguments. 510 511You can also record the return value of that function call, 512and later return that value, without knowing what data type 513the function tried to return (as long as your caller expects 514that data type). 515 516However, these built-in functions may interact badly with some 517sophisticated features or other extensions of the language. It 518is, therefore, not recommended to use them outside very simple 519functions acting as mere forwarders for their arguments. 520 521@deftypefn {Built-in Function} {void *} __builtin_apply_args () 522This built-in function returns a pointer to data 523describing how to perform a call with the same arguments as were passed 524to the current function. 525 526The function saves the arg pointer register, structure value address, 527and all registers that might be used to pass arguments to a function 528into a block of memory allocated on the stack. Then it returns the 529address of that block. 530@end deftypefn 531 532@deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size}) 533This built-in function invokes @var{function} 534with a copy of the parameters described by @var{arguments} 535and @var{size}. 536 537The value of @var{arguments} should be the value returned by 538@code{__builtin_apply_args}. The argument @var{size} specifies the size 539of the stack argument data, in bytes. 540 541This function returns a pointer to data describing 542how to return whatever value was returned by @var{function}. The data 543is saved in a block of memory allocated on the stack. 544 545It is not always simple to compute the proper value for @var{size}. The 546value is used by @code{__builtin_apply} to compute the amount of data 547that should be pushed on the stack and copied from the incoming argument 548area. 549@end deftypefn 550 551@deftypefn {Built-in Function} {void} __builtin_return (void *@var{result}) 552This built-in function returns the value described by @var{result} from 553the containing function. You should specify, for @var{result}, a value 554returned by @code{__builtin_apply}. 555@end deftypefn 556 557@node Typeof 558@section Referring to a Type with @code{typeof} 559@findex typeof 560@findex sizeof 561@cindex macros, types of arguments 562 563Another way to refer to the type of an expression is with @code{typeof}. 564The syntax of using of this keyword looks like @code{sizeof}, but the 565construct acts semantically like a type name defined with @code{typedef}. 566 567There are two ways of writing the argument to @code{typeof}: with an 568expression or with a type. Here is an example with an expression: 569 570@smallexample 571typeof (x[0](1)) 572@end smallexample 573 574@noindent 575This assumes that @code{x} is an array of pointers to functions; 576the type described is that of the values of the functions. 577 578Here is an example with a typename as the argument: 579 580@smallexample 581typeof (int *) 582@end smallexample 583 584@noindent 585Here the type described is that of pointers to @code{int}. 586 587If you are writing a header file that must work when included in ISO C 588programs, write @code{__typeof__} instead of @code{typeof}. 589@xref{Alternate Keywords}. 590 591A @code{typeof}-construct can be used anywhere a typedef name could be 592used. For example, you can use it in a declaration, in a cast, or inside 593of @code{sizeof} or @code{typeof}. 594 595@code{typeof} is often useful in conjunction with the 596statements-within-expressions feature. Here is how the two together can 597be used to define a safe ``maximum'' macro that operates on any 598arithmetic type and evaluates each of its arguments exactly once: 599 600@smallexample 601#define max(a,b) \ 602 (@{ typeof (a) _a = (a); \ 603 typeof (b) _b = (b); \ 604 _a > _b ? _a : _b; @}) 605@end smallexample 606 607@cindex underscores in variables in macros 608@cindex @samp{_} in variables in macros 609@cindex local variables in macros 610@cindex variables, local, in macros 611@cindex macros, local variables in 612 613The reason for using names that start with underscores for the local 614variables is to avoid conflicts with variable names that occur within the 615expressions that are substituted for @code{a} and @code{b}. Eventually we 616hope to design a new form of declaration syntax that allows you to declare 617variables whose scopes start only after their initializers; this will be a 618more reliable way to prevent such conflicts. 619 620@noindent 621Some more examples of the use of @code{typeof}: 622 623@itemize @bullet 624@item 625This declares @code{y} with the type of what @code{x} points to. 626 627@smallexample 628typeof (*x) y; 629@end smallexample 630 631@item 632This declares @code{y} as an array of such values. 633 634@smallexample 635typeof (*x) y[4]; 636@end smallexample 637 638@item 639This declares @code{y} as an array of pointers to characters: 640 641@smallexample 642typeof (typeof (char *)[4]) y; 643@end smallexample 644 645@noindent 646It is equivalent to the following traditional C declaration: 647 648@smallexample 649char *y[4]; 650@end smallexample 651 652To see the meaning of the declaration using @code{typeof}, and why it 653might be a useful way to write, rewrite it with these macros: 654 655@smallexample 656#define pointer(T) typeof(T *) 657#define array(T, N) typeof(T [N]) 658@end smallexample 659 660@noindent 661Now the declaration can be rewritten this way: 662 663@smallexample 664array (pointer (char), 4) y; 665@end smallexample 666 667@noindent 668Thus, @code{array (pointer (char), 4)} is the type of arrays of 4 669pointers to @code{char}. 670@end itemize 671 672@emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported 673a more limited extension which permitted one to write 674 675@smallexample 676typedef @var{T} = @var{expr}; 677@end smallexample 678 679@noindent 680with the effect of declaring @var{T} to have the type of the expression 681@var{expr}. This extension does not work with GCC 3 (versions between 6823.0 and 3.2 will crash; 3.2.1 and later give an error). Code which 683relies on it should be rewritten to use @code{typeof}: 684 685@smallexample 686typedef typeof(@var{expr}) @var{T}; 687@end smallexample 688 689@noindent 690This will work with all versions of GCC@. 691 692@node Conditionals 693@section Conditionals with Omitted Operands 694@cindex conditional expressions, extensions 695@cindex omitted middle-operands 696@cindex middle-operands, omitted 697@cindex extensions, @code{?:} 698@cindex @code{?:} extensions 699 700The middle operand in a conditional expression may be omitted. Then 701if the first operand is nonzero, its value is the value of the conditional 702expression. 703 704Therefore, the expression 705 706@smallexample 707x ? : y 708@end smallexample 709 710@noindent 711has the value of @code{x} if that is nonzero; otherwise, the value of 712@code{y}. 713 714This example is perfectly equivalent to 715 716@smallexample 717x ? x : y 718@end smallexample 719 720@cindex side effect in ?: 721@cindex ?: side effect 722@noindent 723In this simple case, the ability to omit the middle operand is not 724especially useful. When it becomes useful is when the first operand does, 725or may (if it is a macro argument), contain a side effect. Then repeating 726the operand in the middle would perform the side effect twice. Omitting 727the middle operand uses the value already computed without the undesirable 728effects of recomputing it. 729 730@node Long Long 731@section Double-Word Integers 732@cindex @code{long long} data types 733@cindex double-word arithmetic 734@cindex multiprecision arithmetic 735@cindex @code{LL} integer suffix 736@cindex @code{ULL} integer suffix 737 738ISO C99 supports data types for integers that are at least 64 bits wide, 739and as an extension GCC supports them in C89 mode and in C++. 740Simply write @code{long long int} for a signed integer, or 741@code{unsigned long long int} for an unsigned integer. To make an 742integer constant of type @code{long long int}, add the suffix @samp{LL} 743to the integer. To make an integer constant of type @code{unsigned long 744long int}, add the suffix @samp{ULL} to the integer. 745 746You can use these types in arithmetic like any other integer types. 747Addition, subtraction, and bitwise boolean operations on these types 748are open-coded on all types of machines. Multiplication is open-coded 749if the machine supports fullword-to-doubleword a widening multiply 750instruction. Division and shifts are open-coded only on machines that 751provide special support. The operations that are not open-coded use 752special library routines that come with GCC@. 753 754There may be pitfalls when you use @code{long long} types for function 755arguments, unless you declare function prototypes. If a function 756expects type @code{int} for its argument, and you pass a value of type 757@code{long long int}, confusion will result because the caller and the 758subroutine will disagree about the number of bytes for the argument. 759Likewise, if the function expects @code{long long int} and you pass 760@code{int}. The best way to avoid such problems is to use prototypes. 761 762@node Complex 763@section Complex Numbers 764@cindex complex numbers 765@cindex @code{_Complex} keyword 766@cindex @code{__complex__} keyword 767 768ISO C99 supports complex floating data types, and as an extension GCC 769supports them in C89 mode and in C++, and supports complex integer data 770types which are not part of ISO C99. You can declare complex types 771using the keyword @code{_Complex}. As an extension, the older GNU 772keyword @code{__complex__} is also supported. 773 774For example, @samp{_Complex double x;} declares @code{x} as a 775variable whose real part and imaginary part are both of type 776@code{double}. @samp{_Complex short int y;} declares @code{y} to 777have real and imaginary parts of type @code{short int}; this is not 778likely to be useful, but it shows that the set of complex types is 779complete. 780 781To write a constant with a complex data type, use the suffix @samp{i} or 782@samp{j} (either one; they are equivalent). For example, @code{2.5fi} 783has type @code{_Complex float} and @code{3i} has type 784@code{_Complex int}. Such a constant always has a pure imaginary 785value, but you can form any complex value you like by adding one to a 786real constant. This is a GNU extension; if you have an ISO C99 787conforming C library (such as GNU libc), and want to construct complex 788constants of floating type, you should include @code{<complex.h>} and 789use the macros @code{I} or @code{_Complex_I} instead. 790 791@cindex @code{__real__} keyword 792@cindex @code{__imag__} keyword 793To extract the real part of a complex-valued expression @var{exp}, write 794@code{__real__ @var{exp}}. Likewise, use @code{__imag__} to 795extract the imaginary part. This is a GNU extension; for values of 796floating type, you should use the ISO C99 functions @code{crealf}, 797@code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and 798@code{cimagl}, declared in @code{<complex.h>} and also provided as 799built-in functions by GCC@. 800 801@cindex complex conjugation 802The operator @samp{~} performs complex conjugation when used on a value 803with a complex type. This is a GNU extension; for values of 804floating type, you should use the ISO C99 functions @code{conjf}, 805@code{conj} and @code{conjl}, declared in @code{<complex.h>} and also 806provided as built-in functions by GCC@. 807 808GCC can allocate complex automatic variables in a noncontiguous 809fashion; it's even possible for the real part to be in a register while 810the imaginary part is on the stack (or vice-versa). Only the DWARF2 811debug info format can represent this, so use of DWARF2 is recommended. 812If you are using the stabs debug info format, GCC describes a noncontiguous 813complex variable as if it were two separate variables of noncomplex type. 814If the variable's actual name is @code{foo}, the two fictitious 815variables are named @code{foo$real} and @code{foo$imag}. You can 816examine and set these two fictitious variables with your debugger. 817 818@node Decimal Float 819@section Decimal Floating Types 820@cindex decimal floating types 821@cindex @code{_Decimal32} data type 822@cindex @code{_Decimal64} data type 823@cindex @code{_Decimal128} data type 824@cindex @code{df} integer suffix 825@cindex @code{dd} integer suffix 826@cindex @code{dl} integer suffix 827@cindex @code{DF} integer suffix 828@cindex @code{DD} integer suffix 829@cindex @code{DL} integer suffix 830 831As an extension, the GNU C compiler supports decimal floating types as 832defined in the N1176 draft of ISO/IEC WDTR24732. Support for decimal 833floating types in GCC will evolve as the draft technical report changes. 834Calling conventions for any target might also change. Not all targets 835support decimal floating types. 836 837The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and 838@code{_Decimal128}. They use a radix of ten, unlike the floating types 839@code{float}, @code{double}, and @code{long double} whose radix is not 840specified by the C standard but is usually two. 841 842Support for decimal floating types includes the arithmetic operators 843add, subtract, multiply, divide; unary arithmetic operators; 844relational operators; equality operators; and conversions to and from 845integer and other floating types. Use a suffix @samp{df} or 846@samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd} 847or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for 848@code{_Decimal128}. 849 850GCC support of decimal float as specified by the draft technical report 851is incomplete: 852 853@itemize @bullet 854@item 855Translation time data type (TTDT) is not supported. 856 857@item 858Characteristics of decimal floating types are defined in header file 859@file{decfloat.h} rather than @file{float.h}. 860 861@item 862When the value of a decimal floating type cannot be represented in the 863integer type to which it is being converted, the result is undefined 864rather than the result value specified by the draft technical report. 865@end itemize 866 867Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128} 868are supported by the DWARF2 debug information format. 869 870@node Hex Floats 871@section Hex Floats 872@cindex hex floats 873 874ISO C99 supports floating-point numbers written not only in the usual 875decimal notation, such as @code{1.55e1}, but also numbers such as 876@code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC 877supports this in C89 mode (except in some cases when strictly 878conforming) and in C++. In that format the 879@samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are 880mandatory. The exponent is a decimal number that indicates the power of 8812 by which the significant part will be multiplied. Thus @samp{0x1.f} is 882@tex 883$1 {15\over16}$, 884@end tex 885@ifnottex 8861 15/16, 887@end ifnottex 888@samp{p3} multiplies it by 8, and the value of @code{0x1.fp3} 889is the same as @code{1.55e1}. 890 891Unlike for floating-point numbers in the decimal notation the exponent 892is always required in the hexadecimal notation. Otherwise the compiler 893would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This 894could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the 895extension for floating-point constants of type @code{float}. 896 897@node Zero Length 898@section Arrays of Length Zero 899@cindex arrays of length zero 900@cindex zero-length arrays 901@cindex length-zero arrays 902@cindex flexible array members 903 904Zero-length arrays are allowed in GNU C@. They are very useful as the 905last element of a structure which is really a header for a variable-length 906object: 907 908@smallexample 909struct line @{ 910 int length; 911 char contents[0]; 912@}; 913 914struct line *thisline = (struct line *) 915 malloc (sizeof (struct line) + this_length); 916thisline->length = this_length; 917@end smallexample 918 919In ISO C90, you would have to give @code{contents} a length of 1, which 920means either you waste space or complicate the argument to @code{malloc}. 921 922In ISO C99, you would use a @dfn{flexible array member}, which is 923slightly different in syntax and semantics: 924 925@itemize @bullet 926@item 927Flexible array members are written as @code{contents[]} without 928the @code{0}. 929 930@item 931Flexible array members have incomplete type, and so the @code{sizeof} 932operator may not be applied. As a quirk of the original implementation 933of zero-length arrays, @code{sizeof} evaluates to zero. 934 935@item 936Flexible array members may only appear as the last member of a 937@code{struct} that is otherwise non-empty. 938 939@item 940A structure containing a flexible array member, or a union containing 941such a structure (possibly recursively), may not be a member of a 942structure or an element of an array. (However, these uses are 943permitted by GCC as extensions.) 944@end itemize 945 946GCC versions before 3.0 allowed zero-length arrays to be statically 947initialized, as if they were flexible arrays. In addition to those 948cases that were useful, it also allowed initializations in situations 949that would corrupt later data. Non-empty initialization of zero-length 950arrays is now treated like any case where there are more initializer 951elements than the array holds, in that a suitable warning about "excess 952elements in array" is given, and the excess elements (all of them, in 953this case) are ignored. 954 955Instead GCC allows static initialization of flexible array members. 956This is equivalent to defining a new structure containing the original 957structure followed by an array of sufficient size to contain the data. 958I.e.@: in the following, @code{f1} is constructed as if it were declared 959like @code{f2}. 960 961@smallexample 962struct f1 @{ 963 int x; int y[]; 964@} f1 = @{ 1, @{ 2, 3, 4 @} @}; 965 966struct f2 @{ 967 struct f1 f1; int data[3]; 968@} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @}; 969@end smallexample 970 971@noindent 972The convenience of this extension is that @code{f1} has the desired 973type, eliminating the need to consistently refer to @code{f2.f1}. 974 975This has symmetry with normal static arrays, in that an array of 976unknown size is also written with @code{[]}. 977 978Of course, this extension only makes sense if the extra data comes at 979the end of a top-level object, as otherwise we would be overwriting 980data at subsequent offsets. To avoid undue complication and confusion 981with initialization of deeply nested arrays, we simply disallow any 982non-empty initialization except when the structure is the top-level 983object. For example: 984 985@smallexample 986struct foo @{ int x; int y[]; @}; 987struct bar @{ struct foo z; @}; 988 989struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.} 990struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 991struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.} 992struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 993@end smallexample 994 995@node Empty Structures 996@section Structures With No Members 997@cindex empty structures 998@cindex zero-size structures 999 1000GCC permits a C structure to have no members: 1001 1002@smallexample 1003struct empty @{ 1004@}; 1005@end smallexample 1006 1007The structure will have size zero. In C++, empty structures are part 1008of the language. G++ treats empty structures as if they had a single 1009member of type @code{char}. 1010 1011@node Variable Length 1012@section Arrays of Variable Length 1013@cindex variable-length arrays 1014@cindex arrays of variable length 1015@cindex VLAs 1016 1017Variable-length automatic arrays are allowed in ISO C99, and as an 1018extension GCC accepts them in C89 mode and in C++. (However, GCC's 1019implementation of variable-length arrays does not yet conform in detail 1020to the ISO C99 standard.) These arrays are 1021declared like any other automatic arrays, but with a length that is not 1022a constant expression. The storage is allocated at the point of 1023declaration and deallocated when the brace-level is exited. For 1024example: 1025 1026@smallexample 1027FILE * 1028concat_fopen (char *s1, char *s2, char *mode) 1029@{ 1030 char str[strlen (s1) + strlen (s2) + 1]; 1031 strcpy (str, s1); 1032 strcat (str, s2); 1033 return fopen (str, mode); 1034@} 1035@end smallexample 1036 1037@cindex scope of a variable length array 1038@cindex variable-length array scope 1039@cindex deallocating variable length arrays 1040Jumping or breaking out of the scope of the array name deallocates the 1041storage. Jumping into the scope is not allowed; you get an error 1042message for it. 1043 1044@cindex @code{alloca} vs variable-length arrays 1045You can use the function @code{alloca} to get an effect much like 1046variable-length arrays. The function @code{alloca} is available in 1047many other C implementations (but not in all). On the other hand, 1048variable-length arrays are more elegant. 1049 1050There are other differences between these two methods. Space allocated 1051with @code{alloca} exists until the containing @emph{function} returns. 1052The space for a variable-length array is deallocated as soon as the array 1053name's scope ends. (If you use both variable-length arrays and 1054@code{alloca} in the same function, deallocation of a variable-length array 1055will also deallocate anything more recently allocated with @code{alloca}.) 1056 1057You can also use variable-length arrays as arguments to functions: 1058 1059@smallexample 1060struct entry 1061tester (int len, char data[len][len]) 1062@{ 1063 /* @r{@dots{}} */ 1064@} 1065@end smallexample 1066 1067The length of an array is computed once when the storage is allocated 1068and is remembered for the scope of the array in case you access it with 1069@code{sizeof}. 1070 1071If you want to pass the array first and the length afterward, you can 1072use a forward declaration in the parameter list---another GNU extension. 1073 1074@smallexample 1075struct entry 1076tester (int len; char data[len][len], int len) 1077@{ 1078 /* @r{@dots{}} */ 1079@} 1080@end smallexample 1081 1082@cindex parameter forward declaration 1083The @samp{int len} before the semicolon is a @dfn{parameter forward 1084declaration}, and it serves the purpose of making the name @code{len} 1085known when the declaration of @code{data} is parsed. 1086 1087You can write any number of such parameter forward declarations in the 1088parameter list. They can be separated by commas or semicolons, but the 1089last one must end with a semicolon, which is followed by the ``real'' 1090parameter declarations. Each forward declaration must match a ``real'' 1091declaration in parameter name and data type. ISO C99 does not support 1092parameter forward declarations. 1093 1094@node Variadic Macros 1095@section Macros with a Variable Number of Arguments. 1096@cindex variable number of arguments 1097@cindex macro with variable arguments 1098@cindex rest argument (in macro) 1099@cindex variadic macros 1100 1101In the ISO C standard of 1999, a macro can be declared to accept a 1102variable number of arguments much as a function can. The syntax for 1103defining the macro is similar to that of a function. Here is an 1104example: 1105 1106@smallexample 1107#define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) 1108@end smallexample 1109 1110Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of 1111such a macro, it represents the zero or more tokens until the closing 1112parenthesis that ends the invocation, including any commas. This set of 1113tokens replaces the identifier @code{__VA_ARGS__} in the macro body 1114wherever it appears. See the CPP manual for more information. 1115 1116GCC has long supported variadic macros, and used a different syntax that 1117allowed you to give a name to the variable arguments just like any other 1118argument. Here is an example: 1119 1120@smallexample 1121#define debug(format, args...) fprintf (stderr, format, args) 1122@end smallexample 1123 1124This is in all ways equivalent to the ISO C example above, but arguably 1125more readable and descriptive. 1126 1127GNU CPP has two further variadic macro extensions, and permits them to 1128be used with either of the above forms of macro definition. 1129 1130In standard C, you are not allowed to leave the variable argument out 1131entirely; but you are allowed to pass an empty argument. For example, 1132this invocation is invalid in ISO C, because there is no comma after 1133the string: 1134 1135@smallexample 1136debug ("A message") 1137@end smallexample 1138 1139GNU CPP permits you to completely omit the variable arguments in this 1140way. In the above examples, the compiler would complain, though since 1141the expansion of the macro still has the extra comma after the format 1142string. 1143 1144To help solve this problem, CPP behaves specially for variable arguments 1145used with the token paste operator, @samp{##}. If instead you write 1146 1147@smallexample 1148#define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) 1149@end smallexample 1150 1151and if the variable arguments are omitted or empty, the @samp{##} 1152operator causes the preprocessor to remove the comma before it. If you 1153do provide some variable arguments in your macro invocation, GNU CPP 1154does not complain about the paste operation and instead places the 1155variable arguments after the comma. Just like any other pasted macro 1156argument, these arguments are not macro expanded. 1157 1158@node Escaped Newlines 1159@section Slightly Looser Rules for Escaped Newlines 1160@cindex escaped newlines 1161@cindex newlines (escaped) 1162 1163Recently, the preprocessor has relaxed its treatment of escaped 1164newlines. Previously, the newline had to immediately follow a 1165backslash. The current implementation allows whitespace in the form 1166of spaces, horizontal and vertical tabs, and form feeds between the 1167backslash and the subsequent newline. The preprocessor issues a 1168warning, but treats it as a valid escaped newline and combines the two 1169lines to form a single logical line. This works within comments and 1170tokens, as well as between tokens. Comments are @emph{not} treated as 1171whitespace for the purposes of this relaxation, since they have not 1172yet been replaced with spaces. 1173 1174@node Subscripting 1175@section Non-Lvalue Arrays May Have Subscripts 1176@cindex subscripting 1177@cindex arrays, non-lvalue 1178 1179@cindex subscripting and function values 1180In ISO C99, arrays that are not lvalues still decay to pointers, and 1181may be subscripted, although they may not be modified or used after 1182the next sequence point and the unary @samp{&} operator may not be 1183applied to them. As an extension, GCC allows such arrays to be 1184subscripted in C89 mode, though otherwise they do not decay to 1185pointers outside C99 mode. For example, 1186this is valid in GNU C though not valid in C89: 1187 1188@smallexample 1189@group 1190struct foo @{int a[4];@}; 1191 1192struct foo f(); 1193 1194bar (int index) 1195@{ 1196 return f().a[index]; 1197@} 1198@end group 1199@end smallexample 1200 1201@node Pointer Arith 1202@section Arithmetic on @code{void}- and Function-Pointers 1203@cindex void pointers, arithmetic 1204@cindex void, size of pointer to 1205@cindex function pointers, arithmetic 1206@cindex function, size of pointer to 1207 1208In GNU C, addition and subtraction operations are supported on pointers to 1209@code{void} and on pointers to functions. This is done by treating the 1210size of a @code{void} or of a function as 1. 1211 1212A consequence of this is that @code{sizeof} is also allowed on @code{void} 1213and on function types, and returns 1. 1214 1215@opindex Wpointer-arith 1216The option @option{-Wpointer-arith} requests a warning if these extensions 1217are used. 1218 1219@node Initializers 1220@section Non-Constant Initializers 1221@cindex initializers, non-constant 1222@cindex non-constant initializers 1223 1224As in standard C++ and ISO C99, the elements of an aggregate initializer for an 1225automatic variable are not required to be constant expressions in GNU C@. 1226Here is an example of an initializer with run-time varying elements: 1227 1228@smallexample 1229foo (float f, float g) 1230@{ 1231 float beat_freqs[2] = @{ f-g, f+g @}; 1232 /* @r{@dots{}} */ 1233@} 1234@end smallexample 1235 1236@node Compound Literals 1237@section Compound Literals 1238@cindex constructor expressions 1239@cindex initializations in expressions 1240@cindex structures, constructor expression 1241@cindex expressions, constructor 1242@cindex compound literals 1243@c The GNU C name for what C99 calls compound literals was "constructor expressions". 1244 1245ISO C99 supports compound literals. A compound literal looks like 1246a cast containing an initializer. Its value is an object of the 1247type specified in the cast, containing the elements specified in 1248the initializer; it is an lvalue. As an extension, GCC supports 1249compound literals in C89 mode and in C++. 1250 1251Usually, the specified type is a structure. Assume that 1252@code{struct foo} and @code{structure} are declared as shown: 1253 1254@smallexample 1255struct foo @{int a; char b[2];@} structure; 1256@end smallexample 1257 1258@noindent 1259Here is an example of constructing a @code{struct foo} with a compound literal: 1260 1261@smallexample 1262structure = ((struct foo) @{x + y, 'a', 0@}); 1263@end smallexample 1264 1265@noindent 1266This is equivalent to writing the following: 1267 1268@smallexample 1269@{ 1270 struct foo temp = @{x + y, 'a', 0@}; 1271 structure = temp; 1272@} 1273@end smallexample 1274 1275You can also construct an array. If all the elements of the compound literal 1276are (made up of) simple constant expressions, suitable for use in 1277initializers of objects of static storage duration, then the compound 1278literal can be coerced to a pointer to its first element and used in 1279such an initializer, as shown here: 1280 1281@smallexample 1282char **foo = (char *[]) @{ "x", "y", "z" @}; 1283@end smallexample 1284 1285Compound literals for scalar types and union types are is 1286also allowed, but then the compound literal is equivalent 1287to a cast. 1288 1289As a GNU extension, GCC allows initialization of objects with static storage 1290duration by compound literals (which is not possible in ISO C99, because 1291the initializer is not a constant). 1292It is handled as if the object was initialized only with the bracket 1293enclosed list if the types of the compound literal and the object match. 1294The initializer list of the compound literal must be constant. 1295If the object being initialized has array type of unknown size, the size is 1296determined by compound literal size. 1297 1298@smallexample 1299static struct foo x = (struct foo) @{1, 'a', 'b'@}; 1300static int y[] = (int []) @{1, 2, 3@}; 1301static int z[] = (int [3]) @{1@}; 1302@end smallexample 1303 1304@noindent 1305The above lines are equivalent to the following: 1306@smallexample 1307static struct foo x = @{1, 'a', 'b'@}; 1308static int y[] = @{1, 2, 3@}; 1309static int z[] = @{1, 0, 0@}; 1310@end smallexample 1311 1312@node Designated Inits 1313@section Designated Initializers 1314@cindex initializers with labeled elements 1315@cindex labeled elements in initializers 1316@cindex case labels in initializers 1317@cindex designated initializers 1318 1319Standard C89 requires the elements of an initializer to appear in a fixed 1320order, the same as the order of the elements in the array or structure 1321being initialized. 1322 1323In ISO C99 you can give the elements in any order, specifying the array 1324indices or structure field names they apply to, and GNU C allows this as 1325an extension in C89 mode as well. This extension is not 1326implemented in GNU C++. 1327 1328To specify an array index, write 1329@samp{[@var{index}] =} before the element value. For example, 1330 1331@smallexample 1332int a[6] = @{ [4] = 29, [2] = 15 @}; 1333@end smallexample 1334 1335@noindent 1336is equivalent to 1337 1338@smallexample 1339int a[6] = @{ 0, 0, 15, 0, 29, 0 @}; 1340@end smallexample 1341 1342@noindent 1343The index values must be constant expressions, even if the array being 1344initialized is automatic. 1345 1346An alternative syntax for this which has been obsolete since GCC 2.5 but 1347GCC still accepts is to write @samp{[@var{index}]} before the element 1348value, with no @samp{=}. 1349 1350To initialize a range of elements to the same value, write 1351@samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU 1352extension. For example, 1353 1354@smallexample 1355int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @}; 1356@end smallexample 1357 1358@noindent 1359If the value in it has side-effects, the side-effects will happen only once, 1360not for each initialized field by the range initializer. 1361 1362@noindent 1363Note that the length of the array is the highest value specified 1364plus one. 1365 1366In a structure initializer, specify the name of a field to initialize 1367with @samp{.@var{fieldname} =} before the element value. For example, 1368given the following structure, 1369 1370@smallexample 1371struct point @{ int x, y; @}; 1372@end smallexample 1373 1374@noindent 1375the following initialization 1376 1377@smallexample 1378struct point p = @{ .y = yvalue, .x = xvalue @}; 1379@end smallexample 1380 1381@noindent 1382is equivalent to 1383 1384@smallexample 1385struct point p = @{ xvalue, yvalue @}; 1386@end smallexample 1387 1388Another syntax which has the same meaning, obsolete since GCC 2.5, is 1389@samp{@var{fieldname}:}, as shown here: 1390 1391@smallexample 1392struct point p = @{ y: yvalue, x: xvalue @}; 1393@end smallexample 1394 1395@cindex designators 1396The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a 1397@dfn{designator}. You can also use a designator (or the obsolete colon 1398syntax) when initializing a union, to specify which element of the union 1399should be used. For example, 1400 1401@smallexample 1402union foo @{ int i; double d; @}; 1403 1404union foo f = @{ .d = 4 @}; 1405@end smallexample 1406 1407@noindent 1408will convert 4 to a @code{double} to store it in the union using 1409the second element. By contrast, casting 4 to type @code{union foo} 1410would store it into the union as the integer @code{i}, since it is 1411an integer. (@xref{Cast to Union}.) 1412 1413You can combine this technique of naming elements with ordinary C 1414initialization of successive elements. Each initializer element that 1415does not have a designator applies to the next consecutive element of the 1416array or structure. For example, 1417 1418@smallexample 1419int a[6] = @{ [1] = v1, v2, [4] = v4 @}; 1420@end smallexample 1421 1422@noindent 1423is equivalent to 1424 1425@smallexample 1426int a[6] = @{ 0, v1, v2, 0, v4, 0 @}; 1427@end smallexample 1428 1429Labeling the elements of an array initializer is especially useful 1430when the indices are characters or belong to an @code{enum} type. 1431For example: 1432 1433@smallexample 1434int whitespace[256] 1435 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1, 1436 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @}; 1437@end smallexample 1438 1439@cindex designator lists 1440You can also write a series of @samp{.@var{fieldname}} and 1441@samp{[@var{index}]} designators before an @samp{=} to specify a 1442nested subobject to initialize; the list is taken relative to the 1443subobject corresponding to the closest surrounding brace pair. For 1444example, with the @samp{struct point} declaration above: 1445 1446@smallexample 1447struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @}; 1448@end smallexample 1449 1450@noindent 1451If the same field is initialized multiple times, it will have value from 1452the last initialization. If any such overridden initialization has 1453side-effect, it is unspecified whether the side-effect happens or not. 1454Currently, GCC will discard them and issue a warning. 1455 1456@node Case Ranges 1457@section Case Ranges 1458@cindex case ranges 1459@cindex ranges in case statements 1460 1461You can specify a range of consecutive values in a single @code{case} label, 1462like this: 1463 1464@smallexample 1465case @var{low} ... @var{high}: 1466@end smallexample 1467 1468@noindent 1469This has the same effect as the proper number of individual @code{case} 1470labels, one for each integer value from @var{low} to @var{high}, inclusive. 1471 1472This feature is especially useful for ranges of ASCII character codes: 1473 1474@smallexample 1475case 'A' ... 'Z': 1476@end smallexample 1477 1478@strong{Be careful:} Write spaces around the @code{...}, for otherwise 1479it may be parsed wrong when you use it with integer values. For example, 1480write this: 1481 1482@smallexample 1483case 1 ... 5: 1484@end smallexample 1485 1486@noindent 1487rather than this: 1488 1489@smallexample 1490case 1...5: 1491@end smallexample 1492 1493@node Cast to Union 1494@section Cast to a Union Type 1495@cindex cast to a union 1496@cindex union, casting to a 1497 1498A cast to union type is similar to other casts, except that the type 1499specified is a union type. You can specify the type either with 1500@code{union @var{tag}} or with a typedef name. A cast to union is actually 1501a constructor though, not a cast, and hence does not yield an lvalue like 1502normal casts. (@xref{Compound Literals}.) 1503 1504The types that may be cast to the union type are those of the members 1505of the union. Thus, given the following union and variables: 1506 1507@smallexample 1508union foo @{ int i; double d; @}; 1509int x; 1510double y; 1511@end smallexample 1512 1513@noindent 1514both @code{x} and @code{y} can be cast to type @code{union foo}. 1515 1516Using the cast as the right-hand side of an assignment to a variable of 1517union type is equivalent to storing in a member of the union: 1518 1519@smallexample 1520union foo u; 1521/* @r{@dots{}} */ 1522u = (union foo) x @equiv{} u.i = x 1523u = (union foo) y @equiv{} u.d = y 1524@end smallexample 1525 1526You can also use the union cast as a function argument: 1527 1528@smallexample 1529void hack (union foo); 1530/* @r{@dots{}} */ 1531hack ((union foo) x); 1532@end smallexample 1533 1534@node Mixed Declarations 1535@section Mixed Declarations and Code 1536@cindex mixed declarations and code 1537@cindex declarations, mixed with code 1538@cindex code, mixed with declarations 1539 1540ISO C99 and ISO C++ allow declarations and code to be freely mixed 1541within compound statements. As an extension, GCC also allows this in 1542C89 mode. For example, you could do: 1543 1544@smallexample 1545int i; 1546/* @r{@dots{}} */ 1547i++; 1548int j = i + 2; 1549@end smallexample 1550 1551Each identifier is visible from where it is declared until the end of 1552the enclosing block. 1553 1554@node Function Attributes 1555@section Declaring Attributes of Functions 1556@cindex function attributes 1557@cindex declaring attributes of functions 1558@cindex functions that never return 1559@cindex functions that return more than once 1560@cindex functions that have no side effects 1561@cindex functions in arbitrary sections 1562@cindex functions that behave like malloc 1563@cindex @code{volatile} applied to function 1564@cindex @code{const} applied to function 1565@cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments 1566@cindex functions with non-null pointer arguments 1567@cindex functions that are passed arguments in registers on the 386 1568@cindex functions that pop the argument stack on the 386 1569@cindex functions that do not pop the argument stack on the 386 1570 1571In GNU C, you declare certain things about functions called in your program 1572which help the compiler optimize function calls and check your code more 1573carefully. 1574 1575The keyword @code{__attribute__} allows you to specify special 1576attributes when making a declaration. This keyword is followed by an 1577attribute specification inside double parentheses. The following 1578attributes are currently defined for functions on all targets: 1579@code{noreturn}, @code{returns_twice}, @code{noinline}, @code{always_inline}, 1580@code{flatten}, @code{pure}, @code{const}, @code{nothrow}, @code{sentinel}, 1581@code{format}, @code{format_arg}, @code{no_instrument_function}, 1582@code{section}, @code{constructor}, @code{destructor}, @code{used}, 1583@code{unused}, @code{deprecated}, @code{weak}, @code{malloc}, 1584@code{alias}, @code{warn_unused_result}, @code{nonnull}, 1585@code{gnu_inline} and @code{externally_visible}. Several other 1586attributes are defined for functions on particular target systems. Other 1587attributes, including @code{section} are supported for variables declarations 1588(@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}). 1589 1590You may also specify attributes with @samp{__} preceding and following 1591each keyword. This allows you to use them in header files without 1592being concerned about a possible macro of the same name. For example, 1593you may use @code{__noreturn__} instead of @code{noreturn}. 1594 1595@xref{Attribute Syntax}, for details of the exact syntax for using 1596attributes. 1597 1598@table @code 1599@c Keep this table alphabetized by attribute name. Treat _ as space. 1600 1601@item alias ("@var{target}") 1602@cindex @code{alias} attribute 1603The @code{alias} attribute causes the declaration to be emitted as an 1604alias for another symbol, which must be specified. For instance, 1605 1606@smallexample 1607void __f () @{ /* @r{Do something.} */; @} 1608void f () __attribute__ ((weak, alias ("__f"))); 1609@end smallexample 1610 1611defines @samp{f} to be a weak alias for @samp{__f}. In C++, the 1612mangled name for the target must be used. It is an error if @samp{__f} 1613is not defined in the same translation unit. 1614 1615Not all target machines support this attribute. 1616 1617@item always_inline 1618@cindex @code{always_inline} function attribute 1619Generally, functions are not inlined unless optimization is specified. 1620For functions declared inline, this attribute inlines the function even 1621if no optimization level was specified. 1622 1623@item gnu_inline 1624@cindex @code{gnu_inline} function attribute 1625This attribute should be used with a function which is also declared 1626with the @code{inline} keyword. It directs GCC to treat the function 1627as if it were defined in gnu89 mode even when compiling in C99 or 1628gnu99 mode. 1629 1630If the function is declared @code{extern}, then this definition of the 1631function is used only for inlining. In no case is the function 1632compiled as a standalone function, not even if you take its address 1633explicitly. Such an address becomes an external reference, as if you 1634had only declared the function, and had not defined it. This has 1635almost the effect of a macro. The way to use this is to put a 1636function definition in a header file with this attribute, and put 1637another copy of the function, without @code{extern}, in a library 1638file. The definition in the header file will cause most calls to the 1639function to be inlined. If any uses of the function remain, they will 1640refer to the single copy in the library. Note that the two 1641definitions of the functions need not be precisely the same, although 1642if they do not have the same effect your program may behave oddly. 1643 1644If the function is neither @code{extern} nor @code{static}, then the 1645function is compiled as a standalone function, as well as being 1646inlined where possible. 1647 1648This is how GCC traditionally handled functions declared 1649@code{inline}. Since ISO C99 specifies a different semantics for 1650@code{inline}, this function attribute is provided as a transition 1651measure and as a useful feature in its own right. This attribute is 1652available in GCC 4.1.3 and later. It is available if either of the 1653preprocessor macros @code{__GNUC_GNU_INLINE__} or 1654@code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline 1655Function is As Fast As a Macro}. 1656 1657Note that since the first version of GCC to support C99 inline semantics 1658is 4.3, earlier versions of GCC which accept this attribute effectively 1659assume that it is always present, whether or not it is given explicitly. 1660In versions prior to 4.3, the only effect of explicitly including it is 1661to disable warnings about using inline functions in C99 mode. 1662 1663@cindex @code{flatten} function attribute 1664@item flatten 1665Generally, inlining into a function is limited. For a function marked with 1666this attribute, every call inside this function will be inlined, if possible. 1667Whether the function itself is considered for inlining depends on its size and 1668the current inlining parameters. The @code{flatten} attribute only works 1669reliably in unit-at-a-time mode. 1670 1671@item cdecl 1672@cindex functions that do pop the argument stack on the 386 1673@opindex mrtd 1674On the Intel 386, the @code{cdecl} attribute causes the compiler to 1675assume that the calling function will pop off the stack space used to 1676pass arguments. This is 1677useful to override the effects of the @option{-mrtd} switch. 1678 1679@item const 1680@cindex @code{const} function attribute 1681Many functions do not examine any values except their arguments, and 1682have no effects except the return value. Basically this is just slightly 1683more strict class than the @code{pure} attribute below, since function is not 1684allowed to read global memory. 1685 1686@cindex pointer arguments 1687Note that a function that has pointer arguments and examines the data 1688pointed to must @emph{not} be declared @code{const}. Likewise, a 1689function that calls a non-@code{const} function usually must not be 1690@code{const}. It does not make sense for a @code{const} function to 1691return @code{void}. 1692 1693The attribute @code{const} is not implemented in GCC versions earlier 1694than 2.5. An alternative way to declare that a function has no side 1695effects, which works in the current version and in some older versions, 1696is as follows: 1697 1698@smallexample 1699typedef int intfn (); 1700 1701extern const intfn square; 1702@end smallexample 1703 1704This approach does not work in GNU C++ from 2.6.0 on, since the language 1705specifies that the @samp{const} must be attached to the return value. 1706 1707@item constructor 1708@itemx destructor 1709@cindex @code{constructor} function attribute 1710@cindex @code{destructor} function attribute 1711The @code{constructor} attribute causes the function to be called 1712automatically before execution enters @code{main ()}. Similarly, the 1713@code{destructor} attribute causes the function to be called 1714automatically after @code{main ()} has completed or @code{exit ()} has 1715been called. Functions with these attributes are useful for 1716initializing data that will be used implicitly during the execution of 1717the program. 1718 1719These attributes are not currently implemented for Objective-C@. 1720 1721@item deprecated 1722@cindex @code{deprecated} attribute. 1723The @code{deprecated} attribute results in a warning if the function 1724is used anywhere in the source file. This is useful when identifying 1725functions that are expected to be removed in a future version of a 1726program. The warning also includes the location of the declaration 1727of the deprecated function, to enable users to easily find further 1728information about why the function is deprecated, or what they should 1729do instead. Note that the warnings only occurs for uses: 1730 1731@smallexample 1732int old_fn () __attribute__ ((deprecated)); 1733int old_fn (); 1734int (*fn_ptr)() = old_fn; 1735@end smallexample 1736 1737results in a warning on line 3 but not line 2. 1738 1739The @code{deprecated} attribute can also be used for variables and 1740types (@pxref{Variable Attributes}, @pxref{Type Attributes}.) 1741 1742@item dllexport 1743@cindex @code{__declspec(dllexport)} 1744On Microsoft Windows targets and Symbian OS targets the 1745@code{dllexport} attribute causes the compiler to provide a global 1746pointer to a pointer in a DLL, so that it can be referenced with the 1747@code{dllimport} attribute. On Microsoft Windows targets, the pointer 1748name is formed by combining @code{_imp__} and the function or variable 1749name. 1750 1751You can use @code{__declspec(dllexport)} as a synonym for 1752@code{__attribute__ ((dllexport))} for compatibility with other 1753compilers. 1754 1755On systems that support the @code{visibility} attribute, this 1756attribute also implies ``default'' visibility, unless a 1757@code{visibility} attribute is explicitly specified. You should avoid 1758the use of @code{dllexport} with ``hidden'' or ``internal'' 1759visibility; in the future GCC may issue an error for those cases. 1760 1761Currently, the @code{dllexport} attribute is ignored for inlined 1762functions, unless the @option{-fkeep-inline-functions} flag has been 1763used. The attribute is also ignored for undefined symbols. 1764 1765When applied to C++ classes, the attribute marks defined non-inlined 1766member functions and static data members as exports. Static consts 1767initialized in-class are not marked unless they are also defined 1768out-of-class. 1769 1770For Microsoft Windows targets there are alternative methods for 1771including the symbol in the DLL's export table such as using a 1772@file{.def} file with an @code{EXPORTS} section or, with GNU ld, using 1773the @option{--export-all} linker flag. 1774 1775@item dllimport 1776@cindex @code{__declspec(dllimport)} 1777On Microsoft Windows and Symbian OS targets, the @code{dllimport} 1778attribute causes the compiler to reference a function or variable via 1779a global pointer to a pointer that is set up by the DLL exporting the 1780symbol. The attribute implies @code{extern} storage. On Microsoft 1781Windows targets, the pointer name is formed by combining @code{_imp__} 1782and the function or variable name. 1783 1784You can use @code{__declspec(dllimport)} as a synonym for 1785@code{__attribute__ ((dllimport))} for compatibility with other 1786compilers. 1787 1788Currently, the attribute is ignored for inlined functions. If the 1789attribute is applied to a symbol @emph{definition}, an error is reported. 1790If a symbol previously declared @code{dllimport} is later defined, the 1791attribute is ignored in subsequent references, and a warning is emitted. 1792The attribute is also overridden by a subsequent declaration as 1793@code{dllexport}. 1794 1795When applied to C++ classes, the attribute marks non-inlined 1796member functions and static data members as imports. However, the 1797attribute is ignored for virtual methods to allow creation of vtables 1798using thunks. 1799 1800On the SH Symbian OS target the @code{dllimport} attribute also has 1801another affect---it can cause the vtable and run-time type information 1802for a class to be exported. This happens when the class has a 1803dllimport'ed constructor or a non-inline, non-pure virtual function 1804and, for either of those two conditions, the class also has a inline 1805constructor or destructor and has a key function that is defined in 1806the current translation unit. 1807 1808For Microsoft Windows based targets the use of the @code{dllimport} 1809attribute on functions is not necessary, but provides a small 1810performance benefit by eliminating a thunk in the DLL@. The use of the 1811@code{dllimport} attribute on imported variables was required on older 1812versions of the GNU linker, but can now be avoided by passing the 1813@option{--enable-auto-import} switch to the GNU linker. As with 1814functions, using the attribute for a variable eliminates a thunk in 1815the DLL@. 1816 1817One drawback to using this attribute is that a pointer to a function 1818or variable marked as @code{dllimport} cannot be used as a constant 1819address. On Microsoft Windows targets, the attribute can be disabled 1820for functions by setting the @option{-mnop-fun-dllimport} flag. 1821 1822@item eightbit_data 1823@cindex eight bit data on the H8/300, H8/300H, and H8S 1824Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified 1825variable should be placed into the eight bit data section. 1826The compiler will generate more efficient code for certain operations 1827on data in the eight bit data area. Note the eight bit data area is limited to 1828256 bytes of data. 1829 1830You must use GAS and GLD from GNU binutils version 2.7 or later for 1831this attribute to work correctly. 1832 1833@item exception_handler 1834@cindex exception handler functions on the Blackfin processor 1835Use this attribute on the Blackfin to indicate that the specified function 1836is an exception handler. The compiler will generate function entry and 1837exit sequences suitable for use in an exception handler when this 1838attribute is present. 1839 1840@item far 1841@cindex functions which handle memory bank switching 1842On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to 1843use a calling convention that takes care of switching memory banks when 1844entering and leaving a function. This calling convention is also the 1845default when using the @option{-mlong-calls} option. 1846 1847On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions 1848to call and return from a function. 1849 1850On 68HC11 the compiler will generate a sequence of instructions 1851to invoke a board-specific routine to switch the memory bank and call the 1852real function. The board-specific routine simulates a @code{call}. 1853At the end of a function, it will jump to a board-specific routine 1854instead of using @code{rts}. The board-specific return routine simulates 1855the @code{rtc}. 1856 1857@item fastcall 1858@cindex functions that pop the argument stack on the 386 1859On the Intel 386, the @code{fastcall} attribute causes the compiler to 1860pass the first argument (if of integral type) in the register ECX and 1861the second argument (if of integral type) in the register EDX@. Subsequent 1862and other typed arguments are passed on the stack. The called function will 1863pop the arguments off the stack. If the number of arguments is variable all 1864arguments are pushed on the stack. 1865 1866@item format (@var{archetype}, @var{string-index}, @var{first-to-check}) 1867@cindex @code{format} function attribute 1868@opindex Wformat 1869The @code{format} attribute specifies that a function takes @code{printf}, 1870@code{scanf}, @code{strftime} or @code{strfmon} style arguments which 1871should be type-checked against a format string. For example, the 1872declaration: 1873 1874@smallexample 1875extern int 1876my_printf (void *my_object, const char *my_format, ...) 1877 __attribute__ ((format (printf, 2, 3))); 1878@end smallexample 1879 1880@noindent 1881causes the compiler to check the arguments in calls to @code{my_printf} 1882for consistency with the @code{printf} style format string argument 1883@code{my_format}. 1884 1885The parameter @var{archetype} determines how the format string is 1886interpreted, and should be @code{printf}, @code{scanf}, @code{strftime} 1887or @code{strfmon}. (You can also use @code{__printf__}, 1888@code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The 1889parameter @var{string-index} specifies which argument is the format 1890string argument (starting from 1), while @var{first-to-check} is the 1891number of the first argument to check against the format string. For 1892functions where the arguments are not available to be checked (such as 1893@code{vprintf}), specify the third parameter as zero. In this case the 1894compiler only checks the format string for consistency. For 1895@code{strftime} formats, the third parameter is required to be zero. 1896Since non-static C++ methods have an implicit @code{this} argument, the 1897arguments of such methods should be counted from two, not one, when 1898giving values for @var{string-index} and @var{first-to-check}. 1899 1900In the example above, the format string (@code{my_format}) is the second 1901argument of the function @code{my_print}, and the arguments to check 1902start with the third argument, so the correct parameters for the format 1903attribute are 2 and 3. 1904 1905@opindex ffreestanding 1906@opindex fno-builtin 1907The @code{format} attribute allows you to identify your own functions 1908which take format strings as arguments, so that GCC can check the 1909calls to these functions for errors. The compiler always (unless 1910@option{-ffreestanding} or @option{-fno-builtin} is used) checks formats 1911for the standard library functions @code{printf}, @code{fprintf}, 1912@code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime}, 1913@code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such 1914warnings are requested (using @option{-Wformat}), so there is no need to 1915modify the header file @file{stdio.h}. In C99 mode, the functions 1916@code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and 1917@code{vsscanf} are also checked. Except in strictly conforming C 1918standard modes, the X/Open function @code{strfmon} is also checked as 1919are @code{printf_unlocked} and @code{fprintf_unlocked}. 1920@xref{C Dialect Options,,Options Controlling C Dialect}. 1921 1922The target may provide additional types of format checks. 1923@xref{Target Format Checks,,Format Checks Specific to Particular 1924Target Machines}. 1925 1926@item format_arg (@var{string-index}) 1927@cindex @code{format_arg} function attribute 1928@opindex Wformat-nonliteral 1929The @code{format_arg} attribute specifies that a function takes a format 1930string for a @code{printf}, @code{scanf}, @code{strftime} or 1931@code{strfmon} style function and modifies it (for example, to translate 1932it into another language), so the result can be passed to a 1933@code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style 1934function (with the remaining arguments to the format function the same 1935as they would have been for the unmodified string). For example, the 1936declaration: 1937 1938@smallexample 1939extern char * 1940my_dgettext (char *my_domain, const char *my_format) 1941 __attribute__ ((format_arg (2))); 1942@end smallexample 1943 1944@noindent 1945causes the compiler to check the arguments in calls to a @code{printf}, 1946@code{scanf}, @code{strftime} or @code{strfmon} type function, whose 1947format string argument is a call to the @code{my_dgettext} function, for 1948consistency with the format string argument @code{my_format}. If the 1949@code{format_arg} attribute had not been specified, all the compiler 1950could tell in such calls to format functions would be that the format 1951string argument is not constant; this would generate a warning when 1952@option{-Wformat-nonliteral} is used, but the calls could not be checked 1953without the attribute. 1954 1955The parameter @var{string-index} specifies which argument is the format 1956string argument (starting from one). Since non-static C++ methods have 1957an implicit @code{this} argument, the arguments of such methods should 1958be counted from two. 1959 1960The @code{format-arg} attribute allows you to identify your own 1961functions which modify format strings, so that GCC can check the 1962calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} 1963type function whose operands are a call to one of your own function. 1964The compiler always treats @code{gettext}, @code{dgettext}, and 1965@code{dcgettext} in this manner except when strict ISO C support is 1966requested by @option{-ansi} or an appropriate @option{-std} option, or 1967@option{-ffreestanding} or @option{-fno-builtin} 1968is used. @xref{C Dialect Options,,Options 1969Controlling C Dialect}. 1970 1971@item function_vector 1972@cindex calling functions through the function vector on the H8/300 processors 1973Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified 1974function should be called through the function vector. Calling a 1975function through the function vector will reduce code size, however; 1976the function vector has a limited size (maximum 128 entries on the H8/300 1977and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector. 1978 1979You must use GAS and GLD from GNU binutils version 2.7 or later for 1980this attribute to work correctly. 1981 1982@item interrupt 1983@cindex interrupt handler functions 1984Use this attribute on the ARM, AVR, C4x, CRX, M32C, M32R/D, MS1, and Xstormy16 1985ports to indicate that the specified function is an interrupt handler. 1986The compiler will generate function entry and exit sequences suitable 1987for use in an interrupt handler when this attribute is present. 1988 1989Note, interrupt handlers for the Blackfin, m68k, H8/300, H8/300H, H8S, and 1990SH processors can be specified via the @code{interrupt_handler} attribute. 1991 1992Note, on the AVR, interrupts will be enabled inside the function. 1993 1994Note, for the ARM, you can specify the kind of interrupt to be handled by 1995adding an optional parameter to the interrupt attribute like this: 1996 1997@smallexample 1998void f () __attribute__ ((interrupt ("IRQ"))); 1999@end smallexample 2000 2001Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@. 2002 2003@item interrupt_handler 2004@cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors 2005Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to 2006indicate that the specified function is an interrupt handler. The compiler 2007will generate function entry and exit sequences suitable for use in an 2008interrupt handler when this attribute is present. 2009 2010@item kspisusp 2011@cindex User stack pointer in interrupts on the Blackfin 2012When used together with @code{interrupt_handler}, @code{exception_handler} 2013or @code{nmi_handler}, code will be generated to load the stack pointer 2014from the USP register in the function prologue. 2015 2016@item long_call/short_call 2017@cindex indirect calls on ARM 2018This attribute specifies how a particular function is called on 2019ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options}) 2020command line switch and @code{#pragma long_calls} settings. The 2021@code{long_call} attribute indicates that the function might be far 2022away from the call site and require a different (more expensive) 2023calling sequence. The @code{short_call} attribute always places 2024the offset to the function from the call site into the @samp{BL} 2025instruction directly. 2026 2027@item longcall/shortcall 2028@cindex functions called via pointer on the RS/6000 and PowerPC 2029On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute 2030indicates that the function might be far away from the call site and 2031require a different (more expensive) calling sequence. The 2032@code{shortcall} attribute indicates that the function is always close 2033enough for the shorter calling sequence to be used. These attributes 2034override both the @option{-mlongcall} switch and, on the RS/6000 and 2035PowerPC, the @code{#pragma longcall} setting. 2036 2037@xref{RS/6000 and PowerPC Options}, for more information on whether long 2038calls are necessary. 2039 2040@item long_call 2041@cindex indirect calls on MIPS 2042This attribute specifies how a particular function is called on MIPS@. 2043The attribute overrides the @option{-mlong-calls} (@pxref{MIPS Options}) 2044command line switch. This attribute causes the compiler to always call 2045the function by first loading its address into a register, and then using 2046the contents of that register. 2047 2048@item malloc 2049@cindex @code{malloc} attribute 2050The @code{malloc} attribute is used to tell the compiler that a function 2051may be treated as if any non-@code{NULL} pointer it returns cannot 2052alias any other pointer valid when the function returns. 2053This will often improve optimization. 2054Standard functions with this property include @code{malloc} and 2055@code{calloc}. @code{realloc}-like functions have this property as 2056long as the old pointer is never referred to (including comparing it 2057to the new pointer) after the function returns a non-@code{NULL} 2058value. 2059 2060@item model (@var{model-name}) 2061@cindex function addressability on the M32R/D 2062@cindex variable addressability on the IA-64 2063 2064On the M32R/D, use this attribute to set the addressability of an 2065object, and of the code generated for a function. The identifier 2066@var{model-name} is one of @code{small}, @code{medium}, or 2067@code{large}, representing each of the code models. 2068 2069Small model objects live in the lower 16MB of memory (so that their 2070addresses can be loaded with the @code{ld24} instruction), and are 2071callable with the @code{bl} instruction. 2072 2073Medium model objects may live anywhere in the 32-bit address space (the 2074compiler will generate @code{seth/add3} instructions to load their addresses), 2075and are callable with the @code{bl} instruction. 2076 2077Large model objects may live anywhere in the 32-bit address space (the 2078compiler will generate @code{seth/add3} instructions to load their addresses), 2079and may not be reachable with the @code{bl} instruction (the compiler will 2080generate the much slower @code{seth/add3/jl} instruction sequence). 2081 2082On IA-64, use this attribute to set the addressability of an object. 2083At present, the only supported identifier for @var{model-name} is 2084@code{small}, indicating addressability via ``small'' (22-bit) 2085addresses (so that their addresses can be loaded with the @code{addl} 2086instruction). Caveat: such addressing is by definition not position 2087independent and hence this attribute must not be used for objects 2088defined by shared libraries. 2089 2090@item naked 2091@cindex function without a prologue/epilogue code 2092Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate that the 2093specified function does not need prologue/epilogue sequences generated by 2094the compiler. It is up to the programmer to provide these sequences. 2095 2096@item near 2097@cindex functions which do not handle memory bank switching on 68HC11/68HC12 2098On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to 2099use the normal calling convention based on @code{jsr} and @code{rts}. 2100This attribute can be used to cancel the effect of the @option{-mlong-calls} 2101option. 2102 2103@item nesting 2104@cindex Allow nesting in an interrupt handler on the Blackfin processor. 2105Use this attribute together with @code{interrupt_handler}, 2106@code{exception_handler} or @code{nmi_handler} to indicate that the function 2107entry code should enable nested interrupts or exceptions. 2108 2109@item nmi_handler 2110@cindex NMI handler functions on the Blackfin processor 2111Use this attribute on the Blackfin to indicate that the specified function 2112is an NMI handler. The compiler will generate function entry and 2113exit sequences suitable for use in an NMI handler when this 2114attribute is present. 2115 2116@item no_instrument_function 2117@cindex @code{no_instrument_function} function attribute 2118@opindex finstrument-functions 2119If @option{-finstrument-functions} is given, profiling function calls will 2120be generated at entry and exit of most user-compiled functions. 2121Functions with this attribute will not be so instrumented. 2122 2123@item noinline 2124@cindex @code{noinline} function attribute 2125This function attribute prevents a function from being considered for 2126inlining. 2127 2128@item nonnull (@var{arg-index}, @dots{}) 2129@cindex @code{nonnull} function attribute 2130The @code{nonnull} attribute specifies that some function parameters should 2131be non-null pointers. For instance, the declaration: 2132 2133@smallexample 2134extern void * 2135my_memcpy (void *dest, const void *src, size_t len) 2136 __attribute__((nonnull (1, 2))); 2137@end smallexample 2138 2139@noindent 2140causes the compiler to check that, in calls to @code{my_memcpy}, 2141arguments @var{dest} and @var{src} are non-null. If the compiler 2142determines that a null pointer is passed in an argument slot marked 2143as non-null, and the @option{-Wnonnull} option is enabled, a warning 2144is issued. The compiler may also choose to make optimizations based 2145on the knowledge that certain function arguments will not be null. 2146 2147If no argument index list is given to the @code{nonnull} attribute, 2148all pointer arguments are marked as non-null. To illustrate, the 2149following declaration is equivalent to the previous example: 2150 2151@smallexample 2152extern void * 2153my_memcpy (void *dest, const void *src, size_t len) 2154 __attribute__((nonnull)); 2155@end smallexample 2156 2157@item noreturn 2158@cindex @code{noreturn} function attribute 2159A few standard library functions, such as @code{abort} and @code{exit}, 2160cannot return. GCC knows this automatically. Some programs define 2161their own functions that never return. You can declare them 2162@code{noreturn} to tell the compiler this fact. For example, 2163 2164@smallexample 2165@group 2166void fatal () __attribute__ ((noreturn)); 2167 2168void 2169fatal (/* @r{@dots{}} */) 2170@{ 2171 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */ 2172 exit (1); 2173@} 2174@end group 2175@end smallexample 2176 2177The @code{noreturn} keyword tells the compiler to assume that 2178@code{fatal} cannot return. It can then optimize without regard to what 2179would happen if @code{fatal} ever did return. This makes slightly 2180better code. More importantly, it helps avoid spurious warnings of 2181uninitialized variables. 2182 2183The @code{noreturn} keyword does not affect the exceptional path when that 2184applies: a @code{noreturn}-marked function may still return to the caller 2185by throwing an exception or calling @code{longjmp}. 2186 2187Do not assume that registers saved by the calling function are 2188restored before calling the @code{noreturn} function. 2189 2190It does not make sense for a @code{noreturn} function to have a return 2191type other than @code{void}. 2192 2193The attribute @code{noreturn} is not implemented in GCC versions 2194earlier than 2.5. An alternative way to declare that a function does 2195not return, which works in the current version and in some older 2196versions, is as follows: 2197 2198@smallexample 2199typedef void voidfn (); 2200 2201volatile voidfn fatal; 2202@end smallexample 2203 2204This approach does not work in GNU C++. 2205 2206@item nothrow 2207@cindex @code{nothrow} function attribute 2208The @code{nothrow} attribute is used to inform the compiler that a 2209function cannot throw an exception. For example, most functions in 2210the standard C library can be guaranteed not to throw an exception 2211with the notable exceptions of @code{qsort} and @code{bsearch} that 2212take function pointer arguments. The @code{nothrow} attribute is not 2213implemented in GCC versions earlier than 3.3. 2214 2215@item pure 2216@cindex @code{pure} function attribute 2217Many functions have no effects except the return value and their 2218return value depends only on the parameters and/or global variables. 2219Such a function can be subject 2220to common subexpression elimination and loop optimization just as an 2221arithmetic operator would be. These functions should be declared 2222with the attribute @code{pure}. For example, 2223 2224@smallexample 2225int square (int) __attribute__ ((pure)); 2226@end smallexample 2227 2228@noindent 2229says that the hypothetical function @code{square} is safe to call 2230fewer times than the program says. 2231 2232Some of common examples of pure functions are @code{strlen} or @code{memcmp}. 2233Interesting non-pure functions are functions with infinite loops or those 2234depending on volatile memory or other system resource, that may change between 2235two consecutive calls (such as @code{feof} in a multithreading environment). 2236 2237The attribute @code{pure} is not implemented in GCC versions earlier 2238than 2.96. 2239 2240@item regparm (@var{number}) 2241@cindex @code{regparm} attribute 2242@cindex functions that are passed arguments in registers on the 386 2243On the Intel 386, the @code{regparm} attribute causes the compiler to 2244pass arguments number one to @var{number} if they are of integral type 2245in registers EAX, EDX, and ECX instead of on the stack. Functions that 2246take a variable number of arguments will continue to be passed all of their 2247arguments on the stack. 2248 2249Beware that on some ELF systems this attribute is unsuitable for 2250global functions in shared libraries with lazy binding (which is the 2251default). Lazy binding will send the first call via resolving code in 2252the loader, which might assume EAX, EDX and ECX can be clobbered, as 2253per the standard calling conventions. Solaris 8 is affected by this. 2254GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be 2255safe since the loaders there save all registers. (Lazy binding can be 2256disabled with the linker or the loader if desired, to avoid the 2257problem.) 2258 2259@item sseregparm 2260@cindex @code{sseregparm} attribute 2261On the Intel 386 with SSE support, the @code{sseregparm} attribute 2262causes the compiler to pass up to 3 floating point arguments in 2263SSE registers instead of on the stack. Functions that take a 2264variable number of arguments will continue to pass all of their 2265floating point arguments on the stack. 2266 2267@item force_align_arg_pointer 2268@cindex @code{force_align_arg_pointer} attribute 2269On the Intel x86, the @code{force_align_arg_pointer} attribute may be 2270applied to individual function definitions, generating an alternate 2271prologue and epilogue that realigns the runtime stack. This supports 2272mixing legacy codes that run with a 4-byte aligned stack with modern 2273codes that keep a 16-byte stack for SSE compatibility. The alternate 2274prologue and epilogue are slower and bigger than the regular ones, and 2275the alternate prologue requires a scratch register; this lowers the 2276number of registers available if used in conjunction with the 2277@code{regparm} attribute. The @code{force_align_arg_pointer} 2278attribute is incompatible with nested functions; this is considered a 2279hard error. 2280 2281@item returns_twice 2282@cindex @code{returns_twice} attribute 2283The @code{returns_twice} attribute tells the compiler that a function may 2284return more than one time. The compiler will ensure that all registers 2285are dead before calling such a function and will emit a warning about 2286the variables that may be clobbered after the second return from the 2287function. Examples of such functions are @code{setjmp} and @code{vfork}. 2288The @code{longjmp}-like counterpart of such function, if any, might need 2289to be marked with the @code{noreturn} attribute. 2290 2291@item saveall 2292@cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S 2293Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that 2294all registers except the stack pointer should be saved in the prologue 2295regardless of whether they are used or not. 2296 2297@item section ("@var{section-name}") 2298@cindex @code{section} function attribute 2299Normally, the compiler places the code it generates in the @code{text} section. 2300Sometimes, however, you need additional sections, or you need certain 2301particular functions to appear in special sections. The @code{section} 2302attribute specifies that a function lives in a particular section. 2303For example, the declaration: 2304 2305@smallexample 2306extern void foobar (void) __attribute__ ((section ("bar"))); 2307@end smallexample 2308 2309@noindent 2310puts the function @code{foobar} in the @code{bar} section. 2311 2312Some file formats do not support arbitrary sections so the @code{section} 2313attribute is not available on all platforms. 2314If you need to map the entire contents of a module to a particular 2315section, consider using the facilities of the linker instead. 2316 2317@item sentinel 2318@cindex @code{sentinel} function attribute 2319This function attribute ensures that a parameter in a function call is 2320an explicit @code{NULL}. The attribute is only valid on variadic 2321functions. By default, the sentinel is located at position zero, the 2322last parameter of the function call. If an optional integer position 2323argument P is supplied to the attribute, the sentinel must be located at 2324position P counting backwards from the end of the argument list. 2325 2326@smallexample 2327__attribute__ ((sentinel)) 2328is equivalent to 2329__attribute__ ((sentinel(0))) 2330@end smallexample 2331 2332The attribute is automatically set with a position of 0 for the built-in 2333functions @code{execl} and @code{execlp}. The built-in function 2334@code{execle} has the attribute set with a position of 1. 2335 2336A valid @code{NULL} in this context is defined as zero with any pointer 2337type. If your system defines the @code{NULL} macro with an integer type 2338then you need to add an explicit cast. GCC replaces @code{stddef.h} 2339with a copy that redefines NULL appropriately. 2340 2341The warnings for missing or incorrect sentinels are enabled with 2342@option{-Wformat}. 2343 2344@item short_call 2345See long_call/short_call. 2346 2347@item shortcall 2348See longcall/shortcall. 2349 2350@item signal 2351@cindex signal handler functions on the AVR processors 2352Use this attribute on the AVR to indicate that the specified 2353function is a signal handler. The compiler will generate function 2354entry and exit sequences suitable for use in a signal handler when this 2355attribute is present. Interrupts will be disabled inside the function. 2356 2357@item sp_switch 2358Use this attribute on the SH to indicate an @code{interrupt_handler} 2359function should switch to an alternate stack. It expects a string 2360argument that names a global variable holding the address of the 2361alternate stack. 2362 2363@smallexample 2364void *alt_stack; 2365void f () __attribute__ ((interrupt_handler, 2366 sp_switch ("alt_stack"))); 2367@end smallexample 2368 2369@item stdcall 2370@cindex functions that pop the argument stack on the 386 2371On the Intel 386, the @code{stdcall} attribute causes the compiler to 2372assume that the called function will pop off the stack space used to 2373pass arguments, unless it takes a variable number of arguments. 2374 2375@item tiny_data 2376@cindex tiny data section on the H8/300H and H8S 2377Use this attribute on the H8/300H and H8S to indicate that the specified 2378variable should be placed into the tiny data section. 2379The compiler will generate more efficient code for loads and stores 2380on data in the tiny data section. Note the tiny data area is limited to 2381slightly under 32kbytes of data. 2382 2383@item trap_exit 2384Use this attribute on the SH for an @code{interrupt_handler} to return using 2385@code{trapa} instead of @code{rte}. This attribute expects an integer 2386argument specifying the trap number to be used. 2387 2388@item unused 2389@cindex @code{unused} attribute. 2390This attribute, attached to a function, means that the function is meant 2391to be possibly unused. GCC will not produce a warning for this 2392function. 2393 2394@item used 2395@cindex @code{used} attribute. 2396This attribute, attached to a function, means that code must be emitted 2397for the function even if it appears that the function is not referenced. 2398This is useful, for example, when the function is referenced only in 2399inline assembly. 2400 2401@item visibility ("@var{visibility_type}") 2402@cindex @code{visibility} attribute 2403This attribute affects the linkage of the declaration to which it is attached. 2404There are four supported @var{visibility_type} values: default, 2405hidden, protected or internal visibility. 2406 2407@smallexample 2408void __attribute__ ((visibility ("protected"))) 2409f () @{ /* @r{Do something.} */; @} 2410int i __attribute__ ((visibility ("hidden"))); 2411@end smallexample 2412 2413The possible values of @var{visibility_type} correspond to the 2414visibility settings in the ELF gABI. 2415 2416@table @dfn 2417@c keep this list of visibilities in alphabetical order. 2418 2419@item default 2420Default visibility is the normal case for the object file format. 2421This value is available for the visibility attribute to override other 2422options that may change the assumed visibility of entities. 2423 2424On ELF, default visibility means that the declaration is visible to other 2425modules and, in shared libraries, means that the declared entity may be 2426overridden. 2427 2428On Darwin, default visibility means that the declaration is visible to 2429other modules. 2430 2431Default visibility corresponds to ``external linkage'' in the language. 2432 2433@item hidden 2434Hidden visibility indicates that the entity declared will have a new 2435form of linkage, which we'll call ``hidden linkage''. Two 2436declarations of an object with hidden linkage refer to the same object 2437if they are in the same shared object. 2438 2439@item internal 2440Internal visibility is like hidden visibility, but with additional 2441processor specific semantics. Unless otherwise specified by the 2442psABI, GCC defines internal visibility to mean that a function is 2443@emph{never} called from another module. Compare this with hidden 2444functions which, while they cannot be referenced directly by other 2445modules, can be referenced indirectly via function pointers. By 2446indicating that a function cannot be called from outside the module, 2447GCC may for instance omit the load of a PIC register since it is known 2448that the calling function loaded the correct value. 2449 2450@item protected 2451Protected visibility is like default visibility except that it 2452indicates that references within the defining module will bind to the 2453definition in that module. That is, the declared entity cannot be 2454overridden by another module. 2455 2456@end table 2457 2458All visibilities are supported on many, but not all, ELF targets 2459(supported when the assembler supports the @samp{.visibility} 2460pseudo-op). Default visibility is supported everywhere. Hidden 2461visibility is supported on Darwin targets. 2462 2463The visibility attribute should be applied only to declarations which 2464would otherwise have external linkage. The attribute should be applied 2465consistently, so that the same entity should not be declared with 2466different settings of the attribute. 2467 2468In C++, the visibility attribute applies to types as well as functions 2469and objects, because in C++ types have linkage. A class must not have 2470greater visibility than its non-static data member types and bases, 2471and class members default to the visibility of their class. Also, a 2472declaration without explicit visibility is limited to the visibility 2473of its type. 2474 2475In C++, you can mark member functions and static member variables of a 2476class with the visibility attribute. This is useful if if you know a 2477particular method or static member variable should only be used from 2478one shared object; then you can mark it hidden while the rest of the 2479class has default visibility. Care must be taken to avoid breaking 2480the One Definition Rule; for example, it is usually not useful to mark 2481an inline method as hidden without marking the whole class as hidden. 2482 2483A C++ namespace declaration can also have the visibility attribute. 2484This attribute applies only to the particular namespace body, not to 2485other definitions of the same namespace; it is equivalent to using 2486@samp{#pragma GCC visibility} before and after the namespace 2487definition (@pxref{Visibility Pragmas}). 2488 2489In C++, if a template argument has limited visibility, this 2490restriction is implicitly propagated to the template instantiation. 2491Otherwise, template instantiations and specializations default to the 2492visibility of their template. 2493 2494If both the template and enclosing class have explicit visibility, the 2495visibility from the template is used. 2496 2497@item warn_unused_result 2498@cindex @code{warn_unused_result} attribute 2499The @code{warn_unused_result} attribute causes a warning to be emitted 2500if a caller of the function with this attribute does not use its 2501return value. This is useful for functions where not checking 2502the result is either a security problem or always a bug, such as 2503@code{realloc}. 2504 2505@smallexample 2506int fn () __attribute__ ((warn_unused_result)); 2507int foo () 2508@{ 2509 if (fn () < 0) return -1; 2510 fn (); 2511 return 0; 2512@} 2513@end smallexample 2514 2515results in warning on line 5. 2516 2517@item weak 2518@cindex @code{weak} attribute 2519The @code{weak} attribute causes the declaration to be emitted as a weak 2520symbol rather than a global. This is primarily useful in defining 2521library functions which can be overridden in user code, though it can 2522also be used with non-function declarations. Weak symbols are supported 2523for ELF targets, and also for a.out targets when using the GNU assembler 2524and linker. 2525 2526@item weakref 2527@itemx weakref ("@var{target}") 2528@cindex @code{weakref} attribute 2529The @code{weakref} attribute marks a declaration as a weak reference. 2530Without arguments, it should be accompanied by an @code{alias} attribute 2531naming the target symbol. Optionally, the @var{target} may be given as 2532an argument to @code{weakref} itself. In either case, @code{weakref} 2533implicitly marks the declaration as @code{weak}. Without a 2534@var{target}, given as an argument to @code{weakref} or to @code{alias}, 2535@code{weakref} is equivalent to @code{weak}. 2536 2537@smallexample 2538static int x() __attribute__ ((weakref ("y"))); 2539/* is equivalent to... */ 2540static int x() __attribute__ ((weak, weakref, alias ("y"))); 2541/* and to... */ 2542static int x() __attribute__ ((weakref)); 2543static int x() __attribute__ ((alias ("y"))); 2544@end smallexample 2545 2546A weak reference is an alias that does not by itself require a 2547definition to be given for the target symbol. If the target symbol is 2548only referenced through weak references, then the becomes a @code{weak} 2549undefined symbol. If it is directly referenced, however, then such 2550strong references prevail, and a definition will be required for the 2551symbol, not necessarily in the same translation unit. 2552 2553The effect is equivalent to moving all references to the alias to a 2554separate translation unit, renaming the alias to the aliased symbol, 2555declaring it as weak, compiling the two separate translation units and 2556performing a reloadable link on them. 2557 2558At present, a declaration to which @code{weakref} is attached can 2559only be @code{static}. 2560 2561@item externally_visible 2562@cindex @code{externally_visible} attribute. 2563This attribute, attached to a global variable or function nullify 2564effect of @option{-fwhole-program} command line option, so the object 2565remain visible outside the current compilation unit 2566 2567@end table 2568 2569You can specify multiple attributes in a declaration by separating them 2570by commas within the double parentheses or by immediately following an 2571attribute declaration with another attribute declaration. 2572 2573@cindex @code{#pragma}, reason for not using 2574@cindex pragma, reason for not using 2575Some people object to the @code{__attribute__} feature, suggesting that 2576ISO C's @code{#pragma} should be used instead. At the time 2577@code{__attribute__} was designed, there were two reasons for not doing 2578this. 2579 2580@enumerate 2581@item 2582It is impossible to generate @code{#pragma} commands from a macro. 2583 2584@item 2585There is no telling what the same @code{#pragma} might mean in another 2586compiler. 2587@end enumerate 2588 2589These two reasons applied to almost any application that might have been 2590proposed for @code{#pragma}. It was basically a mistake to use 2591@code{#pragma} for @emph{anything}. 2592 2593The ISO C99 standard includes @code{_Pragma}, which now allows pragmas 2594to be generated from macros. In addition, a @code{#pragma GCC} 2595namespace is now in use for GCC-specific pragmas. However, it has been 2596found convenient to use @code{__attribute__} to achieve a natural 2597attachment of attributes to their corresponding declarations, whereas 2598@code{#pragma GCC} is of use for constructs that do not naturally form 2599part of the grammar. @xref{Other Directives,,Miscellaneous 2600Preprocessing Directives, cpp, The GNU C Preprocessor}. 2601 2602@node Attribute Syntax 2603@section Attribute Syntax 2604@cindex attribute syntax 2605 2606This section describes the syntax with which @code{__attribute__} may be 2607used, and the constructs to which attribute specifiers bind, for the C 2608language. Some details may vary for C++ and Objective-C@. Because of 2609infelicities in the grammar for attributes, some forms described here 2610may not be successfully parsed in all cases. 2611 2612There are some problems with the semantics of attributes in C++. For 2613example, there are no manglings for attributes, although they may affect 2614code generation, so problems may arise when attributed types are used in 2615conjunction with templates or overloading. Similarly, @code{typeid} 2616does not distinguish between types with different attributes. Support 2617for attributes in C++ may be restricted in future to attributes on 2618declarations only, but not on nested declarators. 2619 2620@xref{Function Attributes}, for details of the semantics of attributes 2621applying to functions. @xref{Variable Attributes}, for details of the 2622semantics of attributes applying to variables. @xref{Type Attributes}, 2623for details of the semantics of attributes applying to structure, union 2624and enumerated types. 2625 2626An @dfn{attribute specifier} is of the form 2627@code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list} 2628is a possibly empty comma-separated sequence of @dfn{attributes}, where 2629each attribute is one of the following: 2630 2631@itemize @bullet 2632@item 2633Empty. Empty attributes are ignored. 2634 2635@item 2636A word (which may be an identifier such as @code{unused}, or a reserved 2637word such as @code{const}). 2638 2639@item 2640A word, followed by, in parentheses, parameters for the attribute. 2641These parameters take one of the following forms: 2642 2643@itemize @bullet 2644@item 2645An identifier. For example, @code{mode} attributes use this form. 2646 2647@item 2648An identifier followed by a comma and a non-empty comma-separated list 2649of expressions. For example, @code{format} attributes use this form. 2650 2651@item 2652A possibly empty comma-separated list of expressions. For example, 2653@code{format_arg} attributes use this form with the list being a single 2654integer constant expression, and @code{alias} attributes use this form 2655with the list being a single string constant. 2656@end itemize 2657@end itemize 2658 2659An @dfn{attribute specifier list} is a sequence of one or more attribute 2660specifiers, not separated by any other tokens. 2661 2662In GNU C, an attribute specifier list may appear after the colon following a 2663label, other than a @code{case} or @code{default} label. The only 2664attribute it makes sense to use after a label is @code{unused}. This 2665feature is intended for code generated by programs which contains labels 2666that may be unused but which is compiled with @option{-Wall}. It would 2667not normally be appropriate to use in it human-written code, though it 2668could be useful in cases where the code that jumps to the label is 2669contained within an @code{#ifdef} conditional. GNU C++ does not permit 2670such placement of attribute lists, as it is permissible for a 2671declaration, which could begin with an attribute list, to be labelled in 2672C++. Declarations cannot be labelled in C90 or C99, so the ambiguity 2673does not arise there. 2674 2675An attribute specifier list may appear as part of a @code{struct}, 2676@code{union} or @code{enum} specifier. It may go either immediately 2677after the @code{struct}, @code{union} or @code{enum} keyword, or after 2678the closing brace. The former syntax is preferred. 2679Where attribute specifiers follow the closing brace, they are considered 2680to relate to the structure, union or enumerated type defined, not to any 2681enclosing declaration the type specifier appears in, and the type 2682defined is not complete until after the attribute specifiers. 2683@c Otherwise, there would be the following problems: a shift/reduce 2684@c conflict between attributes binding the struct/union/enum and 2685@c binding to the list of specifiers/qualifiers; and "aligned" 2686@c attributes could use sizeof for the structure, but the size could be 2687@c changed later by "packed" attributes. 2688 2689Otherwise, an attribute specifier appears as part of a declaration, 2690counting declarations of unnamed parameters and type names, and relates 2691to that declaration (which may be nested in another declaration, for 2692example in the case of a parameter declaration), or to a particular declarator 2693within a declaration. Where an 2694attribute specifier is applied to a parameter declared as a function or 2695an array, it should apply to the function or array rather than the 2696pointer to which the parameter is implicitly converted, but this is not 2697yet correctly implemented. 2698 2699Any list of specifiers and qualifiers at the start of a declaration may 2700contain attribute specifiers, whether or not such a list may in that 2701context contain storage class specifiers. (Some attributes, however, 2702are essentially in the nature of storage class specifiers, and only make 2703sense where storage class specifiers may be used; for example, 2704@code{section}.) There is one necessary limitation to this syntax: the 2705first old-style parameter declaration in a function definition cannot 2706begin with an attribute specifier, because such an attribute applies to 2707the function instead by syntax described below (which, however, is not 2708yet implemented in this case). In some other cases, attribute 2709specifiers are permitted by this grammar but not yet supported by the 2710compiler. All attribute specifiers in this place relate to the 2711declaration as a whole. In the obsolescent usage where a type of 2712@code{int} is implied by the absence of type specifiers, such a list of 2713specifiers and qualifiers may be an attribute specifier list with no 2714other specifiers or qualifiers. 2715 2716At present, the first parameter in a function prototype must have some 2717type specifier which is not an attribute specifier; this resolves an 2718ambiguity in the interpretation of @code{void f(int 2719(__attribute__((foo)) x))}, but is subject to change. At present, if 2720the parentheses of a function declarator contain only attributes then 2721those attributes are ignored, rather than yielding an error or warning 2722or implying a single parameter of type int, but this is subject to 2723change. 2724 2725An attribute specifier list may appear immediately before a declarator 2726(other than the first) in a comma-separated list of declarators in a 2727declaration of more than one identifier using a single list of 2728specifiers and qualifiers. Such attribute specifiers apply 2729only to the identifier before whose declarator they appear. For 2730example, in 2731 2732@smallexample 2733__attribute__((noreturn)) void d0 (void), 2734 __attribute__((format(printf, 1, 2))) d1 (const char *, ...), 2735 d2 (void) 2736@end smallexample 2737 2738@noindent 2739the @code{noreturn} attribute applies to all the functions 2740declared; the @code{format} attribute only applies to @code{d1}. 2741 2742An attribute specifier list may appear immediately before the comma, 2743@code{=} or semicolon terminating the declaration of an identifier other 2744than a function definition. At present, such attribute specifiers apply 2745to the declared object or function, but in future they may attach to the 2746outermost adjacent declarator. In simple cases there is no difference, 2747but, for example, in 2748 2749@smallexample 2750void (****f)(void) __attribute__((noreturn)); 2751@end smallexample 2752 2753@noindent 2754at present the @code{noreturn} attribute applies to @code{f}, which 2755causes a warning since @code{f} is not a function, but in future it may 2756apply to the function @code{****f}. The precise semantics of what 2757attributes in such cases will apply to are not yet specified. Where an 2758assembler name for an object or function is specified (@pxref{Asm 2759Labels}), at present the attribute must follow the @code{asm} 2760specification; in future, attributes before the @code{asm} specification 2761may apply to the adjacent declarator, and those after it to the declared 2762object or function. 2763 2764An attribute specifier list may, in future, be permitted to appear after 2765the declarator in a function definition (before any old-style parameter 2766declarations or the function body). 2767 2768Attribute specifiers may be mixed with type qualifiers appearing inside 2769the @code{[]} of a parameter array declarator, in the C99 construct by 2770which such qualifiers are applied to the pointer to which the array is 2771implicitly converted. Such attribute specifiers apply to the pointer, 2772not to the array, but at present this is not implemented and they are 2773ignored. 2774 2775An attribute specifier list may appear at the start of a nested 2776declarator. At present, there are some limitations in this usage: the 2777attributes correctly apply to the declarator, but for most individual 2778attributes the semantics this implies are not implemented. 2779When attribute specifiers follow the @code{*} of a pointer 2780declarator, they may be mixed with any type qualifiers present. 2781The following describes the formal semantics of this syntax. It will make the 2782most sense if you are familiar with the formal specification of 2783declarators in the ISO C standard. 2784 2785Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T 2786D1}, where @code{T} contains declaration specifiers that specify a type 2787@var{Type} (such as @code{int}) and @code{D1} is a declarator that 2788contains an identifier @var{ident}. The type specified for @var{ident} 2789for derived declarators whose type does not include an attribute 2790specifier is as in the ISO C standard. 2791 2792If @code{D1} has the form @code{( @var{attribute-specifier-list} D )}, 2793and the declaration @code{T D} specifies the type 2794``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 2795@code{T D1} specifies the type ``@var{derived-declarator-type-list} 2796@var{attribute-specifier-list} @var{Type}'' for @var{ident}. 2797 2798If @code{D1} has the form @code{* 2799@var{type-qualifier-and-attribute-specifier-list} D}, and the 2800declaration @code{T D} specifies the type 2801``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 2802@code{T D1} specifies the type ``@var{derived-declarator-type-list} 2803@var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for 2804@var{ident}. 2805 2806For example, 2807 2808@smallexample 2809void (__attribute__((noreturn)) ****f) (void); 2810@end smallexample 2811 2812@noindent 2813specifies the type ``pointer to pointer to pointer to pointer to 2814non-returning function returning @code{void}''. As another example, 2815 2816@smallexample 2817char *__attribute__((aligned(8))) *f; 2818@end smallexample 2819 2820@noindent 2821specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''. 2822Note again that this does not work with most attributes; for example, 2823the usage of @samp{aligned} and @samp{noreturn} attributes given above 2824is not yet supported. 2825 2826For compatibility with existing code written for compiler versions that 2827did not implement attributes on nested declarators, some laxity is 2828allowed in the placing of attributes. If an attribute that only applies 2829to types is applied to a declaration, it will be treated as applying to 2830the type of that declaration. If an attribute that only applies to 2831declarations is applied to the type of a declaration, it will be treated 2832as applying to that declaration; and, for compatibility with code 2833placing the attributes immediately before the identifier declared, such 2834an attribute applied to a function return type will be treated as 2835applying to the function type, and such an attribute applied to an array 2836element type will be treated as applying to the array type. If an 2837attribute that only applies to function types is applied to a 2838pointer-to-function type, it will be treated as applying to the pointer 2839target type; if such an attribute is applied to a function return type 2840that is not a pointer-to-function type, it will be treated as applying 2841to the function type. 2842 2843@node Function Prototypes 2844@section Prototypes and Old-Style Function Definitions 2845@cindex function prototype declarations 2846@cindex old-style function definitions 2847@cindex promotion of formal parameters 2848 2849GNU C extends ISO C to allow a function prototype to override a later 2850old-style non-prototype definition. Consider the following example: 2851 2852@smallexample 2853/* @r{Use prototypes unless the compiler is old-fashioned.} */ 2854#ifdef __STDC__ 2855#define P(x) x 2856#else 2857#define P(x) () 2858#endif 2859 2860/* @r{Prototype function declaration.} */ 2861int isroot P((uid_t)); 2862 2863/* @r{Old-style function definition.} */ 2864int 2865isroot (x) /* @r{??? lossage here ???} */ 2866 uid_t x; 2867@{ 2868 return x == 0; 2869@} 2870@end smallexample 2871 2872Suppose the type @code{uid_t} happens to be @code{short}. ISO C does 2873not allow this example, because subword arguments in old-style 2874non-prototype definitions are promoted. Therefore in this example the 2875function definition's argument is really an @code{int}, which does not 2876match the prototype argument type of @code{short}. 2877 2878This restriction of ISO C makes it hard to write code that is portable 2879to traditional C compilers, because the programmer does not know 2880whether the @code{uid_t} type is @code{short}, @code{int}, or 2881@code{long}. Therefore, in cases like these GNU C allows a prototype 2882to override a later old-style definition. More precisely, in GNU C, a 2883function prototype argument type overrides the argument type specified 2884by a later old-style definition if the former type is the same as the 2885latter type before promotion. Thus in GNU C the above example is 2886equivalent to the following: 2887 2888@smallexample 2889int isroot (uid_t); 2890 2891int 2892isroot (uid_t x) 2893@{ 2894 return x == 0; 2895@} 2896@end smallexample 2897 2898@noindent 2899GNU C++ does not support old-style function definitions, so this 2900extension is irrelevant. 2901 2902@node C++ Comments 2903@section C++ Style Comments 2904@cindex // 2905@cindex C++ comments 2906@cindex comments, C++ style 2907 2908In GNU C, you may use C++ style comments, which start with @samp{//} and 2909continue until the end of the line. Many other C implementations allow 2910such comments, and they are included in the 1999 C standard. However, 2911C++ style comments are not recognized if you specify an @option{-std} 2912option specifying a version of ISO C before C99, or @option{-ansi} 2913(equivalent to @option{-std=c89}). 2914 2915@node Dollar Signs 2916@section Dollar Signs in Identifier Names 2917@cindex $ 2918@cindex dollar signs in identifier names 2919@cindex identifier names, dollar signs in 2920 2921In GNU C, you may normally use dollar signs in identifier names. 2922This is because many traditional C implementations allow such identifiers. 2923However, dollar signs in identifiers are not supported on a few target 2924machines, typically because the target assembler does not allow them. 2925 2926@node Character Escapes 2927@section The Character @key{ESC} in Constants 2928 2929You can use the sequence @samp{\e} in a string or character constant to 2930stand for the ASCII character @key{ESC}. 2931 2932@node Alignment 2933@section Inquiring on Alignment of Types or Variables 2934@cindex alignment 2935@cindex type alignment 2936@cindex variable alignment 2937 2938The keyword @code{__alignof__} allows you to inquire about how an object 2939is aligned, or the minimum alignment usually required by a type. Its 2940syntax is just like @code{sizeof}. 2941 2942For example, if the target machine requires a @code{double} value to be 2943aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8. 2944This is true on many RISC machines. On more traditional machine 2945designs, @code{__alignof__ (double)} is 4 or even 2. 2946 2947Some machines never actually require alignment; they allow reference to any 2948data type even at an odd address. For these machines, @code{__alignof__} 2949reports the @emph{recommended} alignment of a type. 2950 2951If the operand of @code{__alignof__} is an lvalue rather than a type, 2952its value is the required alignment for its type, taking into account 2953any minimum alignment specified with GCC's @code{__attribute__} 2954extension (@pxref{Variable Attributes}). For example, after this 2955declaration: 2956 2957@smallexample 2958struct foo @{ int x; char y; @} foo1; 2959@end smallexample 2960 2961@noindent 2962the value of @code{__alignof__ (foo1.y)} is 1, even though its actual 2963alignment is probably 2 or 4, the same as @code{__alignof__ (int)}. 2964 2965It is an error to ask for the alignment of an incomplete type. 2966 2967@node Variable Attributes 2968@section Specifying Attributes of Variables 2969@cindex attribute of variables 2970@cindex variable attributes 2971 2972The keyword @code{__attribute__} allows you to specify special 2973attributes of variables or structure fields. This keyword is followed 2974by an attribute specification inside double parentheses. Some 2975attributes are currently defined generically for variables. 2976Other attributes are defined for variables on particular target 2977systems. Other attributes are available for functions 2978(@pxref{Function Attributes}) and for types (@pxref{Type Attributes}). 2979Other front ends might define more attributes 2980(@pxref{C++ Extensions,,Extensions to the C++ Language}). 2981 2982You may also specify attributes with @samp{__} preceding and following 2983each keyword. This allows you to use them in header files without 2984being concerned about a possible macro of the same name. For example, 2985you may use @code{__aligned__} instead of @code{aligned}. 2986 2987@xref{Attribute Syntax}, for details of the exact syntax for using 2988attributes. 2989 2990@table @code 2991@cindex @code{aligned} attribute 2992@item aligned (@var{alignment}) 2993This attribute specifies a minimum alignment for the variable or 2994structure field, measured in bytes. For example, the declaration: 2995 2996@smallexample 2997int x __attribute__ ((aligned (16))) = 0; 2998@end smallexample 2999 3000@noindent 3001causes the compiler to allocate the global variable @code{x} on a 300216-byte boundary. On a 68040, this could be used in conjunction with 3003an @code{asm} expression to access the @code{move16} instruction which 3004requires 16-byte aligned operands. 3005 3006You can also specify the alignment of structure fields. For example, to 3007create a double-word aligned @code{int} pair, you could write: 3008 3009@smallexample 3010struct foo @{ int x[2] __attribute__ ((aligned (8))); @}; 3011@end smallexample 3012 3013@noindent 3014This is an alternative to creating a union with a @code{double} member 3015that forces the union to be double-word aligned. 3016 3017As in the preceding examples, you can explicitly specify the alignment 3018(in bytes) that you wish the compiler to use for a given variable or 3019structure field. Alternatively, you can leave out the alignment factor 3020and just ask the compiler to align a variable or field to the maximum 3021useful alignment for the target machine you are compiling for. For 3022example, you could write: 3023 3024@smallexample 3025short array[3] __attribute__ ((aligned)); 3026@end smallexample 3027 3028Whenever you leave out the alignment factor in an @code{aligned} attribute 3029specification, the compiler automatically sets the alignment for the declared 3030variable or field to the largest alignment which is ever used for any data 3031type on the target machine you are compiling for. Doing this can often make 3032copy operations more efficient, because the compiler can use whatever 3033instructions copy the biggest chunks of memory when performing copies to 3034or from the variables or fields that you have aligned this way. 3035 3036The @code{aligned} attribute can only increase the alignment; but you 3037can decrease it by specifying @code{packed} as well. See below. 3038 3039Note that the effectiveness of @code{aligned} attributes may be limited 3040by inherent limitations in your linker. On many systems, the linker is 3041only able to arrange for variables to be aligned up to a certain maximum 3042alignment. (For some linkers, the maximum supported alignment may 3043be very very small.) If your linker is only able to align variables 3044up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} 3045in an @code{__attribute__} will still only provide you with 8 byte 3046alignment. See your linker documentation for further information. 3047 3048@item cleanup (@var{cleanup_function}) 3049@cindex @code{cleanup} attribute 3050The @code{cleanup} attribute runs a function when the variable goes 3051out of scope. This attribute can only be applied to auto function 3052scope variables; it may not be applied to parameters or variables 3053with static storage duration. The function must take one parameter, 3054a pointer to a type compatible with the variable. The return value 3055of the function (if any) is ignored. 3056 3057If @option{-fexceptions} is enabled, then @var{cleanup_function} 3058will be run during the stack unwinding that happens during the 3059processing of the exception. Note that the @code{cleanup} attribute 3060does not allow the exception to be caught, only to perform an action. 3061It is undefined what happens if @var{cleanup_function} does not 3062return normally. 3063 3064@item common 3065@itemx nocommon 3066@cindex @code{common} attribute 3067@cindex @code{nocommon} attribute 3068@opindex fcommon 3069@opindex fno-common 3070The @code{common} attribute requests GCC to place a variable in 3071``common'' storage. The @code{nocommon} attribute requests the 3072opposite---to allocate space for it directly. 3073 3074These attributes override the default chosen by the 3075@option{-fno-common} and @option{-fcommon} flags respectively. 3076 3077@item deprecated 3078@cindex @code{deprecated} attribute 3079The @code{deprecated} attribute results in a warning if the variable 3080is used anywhere in the source file. This is useful when identifying 3081variables that are expected to be removed in a future version of a 3082program. The warning also includes the location of the declaration 3083of the deprecated variable, to enable users to easily find further 3084information about why the variable is deprecated, or what they should 3085do instead. Note that the warning only occurs for uses: 3086 3087@smallexample 3088extern int old_var __attribute__ ((deprecated)); 3089extern int old_var; 3090int new_fn () @{ return old_var; @} 3091@end smallexample 3092 3093results in a warning on line 3 but not line 2. 3094 3095The @code{deprecated} attribute can also be used for functions and 3096types (@pxref{Function Attributes}, @pxref{Type Attributes}.) 3097 3098@item mode (@var{mode}) 3099@cindex @code{mode} attribute 3100This attribute specifies the data type for the declaration---whichever 3101type corresponds to the mode @var{mode}. This in effect lets you 3102request an integer or floating point type according to its width. 3103 3104You may also specify a mode of @samp{byte} or @samp{__byte__} to 3105indicate the mode corresponding to a one-byte integer, @samp{word} or 3106@samp{__word__} for the mode of a one-word integer, and @samp{pointer} 3107or @samp{__pointer__} for the mode used to represent pointers. 3108 3109@item packed 3110@cindex @code{packed} attribute 3111The @code{packed} attribute specifies that a variable or structure field 3112should have the smallest possible alignment---one byte for a variable, 3113and one bit for a field, unless you specify a larger value with the 3114@code{aligned} attribute. 3115 3116Here is a structure in which the field @code{x} is packed, so that it 3117immediately follows @code{a}: 3118 3119@smallexample 3120struct foo 3121@{ 3122 char a; 3123 int x[2] __attribute__ ((packed)); 3124@}; 3125@end smallexample 3126 3127@item section ("@var{section-name}") 3128@cindex @code{section} variable attribute 3129Normally, the compiler places the objects it generates in sections like 3130@code{data} and @code{bss}. Sometimes, however, you need additional sections, 3131or you need certain particular variables to appear in special sections, 3132for example to map to special hardware. The @code{section} 3133attribute specifies that a variable (or function) lives in a particular 3134section. For example, this small program uses several specific section names: 3135 3136@smallexample 3137struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @}; 3138struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @}; 3139char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @}; 3140int init_data __attribute__ ((section ("INITDATA"))) = 0; 3141 3142main() 3143@{ 3144 /* @r{Initialize stack pointer} */ 3145 init_sp (stack + sizeof (stack)); 3146 3147 /* @r{Initialize initialized data} */ 3148 memcpy (&init_data, &data, &edata - &data); 3149 3150 /* @r{Turn on the serial ports} */ 3151 init_duart (&a); 3152 init_duart (&b); 3153@} 3154@end smallexample 3155 3156@noindent 3157Use the @code{section} attribute with an @emph{initialized} definition 3158of a @emph{global} variable, as shown in the example. GCC issues 3159a warning and otherwise ignores the @code{section} attribute in 3160uninitialized variable declarations. 3161 3162You may only use the @code{section} attribute with a fully initialized 3163global definition because of the way linkers work. The linker requires 3164each object be defined once, with the exception that uninitialized 3165variables tentatively go in the @code{common} (or @code{bss}) section 3166and can be multiply ``defined''. You can force a variable to be 3167initialized with the @option{-fno-common} flag or the @code{nocommon} 3168attribute. 3169 3170Some file formats do not support arbitrary sections so the @code{section} 3171attribute is not available on all platforms. 3172If you need to map the entire contents of a module to a particular 3173section, consider using the facilities of the linker instead. 3174 3175@item shared 3176@cindex @code{shared} variable attribute 3177On Microsoft Windows, in addition to putting variable definitions in a named 3178section, the section can also be shared among all running copies of an 3179executable or DLL@. For example, this small program defines shared data 3180by putting it in a named section @code{shared} and marking the section 3181shareable: 3182 3183@smallexample 3184int foo __attribute__((section ("shared"), shared)) = 0; 3185 3186int 3187main() 3188@{ 3189 /* @r{Read and write foo. All running 3190 copies see the same value.} */ 3191 return 0; 3192@} 3193@end smallexample 3194 3195@noindent 3196You may only use the @code{shared} attribute along with @code{section} 3197attribute with a fully initialized global definition because of the way 3198linkers work. See @code{section} attribute for more information. 3199 3200The @code{shared} attribute is only available on Microsoft Windows@. 3201 3202@item tls_model ("@var{tls_model}") 3203@cindex @code{tls_model} attribute 3204The @code{tls_model} attribute sets thread-local storage model 3205(@pxref{Thread-Local}) of a particular @code{__thread} variable, 3206overriding @option{-ftls-model=} command line switch on a per-variable 3207basis. 3208The @var{tls_model} argument should be one of @code{global-dynamic}, 3209@code{local-dynamic}, @code{initial-exec} or @code{local-exec}. 3210 3211Not all targets support this attribute. 3212 3213@item unused 3214This attribute, attached to a variable, means that the variable is meant 3215to be possibly unused. GCC will not produce a warning for this 3216variable. 3217 3218@item used 3219This attribute, attached to a variable, means that the variable must be 3220emitted even if it appears that the variable is not referenced. 3221 3222@item vector_size (@var{bytes}) 3223This attribute specifies the vector size for the variable, measured in 3224bytes. For example, the declaration: 3225 3226@smallexample 3227int foo __attribute__ ((vector_size (16))); 3228@end smallexample 3229 3230@noindent 3231causes the compiler to set the mode for @code{foo}, to be 16 bytes, 3232divided into @code{int} sized units. Assuming a 32-bit int (a vector of 32334 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@. 3234 3235This attribute is only applicable to integral and float scalars, 3236although arrays, pointers, and function return values are allowed in 3237conjunction with this construct. 3238 3239Aggregates with this attribute are invalid, even if they are of the same 3240size as a corresponding scalar. For example, the declaration: 3241 3242@smallexample 3243struct S @{ int a; @}; 3244struct S __attribute__ ((vector_size (16))) foo; 3245@end smallexample 3246 3247@noindent 3248is invalid even if the size of the structure is the same as the size of 3249the @code{int}. 3250 3251@item selectany 3252The @code{selectany} attribute causes an initialized global variable to 3253have link-once semantics. When multiple definitions of the variable are 3254encountered by the linker, the first is selected and the remainder are 3255discarded. Following usage by the Microsoft compiler, the linker is told 3256@emph{not} to warn about size or content differences of the multiple 3257definitions. 3258 3259Although the primary usage of this attribute is for POD types, the 3260attribute can also be applied to global C++ objects that are initialized 3261by a constructor. In this case, the static initialization and destruction 3262code for the object is emitted in each translation defining the object, 3263but the calls to the constructor and destructor are protected by a 3264link-once guard variable. 3265 3266The @code{selectany} attribute is only available on Microsoft Windows 3267targets. You can use @code{__declspec (selectany)} as a synonym for 3268@code{__attribute__ ((selectany))} for compatibility with other 3269compilers. 3270 3271@item weak 3272The @code{weak} attribute is described in @xref{Function Attributes}. 3273 3274@item dllimport 3275The @code{dllimport} attribute is described in @xref{Function Attributes}. 3276 3277@item dllexport 3278The @code{dllexport} attribute is described in @xref{Function Attributes}. 3279 3280@end table 3281 3282@subsection M32R/D Variable Attributes 3283 3284One attribute is currently defined for the M32R/D@. 3285 3286@table @code 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@end table 3300 3301@anchor{i386 Variable Attributes} 3302@subsection i386 Variable Attributes 3303 3304Two attributes are currently defined for i386 configurations: 3305@code{ms_struct} and @code{gcc_struct} 3306 3307@table @code 3308@item ms_struct 3309@itemx gcc_struct 3310@cindex @code{ms_struct} attribute 3311@cindex @code{gcc_struct} attribute 3312 3313If @code{packed} is used on a structure, or if bit-fields are used 3314it may be that the Microsoft ABI packs them differently 3315than GCC would normally pack them. Particularly when moving packed 3316data between functions compiled with GCC and the native Microsoft compiler 3317(either via function call or as data in a file), it may be necessary to access 3318either format. 3319 3320Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86 3321compilers to match the native Microsoft compiler. 3322 3323The Microsoft structure layout algorithm is fairly simple with the exception 3324of the bitfield packing: 3325 3326The padding and alignment of members of structures and whether a bit field 3327can straddle a storage-unit boundary 3328 3329@enumerate 3330@item Structure members are stored sequentially in the order in which they are 3331declared: the first member has the lowest memory address and the last member 3332the highest. 3333 3334@item Every data object has an alignment-requirement. The alignment-requirement 3335for all data except structures, unions, and arrays is either the size of the 3336object or the current packing size (specified with either the aligned attribute 3337or the pack pragma), whichever is less. For structures, unions, and arrays, 3338the alignment-requirement is the largest alignment-requirement of its members. 3339Every object is allocated an offset so that: 3340 3341offset % alignment-requirement == 0 3342 3343@item Adjacent bit fields are packed into the same 1-, 2-, or 4-byte allocation 3344unit if the integral types are the same size and if the next bit field fits 3345into the current allocation unit without crossing the boundary imposed by the 3346common alignment requirements of the bit fields. 3347@end enumerate 3348 3349Handling of zero-length bitfields: 3350 3351MSVC interprets zero-length bitfields in the following ways: 3352 3353@enumerate 3354@item If a zero-length bitfield is inserted between two bitfields that would 3355normally be coalesced, the bitfields will not be coalesced. 3356 3357For example: 3358 3359@smallexample 3360struct 3361 @{ 3362 unsigned long bf_1 : 12; 3363 unsigned long : 0; 3364 unsigned long bf_2 : 12; 3365 @} t1; 3366@end smallexample 3367 3368The size of @code{t1} would be 8 bytes with the zero-length bitfield. If the 3369zero-length bitfield were removed, @code{t1}'s size would be 4 bytes. 3370 3371@item If a zero-length bitfield is inserted after a bitfield, @code{foo}, and the 3372alignment of the zero-length bitfield is greater than the member that follows it, 3373@code{bar}, @code{bar} will be aligned as the type of the zero-length bitfield. 3374 3375For example: 3376 3377@smallexample 3378struct 3379 @{ 3380 char foo : 4; 3381 short : 0; 3382 char bar; 3383 @} t2; 3384 3385struct 3386 @{ 3387 char foo : 4; 3388 short : 0; 3389 double bar; 3390 @} t3; 3391@end smallexample 3392 3393For @code{t2}, @code{bar} will be placed at offset 2, rather than offset 1. 3394Accordingly, the size of @code{t2} will be 4. For @code{t3}, the zero-length 3395bitfield will not affect the alignment of @code{bar} or, as a result, the size 3396of the structure. 3397 3398Taking this into account, it is important to note the following: 3399 3400@enumerate 3401@item If a zero-length bitfield follows a normal bitfield, the type of the 3402zero-length bitfield may affect the alignment of the structure as whole. For 3403example, @code{t2} has a size of 4 bytes, since the zero-length bitfield follows a 3404normal bitfield, and is of type short. 3405 3406@item Even if a zero-length bitfield is not followed by a normal bitfield, it may 3407still affect the alignment of the structure: 3408 3409@smallexample 3410struct 3411 @{ 3412 char foo : 6; 3413 long : 0; 3414 @} t4; 3415@end smallexample 3416 3417Here, @code{t4} will take up 4 bytes. 3418@end enumerate 3419 3420@item Zero-length bitfields following non-bitfield members are ignored: 3421 3422@smallexample 3423struct 3424 @{ 3425 char foo; 3426 long : 0; 3427 char bar; 3428 @} t5; 3429@end smallexample 3430 3431Here, @code{t5} will take up 2 bytes. 3432@end enumerate 3433@end table 3434 3435@subsection PowerPC Variable Attributes 3436 3437Three attributes currently are defined for PowerPC configurations: 3438@code{altivec}, @code{ms_struct} and @code{gcc_struct}. 3439 3440For full documentation of the struct attributes please see the 3441documentation in the @xref{i386 Variable Attributes}, section. 3442 3443For documentation of @code{altivec} attribute please see the 3444documentation in the @xref{PowerPC Type Attributes}, section. 3445 3446@subsection Xstormy16 Variable Attributes 3447 3448One attribute is currently defined for xstormy16 configurations: 3449@code{below100} 3450 3451@table @code 3452@item below100 3453@cindex @code{below100} attribute 3454 3455If a variable has the @code{below100} attribute (@code{BELOW100} is 3456allowed also), GCC will place the variable in the first 0x100 bytes of 3457memory and use special opcodes to access it. Such variables will be 3458placed in either the @code{.bss_below100} section or the 3459@code{.data_below100} section. 3460 3461@end table 3462 3463@node Type Attributes 3464@section Specifying Attributes of Types 3465@cindex attribute of types 3466@cindex type attributes 3467 3468The keyword @code{__attribute__} allows you to specify special 3469attributes of @code{struct} and @code{union} types when you define 3470such types. This keyword is followed by an attribute specification 3471inside double parentheses. Seven attributes are currently defined for 3472types: @code{aligned}, @code{packed}, @code{transparent_union}, 3473@code{unused}, @code{deprecated}, @code{visibility}, and 3474@code{may_alias}. Other attributes are defined for functions 3475(@pxref{Function Attributes}) and for variables (@pxref{Variable 3476Attributes}). 3477 3478You may also specify any one of these attributes with @samp{__} 3479preceding and following its keyword. This allows you to use these 3480attributes in header files without being concerned about a possible 3481macro of the same name. For example, you may use @code{__aligned__} 3482instead of @code{aligned}. 3483 3484You may specify type attributes either in a @code{typedef} declaration 3485or in an enum, struct or union type declaration or definition. 3486 3487For an enum, struct or union type, you may specify attributes either 3488between the enum, struct or union tag and the name of the type, or 3489just past the closing curly brace of the @emph{definition}. The 3490former syntax is preferred. 3491 3492@xref{Attribute Syntax}, for details of the exact syntax for using 3493attributes. 3494 3495@table @code 3496@cindex @code{aligned} attribute 3497@item aligned (@var{alignment}) 3498This attribute specifies a minimum alignment (in bytes) for variables 3499of the specified type. For example, the declarations: 3500 3501@smallexample 3502struct S @{ short f[3]; @} __attribute__ ((aligned (8))); 3503typedef int more_aligned_int __attribute__ ((aligned (8))); 3504@end smallexample 3505 3506@noindent 3507force the compiler to insure (as far as it can) that each variable whose 3508type is @code{struct S} or @code{more_aligned_int} will be allocated and 3509aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all 3510variables of type @code{struct S} aligned to 8-byte boundaries allows 3511the compiler to use the @code{ldd} and @code{std} (doubleword load and 3512store) instructions when copying one variable of type @code{struct S} to 3513another, thus improving run-time efficiency. 3514 3515Note that the alignment of any given @code{struct} or @code{union} type 3516is required by the ISO C standard to be at least a perfect multiple of 3517the lowest common multiple of the alignments of all of the members of 3518the @code{struct} or @code{union} in question. This means that you @emph{can} 3519effectively adjust the alignment of a @code{struct} or @code{union} 3520type by attaching an @code{aligned} attribute to any one of the members 3521of such a type, but the notation illustrated in the example above is a 3522more obvious, intuitive, and readable way to request the compiler to 3523adjust the alignment of an entire @code{struct} or @code{union} type. 3524 3525As in the preceding example, you can explicitly specify the alignment 3526(in bytes) that you wish the compiler to use for a given @code{struct} 3527or @code{union} type. Alternatively, you can leave out the alignment factor 3528and just ask the compiler to align a type to the maximum 3529useful alignment for the target machine you are compiling for. For 3530example, you could write: 3531 3532@smallexample 3533struct S @{ short f[3]; @} __attribute__ ((aligned)); 3534@end smallexample 3535 3536Whenever you leave out the alignment factor in an @code{aligned} 3537attribute specification, the compiler automatically sets the alignment 3538for the type to the largest alignment which is ever used for any data 3539type on the target machine you are compiling for. Doing this can often 3540make copy operations more efficient, because the compiler can use 3541whatever instructions copy the biggest chunks of memory when performing 3542copies to or from the variables which have types that you have aligned 3543this way. 3544 3545In the example above, if the size of each @code{short} is 2 bytes, then 3546the size of the entire @code{struct S} type is 6 bytes. The smallest 3547power of two which is greater than or equal to that is 8, so the 3548compiler sets the alignment for the entire @code{struct S} type to 8 3549bytes. 3550 3551Note that although you can ask the compiler to select a time-efficient 3552alignment for a given type and then declare only individual stand-alone 3553objects of that type, the compiler's ability to select a time-efficient 3554alignment is primarily useful only when you plan to create arrays of 3555variables having the relevant (efficiently aligned) type. If you 3556declare or use arrays of variables of an efficiently-aligned type, then 3557it is likely that your program will also be doing pointer arithmetic (or 3558subscripting, which amounts to the same thing) on pointers to the 3559relevant type, and the code that the compiler generates for these 3560pointer arithmetic operations will often be more efficient for 3561efficiently-aligned types than for other types. 3562 3563The @code{aligned} attribute can only increase the alignment; but you 3564can decrease it by specifying @code{packed} as well. See below. 3565 3566Note that the effectiveness of @code{aligned} attributes may be limited 3567by inherent limitations in your linker. On many systems, the linker is 3568only able to arrange for variables to be aligned up to a certain maximum 3569alignment. (For some linkers, the maximum supported alignment may 3570be very very small.) If your linker is only able to align variables 3571up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} 3572in an @code{__attribute__} will still only provide you with 8 byte 3573alignment. See your linker documentation for further information. 3574 3575@item packed 3576This attribute, attached to @code{struct} or @code{union} type 3577definition, specifies that each member (other than zero-width bitfields) 3578of the structure or union is placed to minimize the memory required. When 3579attached to an @code{enum} definition, it indicates that the smallest 3580integral type should be used. 3581 3582@opindex fshort-enums 3583Specifying this attribute for @code{struct} and @code{union} types is 3584equivalent to specifying the @code{packed} attribute on each of the 3585structure or union members. Specifying the @option{-fshort-enums} 3586flag on the line is equivalent to specifying the @code{packed} 3587attribute on all @code{enum} definitions. 3588 3589In the following example @code{struct my_packed_struct}'s members are 3590packed closely together, but the internal layout of its @code{s} member 3591is not packed---to do that, @code{struct my_unpacked_struct} would need to 3592be packed too. 3593 3594@smallexample 3595struct my_unpacked_struct 3596 @{ 3597 char c; 3598 int i; 3599 @}; 3600 3601struct __attribute__ ((__packed__)) my_packed_struct 3602 @{ 3603 char c; 3604 int i; 3605 struct my_unpacked_struct s; 3606 @}; 3607@end smallexample 3608 3609You may only specify this attribute on the definition of a @code{enum}, 3610@code{struct} or @code{union}, not on a @code{typedef} which does not 3611also define the enumerated type, structure or union. 3612 3613@item transparent_union 3614This attribute, attached to a @code{union} type definition, indicates 3615that any function parameter having that union type causes calls to that 3616function to be treated in a special way. 3617 3618First, the argument corresponding to a transparent union type can be of 3619any type in the union; no cast is required. Also, if the union contains 3620a pointer type, the corresponding argument can be a null pointer 3621constant or a void pointer expression; and if the union contains a void 3622pointer type, the corresponding argument can be any pointer expression. 3623If the union member type is a pointer, qualifiers like @code{const} on 3624the referenced type must be respected, just as with normal pointer 3625conversions. 3626 3627Second, the argument is passed to the function using the calling 3628conventions of the first member of the transparent union, not the calling 3629conventions of the union itself. All members of the union must have the 3630same machine representation; this is necessary for this argument passing 3631to work properly. 3632 3633Transparent unions are designed for library functions that have multiple 3634interfaces for compatibility reasons. For example, suppose the 3635@code{wait} function must accept either a value of type @code{int *} to 3636comply with Posix, or a value of type @code{union wait *} to comply with 3637the 4.1BSD interface. If @code{wait}'s parameter were @code{void *}, 3638@code{wait} would accept both kinds of arguments, but it would also 3639accept any other pointer type and this would make argument type checking 3640less useful. Instead, @code{<sys/wait.h>} might define the interface 3641as follows: 3642 3643@smallexample 3644typedef union 3645 @{ 3646 int *__ip; 3647 union wait *__up; 3648 @} wait_status_ptr_t __attribute__ ((__transparent_union__)); 3649 3650pid_t wait (wait_status_ptr_t); 3651@end smallexample 3652 3653This interface allows either @code{int *} or @code{union wait *} 3654arguments to be passed, using the @code{int *} calling convention. 3655The program can call @code{wait} with arguments of either type: 3656 3657@smallexample 3658int w1 () @{ int w; return wait (&w); @} 3659int w2 () @{ union wait w; return wait (&w); @} 3660@end smallexample 3661 3662With this interface, @code{wait}'s implementation might look like this: 3663 3664@smallexample 3665pid_t wait (wait_status_ptr_t p) 3666@{ 3667 return waitpid (-1, p.__ip, 0); 3668@} 3669@end smallexample 3670 3671@item unused 3672When attached to a type (including a @code{union} or a @code{struct}), 3673this attribute means that variables of that type are meant to appear 3674possibly unused. GCC will not produce a warning for any variables of 3675that type, even if the variable appears to do nothing. This is often 3676the case with lock or thread classes, which are usually defined and then 3677not referenced, but contain constructors and destructors that have 3678nontrivial bookkeeping functions. 3679 3680@item deprecated 3681The @code{deprecated} attribute results in a warning if the type 3682is used anywhere in the source file. This is useful when identifying 3683types that are expected to be removed in a future version of a program. 3684If possible, the warning also includes the location of the declaration 3685of the deprecated type, to enable users to easily find further 3686information about why the type is deprecated, or what they should do 3687instead. Note that the warnings only occur for uses and then only 3688if the type is being applied to an identifier that itself is not being 3689declared as deprecated. 3690 3691@smallexample 3692typedef int T1 __attribute__ ((deprecated)); 3693T1 x; 3694typedef T1 T2; 3695T2 y; 3696typedef T1 T3 __attribute__ ((deprecated)); 3697T3 z __attribute__ ((deprecated)); 3698@end smallexample 3699 3700results in a warning on line 2 and 3 but not lines 4, 5, or 6. No 3701warning is issued for line 4 because T2 is not explicitly 3702deprecated. Line 5 has no warning because T3 is explicitly 3703deprecated. Similarly for line 6. 3704 3705The @code{deprecated} attribute can also be used for functions and 3706variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.) 3707 3708@item may_alias 3709Accesses to objects with types with this attribute are not subjected to 3710type-based alias analysis, but are instead assumed to be able to alias 3711any other type of objects, just like the @code{char} type. See 3712@option{-fstrict-aliasing} for more information on aliasing issues. 3713 3714Example of use: 3715 3716@smallexample 3717typedef short __attribute__((__may_alias__)) short_a; 3718 3719int 3720main (void) 3721@{ 3722 int a = 0x12345678; 3723 short_a *b = (short_a *) &a; 3724 3725 b[1] = 0; 3726 3727 if (a == 0x12345678) 3728 abort(); 3729 3730 exit(0); 3731@} 3732@end smallexample 3733 3734If you replaced @code{short_a} with @code{short} in the variable 3735declaration, the above program would abort when compiled with 3736@option{-fstrict-aliasing}, which is on by default at @option{-O2} or 3737above in recent GCC versions. 3738 3739@item visibility 3740In C++, attribute visibility (@pxref{Function Attributes}) can also be 3741applied to class, struct, union and enum types. Unlike other type 3742attributes, the attribute must appear between the initial keyword and 3743the name of the type; it cannot appear after the body of the type. 3744 3745Note that the type visibility is applied to vague linkage entities 3746associated with the class (vtable, typeinfo node, etc.). In 3747particular, if a class is thrown as an exception in one shared object 3748and caught in another, the class must have default visibility. 3749Otherwise the two shared objects will be unable to use the same 3750typeinfo node and exception handling will break. 3751 3752@subsection ARM Type Attributes 3753 3754On those ARM targets that support @code{dllimport} (such as Symbian 3755OS), you can use the @code{notshared} attribute to indicate that the 3756virtual table and other similar data for a class should not be 3757exported from a DLL@. For example: 3758 3759@smallexample 3760class __declspec(notshared) C @{ 3761public: 3762 __declspec(dllimport) C(); 3763 virtual void f(); 3764@} 3765 3766__declspec(dllexport) 3767C::C() @{@} 3768@end smallexample 3769 3770In this code, @code{C::C} is exported from the current DLL, but the 3771virtual table for @code{C} is not exported. (You can use 3772@code{__attribute__} instead of @code{__declspec} if you prefer, but 3773most Symbian OS code uses @code{__declspec}.) 3774 3775@anchor{i386 Type Attributes} 3776@subsection i386 Type Attributes 3777 3778Two attributes are currently defined for i386 configurations: 3779@code{ms_struct} and @code{gcc_struct} 3780 3781@item ms_struct 3782@itemx gcc_struct 3783@cindex @code{ms_struct} 3784@cindex @code{gcc_struct} 3785 3786If @code{packed} is used on a structure, or if bit-fields are used 3787it may be that the Microsoft ABI packs them differently 3788than GCC would normally pack them. Particularly when moving packed 3789data between functions compiled with GCC and the native Microsoft compiler 3790(either via function call or as data in a file), it may be necessary to access 3791either format. 3792 3793Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86 3794compilers to match the native Microsoft compiler. 3795@end table 3796 3797To specify multiple attributes, separate them by commas within the 3798double parentheses: for example, @samp{__attribute__ ((aligned (16), 3799packed))}. 3800 3801@anchor{PowerPC Type Attributes} 3802@subsection PowerPC Type Attributes 3803 3804Three attributes currently are defined for PowerPC configurations: 3805@code{altivec}, @code{ms_struct} and @code{gcc_struct}. 3806 3807For full documentation of the struct attributes please see the 3808documentation in the @xref{i386 Type Attributes}, section. 3809 3810The @code{altivec} attribute allows one to declare AltiVec vector data 3811types supported by the AltiVec Programming Interface Manual. The 3812attribute requires an argument to specify one of three vector types: 3813@code{vector__}, @code{pixel__} (always followed by unsigned short), 3814and @code{bool__} (always followed by unsigned). 3815 3816@smallexample 3817__attribute__((altivec(vector__))) 3818__attribute__((altivec(pixel__))) unsigned short 3819__attribute__((altivec(bool__))) unsigned 3820@end smallexample 3821 3822These attributes mainly are intended to support the @code{__vector}, 3823@code{__pixel}, and @code{__bool} AltiVec keywords. 3824 3825@node Inline 3826@section An Inline Function is As Fast As a Macro 3827@cindex inline functions 3828@cindex integrating function code 3829@cindex open coding 3830@cindex macros, inline alternative 3831 3832By declaring a function @code{inline}, you can direct GCC to 3833integrate that function's code into the code for its callers. This 3834makes execution faster by eliminating the function-call overhead; in 3835addition, if any of the actual argument values are constant, their known 3836values may permit simplifications at compile time so that not all of the 3837inline function's code needs to be included. The effect on code size is 3838less predictable; object code may be larger or smaller with function 3839inlining, depending on the particular case. Inlining of functions is an 3840optimization and it really ``works'' only in optimizing compilation. If 3841you don't use @option{-O}, no function is really inline. 3842 3843Inline functions are included in the ISO C99 standard, but there are 3844currently substantial differences between what GCC implements and what 3845the ISO C99 standard requires. GCC will fully support C99 inline 3846functions in version 4.3. The traditional GCC handling of inline 3847functions will still be available with @option{-std=gnu89}, 3848@option{-fgnu89-inline} or when @code{gnu_inline} attribute is present 3849on all inline declarations. The preprocessor macros 3850@code{__GNUC_GNU_INLINE__} and @code{__GNUC_STDC_INLINE__} may be used 3851to determine the handling of @code{inline} during a particular 3852compilation (@pxref{Common Predefined Macros,,,cpp,The C 3853Preprocessor}). 3854 3855To declare a function inline, use the @code{inline} keyword in its 3856declaration, like this: 3857 3858@smallexample 3859inline int 3860inc (int *a) 3861@{ 3862 (*a)++; 3863@} 3864@end smallexample 3865 3866(If you are writing a header file to be included in ISO C programs, write 3867@code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.) 3868You can also make all ``simple enough'' functions inline with the option 3869@option{-finline-functions}. 3870 3871@opindex Winline 3872Note that certain usages in a function definition can make it unsuitable 3873for inline substitution. Among these usages are: use of varargs, use of 3874alloca, use of variable sized data types (@pxref{Variable Length}), 3875use of computed goto (@pxref{Labels as Values}), use of nonlocal goto, 3876and nested functions (@pxref{Nested Functions}). Using @option{-Winline} 3877will warn when a function marked @code{inline} could not be substituted, 3878and will give the reason for the failure. 3879 3880Note that in C and Objective-C, unlike C++, the @code{inline} keyword 3881does not affect the linkage of the function. 3882 3883@cindex automatic @code{inline} for C++ member fns 3884@cindex @code{inline} automatic for C++ member fns 3885@cindex member fns, automatically @code{inline} 3886@cindex C++ member fns, automatically @code{inline} 3887@opindex fno-default-inline 3888GCC automatically inlines member functions defined within the class 3889body of C++ programs even if they are not explicitly declared 3890@code{inline}. (You can override this with @option{-fno-default-inline}; 3891@pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.) 3892 3893@cindex inline functions, omission of 3894@opindex fkeep-inline-functions 3895When a function is both inline and @code{static}, if all calls to the 3896function are integrated into the caller, and the function's address is 3897never used, then the function's own assembler code is never referenced. 3898In this case, GCC does not actually output assembler code for the 3899function, unless you specify the option @option{-fkeep-inline-functions}. 3900Some calls cannot be integrated for various reasons (in particular, 3901calls that precede the function's definition cannot be integrated, and 3902neither can recursive calls within the definition). If there is a 3903nonintegrated call, then the function is compiled to assembler code as 3904usual. The function must also be compiled as usual if the program 3905refers to its address, because that can't be inlined. 3906 3907@cindex non-static inline function 3908When an inline function is not @code{static}, then the compiler must assume 3909that there may be calls from other source files; since a global symbol can 3910be defined only once in any program, the function must not be defined in 3911the other source files, so the calls therein cannot be integrated. 3912Therefore, a non-@code{static} inline function is always compiled on its 3913own in the usual fashion. 3914 3915If you specify both @code{inline} and @code{extern} in the function 3916definition, then the definition is used only for inlining. In no case 3917is the function compiled on its own, not even if you refer to its 3918address explicitly. Such an address becomes an external reference, as 3919if you had only declared the function, and had not defined it. 3920 3921This combination of @code{inline} and @code{extern} has almost the 3922effect of a macro. The way to use it is to put a function definition in 3923a header file with these keywords, and put another copy of the 3924definition (lacking @code{inline} and @code{extern}) in a library file. 3925The definition in the header file will cause most calls to the function 3926to be inlined. If any uses of the function remain, they will refer to 3927the single copy in the library. 3928 3929Since GCC 4.3 will implement ISO C99 semantics for 3930inline functions, it is simplest to use @code{static inline} only 3931to guarantee compatibility. (The 3932existing semantics will remain available when @option{-std=gnu89} is 3933specified, but eventually the default will be @option{-std=gnu99}; 3934that will implement the C99 semantics, though it does not do so in 3935versions of GCC before 4.3. After the default changes, the existing 3936semantics will still be available via the @option{-fgnu89-inline} 3937option or the @code{gnu_inline} function attribute.) 3938 3939GCC does not inline any functions when not optimizing unless you specify 3940the @samp{always_inline} attribute for the function, like this: 3941 3942@smallexample 3943/* @r{Prototype.} */ 3944inline void foo (const char) __attribute__((always_inline)); 3945@end smallexample 3946 3947@node Extended Asm 3948@section Assembler Instructions with C Expression Operands 3949@cindex extended @code{asm} 3950@cindex @code{asm} expressions 3951@cindex assembler instructions 3952@cindex registers 3953 3954In an assembler instruction using @code{asm}, you can specify the 3955operands of the instruction using C expressions. This means you need not 3956guess which registers or memory locations will contain the data you want 3957to use. 3958 3959You must specify an assembler instruction template much like what 3960appears in a machine description, plus an operand constraint string for 3961each operand. 3962 3963For example, here is how to use the 68881's @code{fsinx} instruction: 3964 3965@smallexample 3966asm ("fsinx %1,%0" : "=f" (result) : "f" (angle)); 3967@end smallexample 3968 3969@noindent 3970Here @code{angle} is the C expression for the input operand while 3971@code{result} is that of the output operand. Each has @samp{"f"} as its 3972operand constraint, saying that a floating point register is required. 3973The @samp{=} in @samp{=f} indicates that the operand is an output; all 3974output operands' constraints must use @samp{=}. The constraints use the 3975same language used in the machine description (@pxref{Constraints}). 3976 3977Each operand is described by an operand-constraint string followed by 3978the C expression in parentheses. A colon separates the assembler 3979template from the first output operand and another separates the last 3980output operand from the first input, if any. Commas separate the 3981operands within each group. The total number of operands is currently 3982limited to 30; this limitation may be lifted in some future version of 3983GCC@. 3984 3985If there are no output operands but there are input operands, you must 3986place two consecutive colons surrounding the place where the output 3987operands would go. 3988 3989As of GCC version 3.1, it is also possible to specify input and output 3990operands using symbolic names which can be referenced within the 3991assembler code. These names are specified inside square brackets 3992preceding the constraint string, and can be referenced inside the 3993assembler code using @code{%[@var{name}]} instead of a percentage sign 3994followed by the operand number. Using named operands the above example 3995could look like: 3996 3997@smallexample 3998asm ("fsinx %[angle],%[output]" 3999 : [output] "=f" (result) 4000 : [angle] "f" (angle)); 4001@end smallexample 4002 4003@noindent 4004Note that the symbolic operand names have no relation whatsoever to 4005other C identifiers. You may use any name you like, even those of 4006existing C symbols, but you must ensure that no two operands within the same 4007assembler construct use the same symbolic name. 4008 4009Output operand expressions must be lvalues; the compiler can check this. 4010The input operands need not be lvalues. The compiler cannot check 4011whether the operands have data types that are reasonable for the 4012instruction being executed. It does not parse the assembler instruction 4013template and does not know what it means or even whether it is valid 4014assembler input. The extended @code{asm} feature is most often used for 4015machine instructions the compiler itself does not know exist. If 4016the output expression cannot be directly addressed (for example, it is a 4017bit-field), your constraint must allow a register. In that case, GCC 4018will use the register as the output of the @code{asm}, and then store 4019that register into the output. 4020 4021The ordinary output operands must be write-only; GCC will assume that 4022the values in these operands before the instruction are dead and need 4023not be generated. Extended asm supports input-output or read-write 4024operands. Use the constraint character @samp{+} to indicate such an 4025operand and list it with the output operands. You should only use 4026read-write operands when the constraints for the operand (or the 4027operand in which only some of the bits are to be changed) allow a 4028register. 4029 4030You may, as an alternative, logically split its function into two 4031separate operands, one input operand and one write-only output 4032operand. The connection between them is expressed by constraints 4033which say they need to be in the same location when the instruction 4034executes. You can use the same C expression for both operands, or 4035different expressions. For example, here we write the (fictitious) 4036@samp{combine} instruction with @code{bar} as its read-only source 4037operand and @code{foo} as its read-write destination: 4038 4039@smallexample 4040asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar)); 4041@end smallexample 4042 4043@noindent 4044The constraint @samp{"0"} for operand 1 says that it must occupy the 4045same location as operand 0. A number in constraint is allowed only in 4046an input operand and it must refer to an output operand. 4047 4048Only a number in the constraint can guarantee that one operand will be in 4049the same place as another. The mere fact that @code{foo} is the value 4050of both operands is not enough to guarantee that they will be in the 4051same place in the generated assembler code. The following would not 4052work reliably: 4053 4054@smallexample 4055asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar)); 4056@end smallexample 4057 4058Various optimizations or reloading could cause operands 0 and 1 to be in 4059different registers; GCC knows no reason not to do so. For example, the 4060compiler might find a copy of the value of @code{foo} in one register and 4061use it for operand 1, but generate the output operand 0 in a different 4062register (copying it afterward to @code{foo}'s own address). Of course, 4063since the register for operand 1 is not even mentioned in the assembler 4064code, the result will not work, but GCC can't tell that. 4065 4066As of GCC version 3.1, one may write @code{[@var{name}]} instead of 4067the operand number for a matching constraint. For example: 4068 4069@smallexample 4070asm ("cmoveq %1,%2,%[result]" 4071 : [result] "=r"(result) 4072 : "r" (test), "r"(new), "[result]"(old)); 4073@end smallexample 4074 4075Sometimes you need to make an @code{asm} operand be a specific register, 4076but there's no matching constraint letter for that register @emph{by 4077itself}. To force the operand into that register, use a local variable 4078for the operand and specify the register in the variable declaration. 4079@xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any 4080register constraint letter that matches the register: 4081 4082@smallexample 4083register int *p1 asm ("r0") = @dots{}; 4084register int *p2 asm ("r1") = @dots{}; 4085register int *result asm ("r0"); 4086asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2)); 4087@end smallexample 4088 4089@anchor{Example of asm with clobbered asm reg} 4090In the above example, beware that a register that is call-clobbered by 4091the target ABI will be overwritten by any function call in the 4092assignment, including library calls for arithmetic operators. 4093Assuming it is a call-clobbered register, this may happen to @code{r0} 4094above by the assignment to @code{p2}. If you have to use such a 4095register, use temporary variables for expressions between the register 4096assignment and use: 4097 4098@smallexample 4099int t1 = @dots{}; 4100register int *p1 asm ("r0") = @dots{}; 4101register int *p2 asm ("r1") = t1; 4102register int *result asm ("r0"); 4103asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2)); 4104@end smallexample 4105 4106Some instructions clobber specific hard registers. To describe this, 4107write a third colon after the input operands, followed by the names of 4108the clobbered hard registers (given as strings). Here is a realistic 4109example for the VAX: 4110 4111@smallexample 4112asm volatile ("movc3 %0,%1,%2" 4113 : /* @r{no outputs} */ 4114 : "g" (from), "g" (to), "g" (count) 4115 : "r0", "r1", "r2", "r3", "r4", "r5"); 4116@end smallexample 4117 4118You may not write a clobber description in a way that overlaps with an 4119input or output operand. For example, you may not have an operand 4120describing a register class with one member if you mention that register 4121in the clobber list. Variables declared to live in specific registers 4122(@pxref{Explicit Reg Vars}), and used as asm input or output operands must 4123have no part mentioned in the clobber description. 4124There is no way for you to specify that an input 4125operand is modified without also specifying it as an output 4126operand. Note that if all the output operands you specify are for this 4127purpose (and hence unused), you will then also need to specify 4128@code{volatile} for the @code{asm} construct, as described below, to 4129prevent GCC from deleting the @code{asm} statement as unused. 4130 4131If you refer to a particular hardware register from the assembler code, 4132you will probably have to list the register after the third colon to 4133tell the compiler the register's value is modified. In some assemblers, 4134the register names begin with @samp{%}; to produce one @samp{%} in the 4135assembler code, you must write @samp{%%} in the input. 4136 4137If your assembler instruction can alter the condition code register, add 4138@samp{cc} to the list of clobbered registers. GCC on some machines 4139represents the condition codes as a specific hardware register; 4140@samp{cc} serves to name this register. On other machines, the 4141condition code is handled differently, and specifying @samp{cc} has no 4142effect. But it is valid no matter what the machine. 4143 4144If your assembler instructions access memory in an unpredictable 4145fashion, add @samp{memory} to the list of clobbered registers. This 4146will cause GCC to not keep memory values cached in registers across the 4147assembler instruction and not optimize stores or loads to that memory. 4148You will also want to add the @code{volatile} keyword if the memory 4149affected is not listed in the inputs or outputs of the @code{asm}, as 4150the @samp{memory} clobber does not count as a side-effect of the 4151@code{asm}. If you know how large the accessed memory is, you can add 4152it as input or output but if this is not known, you should add 4153@samp{memory}. As an example, if you access ten bytes of a string, you 4154can use a memory input like: 4155 4156@smallexample 4157@{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}. 4158@end smallexample 4159 4160Note that in the following example the memory input is necessary, 4161otherwise GCC might optimize the store to @code{x} away: 4162@smallexample 4163int foo () 4164@{ 4165 int x = 42; 4166 int *y = &x; 4167 int result; 4168 asm ("magic stuff accessing an 'int' pointed to by '%1'" 4169 "=&d" (r) : "a" (y), "m" (*y)); 4170 return result; 4171@} 4172@end smallexample 4173 4174You can put multiple assembler instructions together in a single 4175@code{asm} template, separated by the characters normally used in assembly 4176code for the system. A combination that works in most places is a newline 4177to break the line, plus a tab character to move to the instruction field 4178(written as @samp{\n\t}). Sometimes semicolons can be used, if the 4179assembler allows semicolons as a line-breaking character. Note that some 4180assembler dialects use semicolons to start a comment. 4181The input operands are guaranteed not to use any of the clobbered 4182registers, and neither will the output operands' addresses, so you can 4183read and write the clobbered registers as many times as you like. Here 4184is an example of multiple instructions in a template; it assumes the 4185subroutine @code{_foo} accepts arguments in registers 9 and 10: 4186 4187@smallexample 4188asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo" 4189 : /* no outputs */ 4190 : "g" (from), "g" (to) 4191 : "r9", "r10"); 4192@end smallexample 4193 4194Unless an output operand has the @samp{&} constraint modifier, GCC 4195may allocate it in the same register as an unrelated input operand, on 4196the assumption the inputs are consumed before the outputs are produced. 4197This assumption may be false if the assembler code actually consists of 4198more than one instruction. In such a case, use @samp{&} for each output 4199operand that may not overlap an input. @xref{Modifiers}. 4200 4201If you want to test the condition code produced by an assembler 4202instruction, you must include a branch and a label in the @code{asm} 4203construct, as follows: 4204 4205@smallexample 4206asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:" 4207 : "g" (result) 4208 : "g" (input)); 4209@end smallexample 4210 4211@noindent 4212This assumes your assembler supports local labels, as the GNU assembler 4213and most Unix assemblers do. 4214 4215Speaking of labels, jumps from one @code{asm} to another are not 4216supported. The compiler's optimizers do not know about these jumps, and 4217therefore they cannot take account of them when deciding how to 4218optimize. 4219 4220@cindex macros containing @code{asm} 4221Usually the most convenient way to use these @code{asm} instructions is to 4222encapsulate them in macros that look like functions. For example, 4223 4224@smallexample 4225#define sin(x) \ 4226(@{ double __value, __arg = (x); \ 4227 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \ 4228 __value; @}) 4229@end smallexample 4230 4231@noindent 4232Here the variable @code{__arg} is used to make sure that the instruction 4233operates on a proper @code{double} value, and to accept only those 4234arguments @code{x} which can convert automatically to a @code{double}. 4235 4236Another way to make sure the instruction operates on the correct data 4237type is to use a cast in the @code{asm}. This is different from using a 4238variable @code{__arg} in that it converts more different types. For 4239example, if the desired type were @code{int}, casting the argument to 4240@code{int} would accept a pointer with no complaint, while assigning the 4241argument to an @code{int} variable named @code{__arg} would warn about 4242using a pointer unless the caller explicitly casts it. 4243 4244If an @code{asm} has output operands, GCC assumes for optimization 4245purposes the instruction has no side effects except to change the output 4246operands. This does not mean instructions with a side effect cannot be 4247used, but you must be careful, because the compiler may eliminate them 4248if the output operands aren't used, or move them out of loops, or 4249replace two with one if they constitute a common subexpression. Also, 4250if your instruction does have a side effect on a variable that otherwise 4251appears not to change, the old value of the variable may be reused later 4252if it happens to be found in a register. 4253 4254You can prevent an @code{asm} instruction from being deleted 4255by writing the keyword @code{volatile} after 4256the @code{asm}. For example: 4257 4258@smallexample 4259#define get_and_set_priority(new) \ 4260(@{ int __old; \ 4261 asm volatile ("get_and_set_priority %0, %1" \ 4262 : "=g" (__old) : "g" (new)); \ 4263 __old; @}) 4264@end smallexample 4265 4266@noindent 4267The @code{volatile} keyword indicates that the instruction has 4268important side-effects. GCC will not delete a volatile @code{asm} if 4269it is reachable. (The instruction can still be deleted if GCC can 4270prove that control-flow will never reach the location of the 4271instruction.) Note that even a volatile @code{asm} instruction 4272can be moved relative to other code, including across jump 4273instructions. For example, on many targets there is a system 4274register which can be set to control the rounding mode of 4275floating point operations. You might try 4276setting it with a volatile @code{asm}, like this PowerPC example: 4277 4278@smallexample 4279 asm volatile("mtfsf 255,%0" : : "f" (fpenv)); 4280 sum = x + y; 4281@end smallexample 4282 4283@noindent 4284This will not work reliably, as the compiler may move the addition back 4285before the volatile @code{asm}. To make it work you need to add an 4286artificial dependency to the @code{asm} referencing a variable in the code 4287you don't want moved, for example: 4288 4289@smallexample 4290 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv)); 4291 sum = x + y; 4292@end smallexample 4293 4294Similarly, you can't expect a 4295sequence of volatile @code{asm} instructions to remain perfectly 4296consecutive. If you want consecutive output, use a single @code{asm}. 4297Also, GCC will perform some optimizations across a volatile @code{asm} 4298instruction; GCC does not ``forget everything'' when it encounters 4299a volatile @code{asm} instruction the way some other compilers do. 4300 4301An @code{asm} instruction without any output operands will be treated 4302identically to a volatile @code{asm} instruction. 4303 4304It is a natural idea to look for a way to give access to the condition 4305code left by the assembler instruction. However, when we attempted to 4306implement this, we found no way to make it work reliably. The problem 4307is that output operands might need reloading, which would result in 4308additional following ``store'' instructions. On most machines, these 4309instructions would alter the condition code before there was time to 4310test it. This problem doesn't arise for ordinary ``test'' and 4311``compare'' instructions because they don't have any output operands. 4312 4313For reasons similar to those described above, it is not possible to give 4314an assembler instruction access to the condition code left by previous 4315instructions. 4316 4317If you are writing a header file that should be includable in ISO C 4318programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate 4319Keywords}. 4320 4321@subsection Size of an @code{asm} 4322 4323Some targets require that GCC track the size of each instruction used in 4324order to generate correct code. Because the final length of an 4325@code{asm} is only known by the assembler, GCC must make an estimate as 4326to how big it will be. The estimate is formed by counting the number of 4327statements in the pattern of the @code{asm} and multiplying that by the 4328length of the longest instruction on that processor. Statements in the 4329@code{asm} are identified by newline characters and whatever statement 4330separator characters are supported by the assembler; on most processors 4331this is the `@code{;}' character. 4332 4333Normally, GCC's estimate is perfectly adequate to ensure that correct 4334code is generated, but it is possible to confuse the compiler if you use 4335pseudo instructions or assembler macros that expand into multiple real 4336instructions or if you use assembler directives that expand to more 4337space in the object file than would be needed for a single instruction. 4338If this happens then the assembler will produce a diagnostic saying that 4339a label is unreachable. 4340 4341@subsection i386 floating point asm operands 4342 4343There are several rules on the usage of stack-like regs in 4344asm_operands insns. These rules apply only to the operands that are 4345stack-like regs: 4346 4347@enumerate 4348@item 4349Given a set of input regs that die in an asm_operands, it is 4350necessary to know which are implicitly popped by the asm, and 4351which must be explicitly popped by gcc. 4352 4353An input reg that is implicitly popped by the asm must be 4354explicitly clobbered, unless it is constrained to match an 4355output operand. 4356 4357@item 4358For any input reg that is implicitly popped by an asm, it is 4359necessary to know how to adjust the stack to compensate for the pop. 4360If any non-popped input is closer to the top of the reg-stack than 4361the implicitly popped reg, it would not be possible to know what the 4362stack looked like---it's not clear how the rest of the stack ``slides 4363up''. 4364 4365All implicitly popped input regs must be closer to the top of 4366the reg-stack than any input that is not implicitly popped. 4367 4368It is possible that if an input dies in an insn, reload might 4369use the input reg for an output reload. Consider this example: 4370 4371@smallexample 4372asm ("foo" : "=t" (a) : "f" (b)); 4373@end smallexample 4374 4375This asm says that input B is not popped by the asm, and that 4376the asm pushes a result onto the reg-stack, i.e., the stack is one 4377deeper after the asm than it was before. But, it is possible that 4378reload will think that it can use the same reg for both the input and 4379the output, if input B dies in this insn. 4380 4381If any input operand uses the @code{f} constraint, all output reg 4382constraints must use the @code{&} earlyclobber. 4383 4384The asm above would be written as 4385 4386@smallexample 4387asm ("foo" : "=&t" (a) : "f" (b)); 4388@end smallexample 4389 4390@item 4391Some operands need to be in particular places on the stack. All 4392output operands fall in this category---there is no other way to 4393know which regs the outputs appear in unless the user indicates 4394this in the constraints. 4395 4396Output operands must specifically indicate which reg an output 4397appears in after an asm. @code{=f} is not allowed: the operand 4398constraints must select a class with a single reg. 4399 4400@item 4401Output operands may not be ``inserted'' between existing stack regs. 4402Since no 387 opcode uses a read/write operand, all output operands 4403are dead before the asm_operands, and are pushed by the asm_operands. 4404It makes no sense to push anywhere but the top of the reg-stack. 4405 4406Output operands must start at the top of the reg-stack: output 4407operands may not ``skip'' a reg. 4408 4409@item 4410Some asm statements may need extra stack space for internal 4411calculations. This can be guaranteed by clobbering stack registers 4412unrelated to the inputs and outputs. 4413 4414@end enumerate 4415 4416Here are a couple of reasonable asms to want to write. This asm 4417takes one input, which is internally popped, and produces two outputs. 4418 4419@smallexample 4420asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); 4421@end smallexample 4422 4423This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode, 4424and replaces them with one output. The user must code the @code{st(1)} 4425clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs. 4426 4427@smallexample 4428asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); 4429@end smallexample 4430 4431@include md.texi 4432 4433@node Asm Labels 4434@section Controlling Names Used in Assembler Code 4435@cindex assembler names for identifiers 4436@cindex names used in assembler code 4437@cindex identifiers, names in assembler code 4438 4439You can specify the name to be used in the assembler code for a C 4440function or variable by writing the @code{asm} (or @code{__asm__}) 4441keyword after the declarator as follows: 4442 4443@smallexample 4444int foo asm ("myfoo") = 2; 4445@end smallexample 4446 4447@noindent 4448This specifies that the name to be used for the variable @code{foo} in 4449the assembler code should be @samp{myfoo} rather than the usual 4450@samp{_foo}. 4451 4452On systems where an underscore is normally prepended to the name of a C 4453function or variable, this feature allows you to define names for the 4454linker that do not start with an underscore. 4455 4456It does not make sense to use this feature with a non-static local 4457variable since such variables do not have assembler names. If you are 4458trying to put the variable in a particular register, see @ref{Explicit 4459Reg Vars}. GCC presently accepts such code with a warning, but will 4460probably be changed to issue an error, rather than a warning, in the 4461future. 4462 4463You cannot use @code{asm} in this way in a function @emph{definition}; but 4464you can get the same effect by writing a declaration for the function 4465before its definition and putting @code{asm} there, like this: 4466 4467@smallexample 4468extern func () asm ("FUNC"); 4469 4470func (x, y) 4471 int x, y; 4472/* @r{@dots{}} */ 4473@end smallexample 4474 4475It is up to you to make sure that the assembler names you choose do not 4476conflict with any other assembler symbols. Also, you must not use a 4477register name; that would produce completely invalid assembler code. GCC 4478does not as yet have the ability to store static variables in registers. 4479Perhaps that will be added. 4480 4481@node Explicit Reg Vars 4482@section Variables in Specified Registers 4483@cindex explicit register variables 4484@cindex variables in specified registers 4485@cindex specified registers 4486@cindex registers, global allocation 4487 4488GNU C allows you to put a few global variables into specified hardware 4489registers. You can also specify the register in which an ordinary 4490register variable should be allocated. 4491 4492@itemize @bullet 4493@item 4494Global register variables reserve registers throughout the program. 4495This may be useful in programs such as programming language 4496interpreters which have a couple of global variables that are accessed 4497very often. 4498 4499@item 4500Local register variables in specific registers do not reserve the 4501registers, except at the point where they are used as input or output 4502operands in an @code{asm} statement and the @code{asm} statement itself is 4503not deleted. The compiler's data flow analysis is capable of determining 4504where the specified registers contain live values, and where they are 4505available for other uses. Stores into local register variables may be deleted 4506when they appear to be dead according to dataflow analysis. References 4507to local register variables may be deleted or moved or simplified. 4508 4509These local variables are sometimes convenient for use with the extended 4510@code{asm} feature (@pxref{Extended Asm}), if you want to write one 4511output of the assembler instruction directly into a particular register. 4512(This will work provided the register you specify fits the constraints 4513specified for that operand in the @code{asm}.) 4514@end itemize 4515 4516@menu 4517* Global Reg Vars:: 4518* Local Reg Vars:: 4519@end menu 4520 4521@node Global Reg Vars 4522@subsection Defining Global Register Variables 4523@cindex global register variables 4524@cindex registers, global variables in 4525 4526You can define a global register variable in GNU C like this: 4527 4528@smallexample 4529register int *foo asm ("a5"); 4530@end smallexample 4531 4532@noindent 4533Here @code{a5} is the name of the register which should be used. Choose a 4534register which is normally saved and restored by function calls on your 4535machine, so that library routines will not clobber it. 4536 4537Naturally the register name is cpu-dependent, so you would need to 4538conditionalize your program according to cpu type. The register 4539@code{a5} would be a good choice on a 68000 for a variable of pointer 4540type. On machines with register windows, be sure to choose a ``global'' 4541register that is not affected magically by the function call mechanism. 4542 4543In addition, operating systems on one type of cpu may differ in how they 4544name the registers; then you would need additional conditionals. For 4545example, some 68000 operating systems call this register @code{%a5}. 4546 4547Eventually there may be a way of asking the compiler to choose a register 4548automatically, but first we need to figure out how it should choose and 4549how to enable you to guide the choice. No solution is evident. 4550 4551Defining a global register variable in a certain register reserves that 4552register entirely for this use, at least within the current compilation. 4553The register will not be allocated for any other purpose in the functions 4554in the current compilation. The register will not be saved and restored by 4555these functions. Stores into this register are never deleted even if they 4556would appear to be dead, but references may be deleted or moved or 4557simplified. 4558 4559It is not safe to access the global register variables from signal 4560handlers, or from more than one thread of control, because the system 4561library routines may temporarily use the register for other things (unless 4562you recompile them specially for the task at hand). 4563 4564@cindex @code{qsort}, and global register variables 4565It is not safe for one function that uses a global register variable to 4566call another such function @code{foo} by way of a third function 4567@code{lose} that was compiled without knowledge of this variable (i.e.@: in a 4568different source file in which the variable wasn't declared). This is 4569because @code{lose} might save the register and put some other value there. 4570For example, you can't expect a global register variable to be available in 4571the comparison-function that you pass to @code{qsort}, since @code{qsort} 4572might have put something else in that register. (If you are prepared to 4573recompile @code{qsort} with the same global register variable, you can 4574solve this problem.) 4575 4576If you want to recompile @code{qsort} or other source files which do not 4577actually use your global register variable, so that they will not use that 4578register for any other purpose, then it suffices to specify the compiler 4579option @option{-ffixed-@var{reg}}. You need not actually add a global 4580register declaration to their source code. 4581 4582A function which can alter the value of a global register variable cannot 4583safely be called from a function compiled without this variable, because it 4584could clobber the value the caller expects to find there on return. 4585Therefore, the function which is the entry point into the part of the 4586program that uses the global register variable must explicitly save and 4587restore the value which belongs to its caller. 4588 4589@cindex register variable after @code{longjmp} 4590@cindex global register after @code{longjmp} 4591@cindex value after @code{longjmp} 4592@findex longjmp 4593@findex setjmp 4594On most machines, @code{longjmp} will restore to each global register 4595variable the value it had at the time of the @code{setjmp}. On some 4596machines, however, @code{longjmp} will not change the value of global 4597register variables. To be portable, the function that called @code{setjmp} 4598should make other arrangements to save the values of the global register 4599variables, and to restore them in a @code{longjmp}. This way, the same 4600thing will happen regardless of what @code{longjmp} does. 4601 4602All global register variable declarations must precede all function 4603definitions. If such a declaration could appear after function 4604definitions, the declaration would be too late to prevent the register from 4605being used for other purposes in the preceding functions. 4606 4607Global register variables may not have initial values, because an 4608executable file has no means to supply initial contents for a register. 4609 4610On the SPARC, there are reports that g3 @dots{} g7 are suitable 4611registers, but certain library functions, such as @code{getwd}, as well 4612as the subroutines for division and remainder, modify g3 and g4. g1 and 4613g2 are local temporaries. 4614 4615On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7. 4616Of course, it will not do to use more than a few of those. 4617 4618@node Local Reg Vars 4619@subsection Specifying Registers for Local Variables 4620@cindex local variables, specifying registers 4621@cindex specifying registers for local variables 4622@cindex registers for local variables 4623 4624You can define a local register variable with a specified register 4625like this: 4626 4627@smallexample 4628register int *foo asm ("a5"); 4629@end smallexample 4630 4631@noindent 4632Here @code{a5} is the name of the register which should be used. Note 4633that this is the same syntax used for defining global register 4634variables, but for a local variable it would appear within a function. 4635 4636Naturally the register name is cpu-dependent, but this is not a 4637problem, since specific registers are most often useful with explicit 4638assembler instructions (@pxref{Extended Asm}). Both of these things 4639generally require that you conditionalize your program according to 4640cpu type. 4641 4642In addition, operating systems on one type of cpu may differ in how they 4643name the registers; then you would need additional conditionals. For 4644example, some 68000 operating systems call this register @code{%a5}. 4645 4646Defining such a register variable does not reserve the register; it 4647remains available for other uses in places where flow control determines 4648the variable's value is not live. 4649 4650This option does not guarantee that GCC will generate code that has 4651this variable in the register you specify at all times. You may not 4652code an explicit reference to this register in the @emph{assembler 4653instruction template} part of an @code{asm} statement and assume it will 4654always refer to this variable. However, using the variable as an 4655@code{asm} @emph{operand} guarantees that the specified register is used 4656for the operand. 4657 4658Stores into local register variables may be deleted when they appear to be dead 4659according to dataflow analysis. References to local register variables may 4660be deleted or moved or simplified. 4661 4662As for global register variables, it's recommended that you choose a 4663register which is normally saved and restored by function calls on 4664your machine, so that library routines will not clobber it. A common 4665pitfall is to initialize multiple call-clobbered registers with 4666arbitrary expressions, where a function call or library call for an 4667arithmetic operator will overwrite a register value from a previous 4668assignment, for example @code{r0} below: 4669@smallexample 4670register int *p1 asm ("r0") = @dots{}; 4671register int *p2 asm ("r1") = @dots{}; 4672@end smallexample 4673In those cases, a solution is to use a temporary variable for 4674each arbitrary expression. @xref{Example of asm with clobbered asm reg}. 4675 4676@node Alternate Keywords 4677@section Alternate Keywords 4678@cindex alternate keywords 4679@cindex keywords, alternate 4680 4681@option{-ansi} and the various @option{-std} options disable certain 4682keywords. This causes trouble when you want to use GNU C extensions, or 4683a general-purpose header file that should be usable by all programs, 4684including ISO C programs. The keywords @code{asm}, @code{typeof} and 4685@code{inline} are not available in programs compiled with 4686@option{-ansi} or @option{-std} (although @code{inline} can be used in a 4687program compiled with @option{-std=c99}). The ISO C99 keyword 4688@code{restrict} is only available when @option{-std=gnu99} (which will 4689eventually be the default) or @option{-std=c99} (or the equivalent 4690@option{-std=iso9899:1999}) is used. 4691 4692The way to solve these problems is to put @samp{__} at the beginning and 4693end of each problematical keyword. For example, use @code{__asm__} 4694instead of @code{asm}, and @code{__inline__} instead of @code{inline}. 4695 4696Other C compilers won't accept these alternative keywords; if you want to 4697compile with another compiler, you can define the alternate keywords as 4698macros to replace them with the customary keywords. It looks like this: 4699 4700@smallexample 4701#ifndef __GNUC__ 4702#define __asm__ asm 4703#endif 4704@end smallexample 4705 4706@findex __extension__ 4707@opindex pedantic 4708@option{-pedantic} and other options cause warnings for many GNU C extensions. 4709You can 4710prevent such warnings within one expression by writing 4711@code{__extension__} before the expression. @code{__extension__} has no 4712effect aside from this. 4713 4714@node Incomplete Enums 4715@section Incomplete @code{enum} Types 4716 4717You can define an @code{enum} tag without specifying its possible values. 4718This results in an incomplete type, much like what you get if you write 4719@code{struct foo} without describing the elements. A later declaration 4720which does specify the possible values completes the type. 4721 4722You can't allocate variables or storage using the type while it is 4723incomplete. However, you can work with pointers to that type. 4724 4725This extension may not be very useful, but it makes the handling of 4726@code{enum} more consistent with the way @code{struct} and @code{union} 4727are handled. 4728 4729This extension is not supported by GNU C++. 4730 4731@node Function Names 4732@section Function Names as Strings 4733@cindex @code{__func__} identifier 4734@cindex @code{__FUNCTION__} identifier 4735@cindex @code{__PRETTY_FUNCTION__} identifier 4736 4737GCC provides three magic variables which hold the name of the current 4738function, as a string. The first of these is @code{__func__}, which 4739is part of the C99 standard: 4740 4741@display 4742The identifier @code{__func__} is implicitly declared by the translator 4743as if, immediately following the opening brace of each function 4744definition, the declaration 4745 4746@smallexample 4747static const char __func__[] = "function-name"; 4748@end smallexample 4749 4750appeared, where function-name is the name of the lexically-enclosing 4751function. This name is the unadorned name of the function. 4752@end display 4753 4754@code{__FUNCTION__} is another name for @code{__func__}. Older 4755versions of GCC recognize only this name. However, it is not 4756standardized. For maximum portability, we recommend you use 4757@code{__func__}, but provide a fallback definition with the 4758preprocessor: 4759 4760@smallexample 4761#if __STDC_VERSION__ < 199901L 4762# if __GNUC__ >= 2 4763# define __func__ __FUNCTION__ 4764# else 4765# define __func__ "<unknown>" 4766# endif 4767#endif 4768@end smallexample 4769 4770In C, @code{__PRETTY_FUNCTION__} is yet another name for 4771@code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains 4772the type signature of the function as well as its bare name. For 4773example, this program: 4774 4775@smallexample 4776extern "C" @{ 4777extern int printf (char *, ...); 4778@} 4779 4780class a @{ 4781 public: 4782 void sub (int i) 4783 @{ 4784 printf ("__FUNCTION__ = %s\n", __FUNCTION__); 4785 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); 4786 @} 4787@}; 4788 4789int 4790main (void) 4791@{ 4792 a ax; 4793 ax.sub (0); 4794 return 0; 4795@} 4796@end smallexample 4797 4798@noindent 4799gives this output: 4800 4801@smallexample 4802__FUNCTION__ = sub 4803__PRETTY_FUNCTION__ = void a::sub(int) 4804@end smallexample 4805 4806These identifiers are not preprocessor macros. In GCC 3.3 and 4807earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} 4808were treated as string literals; they could be used to initialize 4809@code{char} arrays, and they could be concatenated with other string 4810literals. GCC 3.4 and later treat them as variables, like 4811@code{__func__}. In C++, @code{__FUNCTION__} and 4812@code{__PRETTY_FUNCTION__} have always been variables. 4813 4814@node Return Address 4815@section Getting the Return or Frame Address of a Function 4816 4817These functions may be used to get information about the callers of a 4818function. 4819 4820@deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level}) 4821This function returns the return address of the current function, or of 4822one of its callers. The @var{level} argument is number of frames to 4823scan up the call stack. A value of @code{0} yields the return address 4824of the current function, a value of @code{1} yields the return address 4825of the caller of the current function, and so forth. When inlining 4826the expected behavior is that the function will return the address of 4827the function that will be returned to. To work around this behavior use 4828the @code{noinline} function attribute. 4829 4830The @var{level} argument must be a constant integer. 4831 4832On some machines it may be impossible to determine the return address of 4833any function other than the current one; in such cases, or when the top 4834of the stack has been reached, this function will return @code{0} or a 4835random value. In addition, @code{__builtin_frame_address} may be used 4836to determine if the top of the stack has been reached. 4837 4838This function should only be used with a nonzero argument for debugging 4839purposes. 4840@end deftypefn 4841 4842@deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level}) 4843This function is similar to @code{__builtin_return_address}, but it 4844returns the address of the function frame rather than the return address 4845of the function. Calling @code{__builtin_frame_address} with a value of 4846@code{0} yields the frame address of the current function, a value of 4847@code{1} yields the frame address of the caller of the current function, 4848and so forth. 4849 4850The frame is the area on the stack which holds local variables and saved 4851registers. The frame address is normally the address of the first word 4852pushed on to the stack by the function. However, the exact definition 4853depends upon the processor and the calling convention. If the processor 4854has a dedicated frame pointer register, and the function has a frame, 4855then @code{__builtin_frame_address} will return the value of the frame 4856pointer register. 4857 4858On some machines it may be impossible to determine the frame address of 4859any function other than the current one; in such cases, or when the top 4860of the stack has been reached, this function will return @code{0} if 4861the first frame pointer is properly initialized by the startup code. 4862 4863This function should only be used with a nonzero argument for debugging 4864purposes. 4865@end deftypefn 4866 4867@node Vector Extensions 4868@section Using vector instructions through built-in functions 4869 4870On some targets, the instruction set contains SIMD vector instructions that 4871operate on multiple values contained in one large register at the same time. 4872For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used 4873this way. 4874 4875The first step in using these extensions is to provide the necessary data 4876types. This should be done using an appropriate @code{typedef}: 4877 4878@smallexample 4879typedef int v4si __attribute__ ((vector_size (16))); 4880@end smallexample 4881 4882The @code{int} type specifies the base type, while the attribute specifies 4883the vector size for the variable, measured in bytes. For example, the 4884declaration above causes the compiler to set the mode for the @code{v4si} 4885type to be 16 bytes wide and divided into @code{int} sized units. For 4886a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the 4887corresponding mode of @code{foo} will be @acronym{V4SI}. 4888 4889The @code{vector_size} attribute is only applicable to integral and 4890float scalars, although arrays, pointers, and function return values 4891are allowed in conjunction with this construct. 4892 4893All the basic integer types can be used as base types, both as signed 4894and as unsigned: @code{char}, @code{short}, @code{int}, @code{long}, 4895@code{long long}. In addition, @code{float} and @code{double} can be 4896used to build floating-point vector types. 4897 4898Specifying a combination that is not valid for the current architecture 4899will cause GCC to synthesize the instructions using a narrower mode. 4900For example, if you specify a variable of type @code{V4SI} and your 4901architecture does not allow for this specific SIMD type, GCC will 4902produce code that uses 4 @code{SIs}. 4903 4904The types defined in this manner can be used with a subset of normal C 4905operations. Currently, GCC will allow using the following operators 4906on these types: @code{+, -, *, /, unary minus, ^, |, &, ~}@. 4907 4908The operations behave like C++ @code{valarrays}. Addition is defined as 4909the addition of the corresponding elements of the operands. For 4910example, in the code below, each of the 4 elements in @var{a} will be 4911added to the corresponding 4 elements in @var{b} and the resulting 4912vector will be stored in @var{c}. 4913 4914@smallexample 4915typedef int v4si __attribute__ ((vector_size (16))); 4916 4917v4si a, b, c; 4918 4919c = a + b; 4920@end smallexample 4921 4922Subtraction, multiplication, division, and the logical operations 4923operate in a similar manner. Likewise, the result of using the unary 4924minus or complement operators on a vector type is a vector whose 4925elements are the negative or complemented values of the corresponding 4926elements in the operand. 4927 4928You can declare variables and use them in function calls and returns, as 4929well as in assignments and some casts. You can specify a vector type as 4930a return type for a function. Vector types can also be used as function 4931arguments. It is possible to cast from one vector type to another, 4932provided they are of the same size (in fact, you can also cast vectors 4933to and from other datatypes of the same size). 4934 4935You cannot operate between vectors of different lengths or different 4936signedness without a cast. 4937 4938A port that supports hardware vector operations, usually provides a set 4939of built-in functions that can be used to operate on vectors. For 4940example, a function to add two vectors and multiply the result by a 4941third could look like this: 4942 4943@smallexample 4944v4si f (v4si a, v4si b, v4si c) 4945@{ 4946 v4si tmp = __builtin_addv4si (a, b); 4947 return __builtin_mulv4si (tmp, c); 4948@} 4949 4950@end smallexample 4951 4952@node Offsetof 4953@section Offsetof 4954@findex __builtin_offsetof 4955 4956GCC implements for both C and C++ a syntactic extension to implement 4957the @code{offsetof} macro. 4958 4959@smallexample 4960primary: 4961 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")" 4962 4963offsetof_member_designator: 4964 @code{identifier} 4965 | offsetof_member_designator "." @code{identifier} 4966 | offsetof_member_designator "[" @code{expr} "]" 4967@end smallexample 4968 4969This extension is sufficient such that 4970 4971@smallexample 4972#define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member}) 4973@end smallexample 4974 4975is a suitable definition of the @code{offsetof} macro. In C++, @var{type} 4976may be dependent. In either case, @var{member} may consist of a single 4977identifier, or a sequence of member accesses and array references. 4978 4979@node Atomic Builtins 4980@section Built-in functions for atomic memory access 4981 4982The following builtins are intended to be compatible with those described 4983in the @cite{Intel Itanium Processor-specific Application Binary Interface}, 4984section 7.4. As such, they depart from the normal GCC practice of using 4985the ``__builtin_'' prefix, and further that they are overloaded such that 4986they work on multiple types. 4987 4988The definition given in the Intel documentation allows only for the use of 4989the types @code{int}, @code{long}, @code{long long} as well as their unsigned 4990counterparts. GCC will allow any integral scalar or pointer type that is 49911, 2, 4 or 8 bytes in length. 4992 4993Not all operations are supported by all target processors. If a particular 4994operation cannot be implemented on the target processor, a warning will be 4995generated and a call an external function will be generated. The external 4996function will carry the same name as the builtin, with an additional suffix 4997@samp{_@var{n}} where @var{n} is the size of the data type. 4998 4999@c ??? Should we have a mechanism to suppress this warning? This is almost 5000@c useful for implementing the operation under the control of an external 5001@c mutex. 5002 5003In most cases, these builtins are considered a @dfn{full barrier}. That is, 5004no memory operand will be moved across the operation, either forward or 5005backward. Further, instructions will be issued as necessary to prevent the 5006processor from speculating loads across the operation and from queuing stores 5007after the operation. 5008 5009All of the routines are are described in the Intel documentation to take 5010``an optional list of variables protected by the memory barrier''. It's 5011not clear what is meant by that; it could mean that @emph{only} the 5012following variables are protected, or it could mean that these variables 5013should in addition be protected. At present GCC ignores this list and 5014protects all variables which are globally accessible. If in the future 5015we make some use of this list, an empty list will continue to mean all 5016globally accessible variables. 5017 5018@table @code 5019@item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...) 5020@itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...) 5021@itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...) 5022@itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...) 5023@itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...) 5024@itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...) 5025@findex __sync_fetch_and_add 5026@findex __sync_fetch_and_sub 5027@findex __sync_fetch_and_or 5028@findex __sync_fetch_and_and 5029@findex __sync_fetch_and_xor 5030@findex __sync_fetch_and_nand 5031These builtins perform the operation suggested by the name, and 5032returns the value that had previously been in memory. That is, 5033 5034@smallexample 5035@{ tmp = *ptr; *ptr @var{op}= value; return tmp; @} 5036@{ tmp = *ptr; *ptr = ~tmp & value; return tmp; @} // nand 5037@end smallexample 5038 5039@item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...) 5040@itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...) 5041@itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...) 5042@itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...) 5043@itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...) 5044@itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...) 5045@findex __sync_add_and_fetch 5046@findex __sync_sub_and_fetch 5047@findex __sync_or_and_fetch 5048@findex __sync_and_and_fetch 5049@findex __sync_xor_and_fetch 5050@findex __sync_nand_and_fetch 5051These builtins perform the operation suggested by the name, and 5052return the new value. That is, 5053 5054@smallexample 5055@{ *ptr @var{op}= value; return *ptr; @} 5056@{ *ptr = ~*ptr & value; return *ptr; @} // nand 5057@end smallexample 5058 5059@item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...) 5060@itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...) 5061@findex __sync_bool_compare_and_swap 5062@findex __sync_val_compare_and_swap 5063These builtins perform an atomic compare and swap. That is, if the current 5064value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into 5065@code{*@var{ptr}}. 5066 5067The ``bool'' version returns true if the comparison is successful and 5068@var{newval} was written. The ``val'' version returns the contents 5069of @code{*@var{ptr}} before the operation. 5070 5071@item __sync_synchronize (...) 5072@findex __sync_synchronize 5073This builtin issues a full memory barrier. 5074 5075@item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...) 5076@findex __sync_lock_test_and_set 5077This builtin, as described by Intel, is not a traditional test-and-set 5078operation, but rather an atomic exchange operation. It writes @var{value} 5079into @code{*@var{ptr}}, and returns the previous contents of 5080@code{*@var{ptr}}. 5081 5082Many targets have only minimal support for such locks, and do not support 5083a full exchange operation. In this case, a target may support reduced 5084functionality here by which the @emph{only} valid value to store is the 5085immediate constant 1. The exact value actually stored in @code{*@var{ptr}} 5086is implementation defined. 5087 5088This builtin is not a full barrier, but rather an @dfn{acquire barrier}. 5089This means that references after the builtin cannot move to (or be 5090speculated to) before the builtin, but previous memory stores may not 5091be globally visible yet, and previous memory loads may not yet be 5092satisfied. 5093 5094@item void __sync_lock_release (@var{type} *ptr, ...) 5095@findex __sync_lock_release 5096This builtin releases the lock acquired by @code{__sync_lock_test_and_set}. 5097Normally this means writing the constant 0 to @code{*@var{ptr}}. 5098 5099This builtin is not a full barrier, but rather a @dfn{release barrier}. 5100This means that all previous memory stores are globally visible, and all 5101previous memory loads have been satisfied, but following memory reads 5102are not prevented from being speculated to before the barrier. 5103@end table 5104 5105@node Object Size Checking 5106@section Object Size Checking Builtins 5107@findex __builtin_object_size 5108@findex __builtin___memcpy_chk 5109@findex __builtin___mempcpy_chk 5110@findex __builtin___memmove_chk 5111@findex __builtin___memset_chk 5112@findex __builtin___strcpy_chk 5113@findex __builtin___stpcpy_chk 5114@findex __builtin___strncpy_chk 5115@findex __builtin___strcat_chk 5116@findex __builtin___strncat_chk 5117@findex __builtin___sprintf_chk 5118@findex __builtin___snprintf_chk 5119@findex __builtin___vsprintf_chk 5120@findex __builtin___vsnprintf_chk 5121@findex __builtin___printf_chk 5122@findex __builtin___vprintf_chk 5123@findex __builtin___fprintf_chk 5124@findex __builtin___vfprintf_chk 5125 5126GCC implements a limited buffer overflow protection mechanism 5127that can prevent some buffer overflow attacks. 5128 5129@deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type}) 5130is a built-in construct that returns a constant number of bytes from 5131@var{ptr} to the end of the object @var{ptr} pointer points to 5132(if known at compile time). @code{__builtin_object_size} never evaluates 5133its arguments for side-effects. If there are any side-effects in them, it 5134returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0} 5135for @var{type} 2 or 3. If there are multiple objects @var{ptr} can 5136point to and all of them are known at compile time, the returned number 5137is the maximum of remaining byte counts in those objects if @var{type} & 2 is 51380 and minimum if nonzero. If it is not possible to determine which objects 5139@var{ptr} points to at compile time, @code{__builtin_object_size} should 5140return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0} 5141for @var{type} 2 or 3. 5142 5143@var{type} is an integer constant from 0 to 3. If the least significant 5144bit is clear, objects are whole variables, if it is set, a closest 5145surrounding subobject is considered the object a pointer points to. 5146The second bit determines if maximum or minimum of remaining bytes 5147is computed. 5148 5149@smallexample 5150struct V @{ char buf1[10]; int b; char buf2[10]; @} var; 5151char *p = &var.buf1[1], *q = &var.b; 5152 5153/* Here the object p points to is var. */ 5154assert (__builtin_object_size (p, 0) == sizeof (var) - 1); 5155/* The subobject p points to is var.buf1. */ 5156assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1); 5157/* The object q points to is var. */ 5158assert (__builtin_object_size (q, 0) 5159 == (char *) (&var + 1) - (char *) &var.b); 5160/* The subobject q points to is var.b. */ 5161assert (__builtin_object_size (q, 1) == sizeof (var.b)); 5162@end smallexample 5163@end deftypefn 5164 5165There are built-in functions added for many common string operation 5166functions, e.g. for @code{memcpy} @code{__builtin___memcpy_chk} 5167built-in is provided. This built-in has an additional last argument, 5168which is the number of bytes remaining in object the @var{dest} 5169argument points to or @code{(size_t) -1} if the size is not known. 5170 5171The built-in functions are optimized into the normal string functions 5172like @code{memcpy} if the last argument is @code{(size_t) -1} or if 5173it is known at compile time that the destination object will not 5174be overflown. If the compiler can determine at compile time the 5175object will be always overflown, it issues a warning. 5176 5177The intended use can be e.g. 5178 5179@smallexample 5180#undef memcpy 5181#define bos0(dest) __builtin_object_size (dest, 0) 5182#define memcpy(dest, src, n) \ 5183 __builtin___memcpy_chk (dest, src, n, bos0 (dest)) 5184 5185char *volatile p; 5186char buf[10]; 5187/* It is unknown what object p points to, so this is optimized 5188 into plain memcpy - no checking is possible. */ 5189memcpy (p, "abcde", n); 5190/* Destination is known and length too. It is known at compile 5191 time there will be no overflow. */ 5192memcpy (&buf[5], "abcde", 5); 5193/* Destination is known, but the length is not known at compile time. 5194 This will result in __memcpy_chk call that can check for overflow 5195 at runtime. */ 5196memcpy (&buf[5], "abcde", n); 5197/* Destination is known and it is known at compile time there will 5198 be overflow. There will be a warning and __memcpy_chk call that 5199 will abort the program at runtime. */ 5200memcpy (&buf[6], "abcde", 5); 5201@end smallexample 5202 5203Such built-in functions are provided for @code{memcpy}, @code{mempcpy}, 5204@code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy}, 5205@code{strcat} and @code{strncat}. 5206 5207There are also checking built-in functions for formatted output functions. 5208@smallexample 5209int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...); 5210int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os, 5211 const char *fmt, ...); 5212int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt, 5213 va_list ap); 5214int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os, 5215 const char *fmt, va_list ap); 5216@end smallexample 5217 5218The added @var{flag} argument is passed unchanged to @code{__sprintf_chk} 5219etc. functions and can contain implementation specific flags on what 5220additional security measures the checking function might take, such as 5221handling @code{%n} differently. 5222 5223The @var{os} argument is the object size @var{s} points to, like in the 5224other built-in functions. There is a small difference in the behavior 5225though, if @var{os} is @code{(size_t) -1}, the built-in functions are 5226optimized into the non-checking functions only if @var{flag} is 0, otherwise 5227the checking function is called with @var{os} argument set to 5228@code{(size_t) -1}. 5229 5230In addition to this, there are checking built-in functions 5231@code{__builtin___printf_chk}, @code{__builtin___vprintf_chk}, 5232@code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}. 5233These have just one additional argument, @var{flag}, right before 5234format string @var{fmt}. If the compiler is able to optimize them to 5235@code{fputc} etc. functions, it will, otherwise the checking function 5236should be called and the @var{flag} argument passed to it. 5237 5238@node Other Builtins 5239@section Other built-in functions provided by GCC 5240@cindex built-in functions 5241@findex __builtin_isgreater 5242@findex __builtin_isgreaterequal 5243@findex __builtin_isless 5244@findex __builtin_islessequal 5245@findex __builtin_islessgreater 5246@findex __builtin_isunordered 5247@findex __builtin_powi 5248@findex __builtin_powif 5249@findex __builtin_powil 5250@findex _Exit 5251@findex _exit 5252@findex abort 5253@findex abs 5254@findex acos 5255@findex acosf 5256@findex acosh 5257@findex acoshf 5258@findex acoshl 5259@findex acosl 5260@findex alloca 5261@findex asin 5262@findex asinf 5263@findex asinh 5264@findex asinhf 5265@findex asinhl 5266@findex asinl 5267@findex atan 5268@findex atan2 5269@findex atan2f 5270@findex atan2l 5271@findex atanf 5272@findex atanh 5273@findex atanhf 5274@findex atanhl 5275@findex atanl 5276@findex bcmp 5277@findex bzero 5278@findex cabs 5279@findex cabsf 5280@findex cabsl 5281@findex cacos 5282@findex cacosf 5283@findex cacosh 5284@findex cacoshf 5285@findex cacoshl 5286@findex cacosl 5287@findex calloc 5288@findex carg 5289@findex cargf 5290@findex cargl 5291@findex casin 5292@findex casinf 5293@findex casinh 5294@findex casinhf 5295@findex casinhl 5296@findex casinl 5297@findex catan 5298@findex catanf 5299@findex catanh 5300@findex catanhf 5301@findex catanhl 5302@findex catanl 5303@findex cbrt 5304@findex cbrtf 5305@findex cbrtl 5306@findex ccos 5307@findex ccosf 5308@findex ccosh 5309@findex ccoshf 5310@findex ccoshl 5311@findex ccosl 5312@findex ceil 5313@findex ceilf 5314@findex ceill 5315@findex cexp 5316@findex cexpf 5317@findex cexpl 5318@findex cimag 5319@findex cimagf 5320@findex cimagl 5321@findex clog 5322@findex clogf 5323@findex clogl 5324@findex conj 5325@findex conjf 5326@findex conjl 5327@findex copysign 5328@findex copysignf 5329@findex copysignl 5330@findex cos 5331@findex cosf 5332@findex cosh 5333@findex coshf 5334@findex coshl 5335@findex cosl 5336@findex cpow 5337@findex cpowf 5338@findex cpowl 5339@findex cproj 5340@findex cprojf 5341@findex cprojl 5342@findex creal 5343@findex crealf 5344@findex creall 5345@findex csin 5346@findex csinf 5347@findex csinh 5348@findex csinhf 5349@findex csinhl 5350@findex csinl 5351@findex csqrt 5352@findex csqrtf 5353@findex csqrtl 5354@findex ctan 5355@findex ctanf 5356@findex ctanh 5357@findex ctanhf 5358@findex ctanhl 5359@findex ctanl 5360@findex dcgettext 5361@findex dgettext 5362@findex drem 5363@findex dremf 5364@findex dreml 5365@findex erf 5366@findex erfc 5367@findex erfcf 5368@findex erfcl 5369@findex erff 5370@findex erfl 5371@findex exit 5372@findex exp 5373@findex exp10 5374@findex exp10f 5375@findex exp10l 5376@findex exp2 5377@findex exp2f 5378@findex exp2l 5379@findex expf 5380@findex expl 5381@findex expm1 5382@findex expm1f 5383@findex expm1l 5384@findex fabs 5385@findex fabsf 5386@findex fabsl 5387@findex fdim 5388@findex fdimf 5389@findex fdiml 5390@findex ffs 5391@findex floor 5392@findex floorf 5393@findex floorl 5394@findex fma 5395@findex fmaf 5396@findex fmal 5397@findex fmax 5398@findex fmaxf 5399@findex fmaxl 5400@findex fmin 5401@findex fminf 5402@findex fminl 5403@findex fmod 5404@findex fmodf 5405@findex fmodl 5406@findex fprintf 5407@findex fprintf_unlocked 5408@findex fputs 5409@findex fputs_unlocked 5410@findex frexp 5411@findex frexpf 5412@findex frexpl 5413@findex fscanf 5414@findex gamma 5415@findex gammaf 5416@findex gammal 5417@findex gettext 5418@findex hypot 5419@findex hypotf 5420@findex hypotl 5421@findex ilogb 5422@findex ilogbf 5423@findex ilogbl 5424@findex imaxabs 5425@findex index 5426@findex isalnum 5427@findex isalpha 5428@findex isascii 5429@findex isblank 5430@findex iscntrl 5431@findex isdigit 5432@findex isgraph 5433@findex islower 5434@findex isprint 5435@findex ispunct 5436@findex isspace 5437@findex isupper 5438@findex iswalnum 5439@findex iswalpha 5440@findex iswblank 5441@findex iswcntrl 5442@findex iswdigit 5443@findex iswgraph 5444@findex iswlower 5445@findex iswprint 5446@findex iswpunct 5447@findex iswspace 5448@findex iswupper 5449@findex iswxdigit 5450@findex isxdigit 5451@findex j0 5452@findex j0f 5453@findex j0l 5454@findex j1 5455@findex j1f 5456@findex j1l 5457@findex jn 5458@findex jnf 5459@findex jnl 5460@findex labs 5461@findex ldexp 5462@findex ldexpf 5463@findex ldexpl 5464@findex lgamma 5465@findex lgammaf 5466@findex lgammal 5467@findex llabs 5468@findex llrint 5469@findex llrintf 5470@findex llrintl 5471@findex llround 5472@findex llroundf 5473@findex llroundl 5474@findex log 5475@findex log10 5476@findex log10f 5477@findex log10l 5478@findex log1p 5479@findex log1pf 5480@findex log1pl 5481@findex log2 5482@findex log2f 5483@findex log2l 5484@findex logb 5485@findex logbf 5486@findex logbl 5487@findex logf 5488@findex logl 5489@findex lrint 5490@findex lrintf 5491@findex lrintl 5492@findex lround 5493@findex lroundf 5494@findex lroundl 5495@findex malloc 5496@findex memcmp 5497@findex memcpy 5498@findex mempcpy 5499@findex memset 5500@findex modf 5501@findex modff 5502@findex modfl 5503@findex nearbyint 5504@findex nearbyintf 5505@findex nearbyintl 5506@findex nextafter 5507@findex nextafterf 5508@findex nextafterl 5509@findex nexttoward 5510@findex nexttowardf 5511@findex nexttowardl 5512@findex pow 5513@findex pow10 5514@findex pow10f 5515@findex pow10l 5516@findex powf 5517@findex powl 5518@findex printf 5519@findex printf_unlocked 5520@findex putchar 5521@findex puts 5522@findex remainder 5523@findex remainderf 5524@findex remainderl 5525@findex remquo 5526@findex remquof 5527@findex remquol 5528@findex rindex 5529@findex rint 5530@findex rintf 5531@findex rintl 5532@findex round 5533@findex roundf 5534@findex roundl 5535@findex scalb 5536@findex scalbf 5537@findex scalbl 5538@findex scalbln 5539@findex scalblnf 5540@findex scalblnf 5541@findex scalbn 5542@findex scalbnf 5543@findex scanfnl 5544@findex signbit 5545@findex signbitf 5546@findex signbitl 5547@findex significand 5548@findex significandf 5549@findex significandl 5550@findex sin 5551@findex sincos 5552@findex sincosf 5553@findex sincosl 5554@findex sinf 5555@findex sinh 5556@findex sinhf 5557@findex sinhl 5558@findex sinl 5559@findex snprintf 5560@findex sprintf 5561@findex sqrt 5562@findex sqrtf 5563@findex sqrtl 5564@findex sscanf 5565@findex stpcpy 5566@findex stpncpy 5567@findex strcasecmp 5568@findex strcat 5569@findex strchr 5570@findex strcmp 5571@findex strcpy 5572@findex strcspn 5573@findex strdup 5574@findex strfmon 5575@findex strftime 5576@findex strlen 5577@findex strncasecmp 5578@findex strncat 5579@findex strncmp 5580@findex strncpy 5581@findex strndup 5582@findex strpbrk 5583@findex strrchr 5584@findex strspn 5585@findex strstr 5586@findex tan 5587@findex tanf 5588@findex tanh 5589@findex tanhf 5590@findex tanhl 5591@findex tanl 5592@findex tgamma 5593@findex tgammaf 5594@findex tgammal 5595@findex toascii 5596@findex tolower 5597@findex toupper 5598@findex towlower 5599@findex towupper 5600@findex trunc 5601@findex truncf 5602@findex truncl 5603@findex vfprintf 5604@findex vfscanf 5605@findex vprintf 5606@findex vscanf 5607@findex vsnprintf 5608@findex vsprintf 5609@findex vsscanf 5610@findex y0 5611@findex y0f 5612@findex y0l 5613@findex y1 5614@findex y1f 5615@findex y1l 5616@findex yn 5617@findex ynf 5618@findex ynl 5619 5620GCC provides a large number of built-in functions other than the ones 5621mentioned above. Some of these are for internal use in the processing 5622of exceptions or variable-length argument lists and will not be 5623documented here because they may change from time to time; we do not 5624recommend general use of these functions. 5625 5626The remaining functions are provided for optimization purposes. 5627 5628@opindex fno-builtin 5629GCC includes built-in versions of many of the functions in the standard 5630C library. The versions prefixed with @code{__builtin_} will always be 5631treated as having the same meaning as the C library function even if you 5632specify the @option{-fno-builtin} option. (@pxref{C Dialect Options}) 5633Many of these functions are only optimized in certain cases; if they are 5634not optimized in a particular case, a call to the library function will 5635be emitted. 5636 5637@opindex ansi 5638@opindex std 5639Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or 5640@option{-std=c99}), the functions 5641@code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero}, 5642@code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml}, 5643@code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll}, 5644@code{ffsl}, @code{ffs}, @code{fprintf_unlocked}, @code{fputs_unlocked}, 5645@code{gammaf}, @code{gammal}, @code{gamma}, @code{gettext}, 5646@code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0}, 5647@code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn}, 5648@code{mempcpy}, @code{pow10f}, @code{pow10l}, @code{pow10}, 5649@code{printf_unlocked}, @code{rindex}, @code{scalbf}, @code{scalbl}, 5650@code{scalb}, @code{signbit}, @code{signbitf}, @code{signbitl}, 5651@code{significandf}, @code{significandl}, @code{significand}, 5652@code{sincosf}, @code{sincosl}, @code{sincos}, @code{stpcpy}, 5653@code{stpncpy}, @code{strcasecmp}, @code{strdup}, @code{strfmon}, 5654@code{strncasecmp}, @code{strndup}, @code{toascii}, @code{y0f}, 5655@code{y0l}, @code{y0}, @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, 5656@code{ynl} and @code{yn} 5657may be handled as built-in functions. 5658All these functions have corresponding versions 5659prefixed with @code{__builtin_}, which may be used even in strict C89 5660mode. 5661 5662The ISO C99 functions 5663@code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf}, 5664@code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh}, 5665@code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf}, 5666@code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos}, 5667@code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf}, 5668@code{casinhl}, @code{casinh}, @code{casinl}, @code{casin}, 5669@code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh}, 5670@code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt}, 5671@code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl}, 5672@code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf}, 5673@code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog}, 5674@code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl}, 5675@code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf}, 5676@code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal}, 5677@code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl}, 5678@code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf}, 5679@code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan}, 5680@code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl}, 5681@code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f}, 5682@code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim}, 5683@code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax}, 5684@code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf}, 5685@code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb}, 5686@code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf}, 5687@code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl}, 5688@code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround}, 5689@code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l}, 5690@code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf}, 5691@code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl}, 5692@code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint}, 5693@code{nextafterf}, @code{nextafterl}, @code{nextafter}, 5694@code{nexttowardf}, @code{nexttowardl}, @code{nexttoward}, 5695@code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof}, 5696@code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint}, 5697@code{roundf}, @code{roundl}, @code{round}, @code{scalblnf}, 5698@code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl}, 5699@code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal}, 5700@code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc}, 5701@code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf} 5702are handled as built-in functions 5703except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}). 5704 5705There are also built-in versions of the ISO C99 functions 5706@code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f}, 5707@code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill}, 5708@code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf}, 5709@code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl}, 5710@code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf}, 5711@code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl}, 5712@code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf}, 5713@code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl}, 5714@code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl} 5715that are recognized in any mode since ISO C90 reserves these names for 5716the purpose to which ISO C99 puts them. All these functions have 5717corresponding versions prefixed with @code{__builtin_}. 5718 5719The ISO C94 functions 5720@code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit}, 5721@code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct}, 5722@code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and 5723@code{towupper} 5724are handled as built-in functions 5725except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}). 5726 5727The ISO C90 functions 5728@code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2}, 5729@code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos}, 5730@code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod}, 5731@code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf}, 5732@code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit}, 5733@code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct}, 5734@code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower}, 5735@code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log}, 5736@code{malloc}, @code{memcmp}, @code{memcpy}, @code{memset}, @code{modf}, 5737@code{pow}, @code{printf}, @code{putchar}, @code{puts}, @code{scanf}, 5738@code{sinh}, @code{sin}, @code{snprintf}, @code{sprintf}, @code{sqrt}, 5739@code{sscanf}, @code{strcat}, @code{strchr}, @code{strcmp}, 5740@code{strcpy}, @code{strcspn}, @code{strlen}, @code{strncat}, 5741@code{strncmp}, @code{strncpy}, @code{strpbrk}, @code{strrchr}, 5742@code{strspn}, @code{strstr}, @code{tanh}, @code{tan}, @code{vfprintf}, 5743@code{vprintf} and @code{vsprintf} 5744are all recognized as built-in functions unless 5745@option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}} 5746is specified for an individual function). All of these functions have 5747corresponding versions prefixed with @code{__builtin_}. 5748 5749GCC provides built-in versions of the ISO C99 floating point comparison 5750macros that avoid raising exceptions for unordered operands. They have 5751the same names as the standard macros ( @code{isgreater}, 5752@code{isgreaterequal}, @code{isless}, @code{islessequal}, 5753@code{islessgreater}, and @code{isunordered}) , with @code{__builtin_} 5754prefixed. We intend for a library implementor to be able to simply 5755@code{#define} each standard macro to its built-in equivalent. 5756 5757@deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2}) 5758 5759You can use the built-in function @code{__builtin_types_compatible_p} to 5760determine whether two types are the same. 5761 5762This built-in function returns 1 if the unqualified versions of the 5763types @var{type1} and @var{type2} (which are types, not expressions) are 5764compatible, 0 otherwise. The result of this built-in function can be 5765used in integer constant expressions. 5766 5767This built-in function ignores top level qualifiers (e.g., @code{const}, 5768@code{volatile}). For example, @code{int} is equivalent to @code{const 5769int}. 5770 5771The type @code{int[]} and @code{int[5]} are compatible. On the other 5772hand, @code{int} and @code{char *} are not compatible, even if the size 5773of their types, on the particular architecture are the same. Also, the 5774amount of pointer indirection is taken into account when determining 5775similarity. Consequently, @code{short *} is not similar to 5776@code{short **}. Furthermore, two types that are typedefed are 5777considered compatible if their underlying types are compatible. 5778 5779An @code{enum} type is not considered to be compatible with another 5780@code{enum} type even if both are compatible with the same integer 5781type; this is what the C standard specifies. 5782For example, @code{enum @{foo, bar@}} is not similar to 5783@code{enum @{hot, dog@}}. 5784 5785You would typically use this function in code whose execution varies 5786depending on the arguments' types. For example: 5787 5788@smallexample 5789#define foo(x) \ 5790 (@{ \ 5791 typeof (x) tmp = (x); \ 5792 if (__builtin_types_compatible_p (typeof (x), long double)) \ 5793 tmp = foo_long_double (tmp); \ 5794 else if (__builtin_types_compatible_p (typeof (x), double)) \ 5795 tmp = foo_double (tmp); \ 5796 else if (__builtin_types_compatible_p (typeof (x), float)) \ 5797 tmp = foo_float (tmp); \ 5798 else \ 5799 abort (); \ 5800 tmp; \ 5801 @}) 5802@end smallexample 5803 5804@emph{Note:} This construct is only available for C@. 5805 5806@end deftypefn 5807 5808@deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2}) 5809 5810You can use the built-in function @code{__builtin_choose_expr} to 5811evaluate code depending on the value of a constant expression. This 5812built-in function returns @var{exp1} if @var{const_exp}, which is a 5813constant expression that must be able to be determined at compile time, 5814is nonzero. Otherwise it returns 0. 5815 5816This built-in function is analogous to the @samp{? :} operator in C, 5817except that the expression returned has its type unaltered by promotion 5818rules. Also, the built-in function does not evaluate the expression 5819that was not chosen. For example, if @var{const_exp} evaluates to true, 5820@var{exp2} is not evaluated even if it has side-effects. 5821 5822This built-in function can return an lvalue if the chosen argument is an 5823lvalue. 5824 5825If @var{exp1} is returned, the return type is the same as @var{exp1}'s 5826type. Similarly, if @var{exp2} is returned, its return type is the same 5827as @var{exp2}. 5828 5829Example: 5830 5831@smallexample 5832#define foo(x) \ 5833 __builtin_choose_expr ( \ 5834 __builtin_types_compatible_p (typeof (x), double), \ 5835 foo_double (x), \ 5836 __builtin_choose_expr ( \ 5837 __builtin_types_compatible_p (typeof (x), float), \ 5838 foo_float (x), \ 5839 /* @r{The void expression results in a compile-time error} \ 5840 @r{when assigning the result to something.} */ \ 5841 (void)0)) 5842@end smallexample 5843 5844@emph{Note:} This construct is only available for C@. Furthermore, the 5845unused expression (@var{exp1} or @var{exp2} depending on the value of 5846@var{const_exp}) may still generate syntax errors. This may change in 5847future revisions. 5848 5849@end deftypefn 5850 5851@deftypefn {Built-in Function} int __builtin_constant_p (@var{exp}) 5852You can use the built-in function @code{__builtin_constant_p} to 5853determine if a value is known to be constant at compile-time and hence 5854that GCC can perform constant-folding on expressions involving that 5855value. The argument of the function is the value to test. The function 5856returns the integer 1 if the argument is known to be a compile-time 5857constant and 0 if it is not known to be a compile-time constant. A 5858return of 0 does not indicate that the value is @emph{not} a constant, 5859but merely that GCC cannot prove it is a constant with the specified 5860value of the @option{-O} option. 5861 5862You would typically use this function in an embedded application where 5863memory was a critical resource. If you have some complex calculation, 5864you may want it to be folded if it involves constants, but need to call 5865a function if it does not. For example: 5866 5867@smallexample 5868#define Scale_Value(X) \ 5869 (__builtin_constant_p (X) \ 5870 ? ((X) * SCALE + OFFSET) : Scale (X)) 5871@end smallexample 5872 5873You may use this built-in function in either a macro or an inline 5874function. However, if you use it in an inlined function and pass an 5875argument of the function as the argument to the built-in, GCC will 5876never return 1 when you call the inline function with a string constant 5877or compound literal (@pxref{Compound Literals}) and will not return 1 5878when you pass a constant numeric value to the inline function unless you 5879specify the @option{-O} option. 5880 5881You may also use @code{__builtin_constant_p} in initializers for static 5882data. For instance, you can write 5883 5884@smallexample 5885static const int table[] = @{ 5886 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1, 5887 /* @r{@dots{}} */ 5888@}; 5889@end smallexample 5890 5891@noindent 5892This is an acceptable initializer even if @var{EXPRESSION} is not a 5893constant expression. GCC must be more conservative about evaluating the 5894built-in in this case, because it has no opportunity to perform 5895optimization. 5896 5897Previous versions of GCC did not accept this built-in in data 5898initializers. The earliest version where it is completely safe is 58993.0.1. 5900@end deftypefn 5901 5902@deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c}) 5903@opindex fprofile-arcs 5904You may use @code{__builtin_expect} to provide the compiler with 5905branch prediction information. In general, you should prefer to 5906use actual profile feedback for this (@option{-fprofile-arcs}), as 5907programmers are notoriously bad at predicting how their programs 5908actually perform. However, there are applications in which this 5909data is hard to collect. 5910 5911The return value is the value of @var{exp}, which should be an 5912integral expression. The value of @var{c} must be a compile-time 5913constant. The semantics of the built-in are that it is expected 5914that @var{exp} == @var{c}. For example: 5915 5916@smallexample 5917if (__builtin_expect (x, 0)) 5918 foo (); 5919@end smallexample 5920 5921@noindent 5922would indicate that we do not expect to call @code{foo}, since 5923we expect @code{x} to be zero. Since you are limited to integral 5924expressions for @var{exp}, you should use constructions such as 5925 5926@smallexample 5927if (__builtin_expect (ptr != NULL, 1)) 5928 error (); 5929@end smallexample 5930 5931@noindent 5932when testing pointer or floating-point values. 5933@end deftypefn 5934 5935@deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...) 5936This function is used to minimize cache-miss latency by moving data into 5937a cache before it is accessed. 5938You can insert calls to @code{__builtin_prefetch} into code for which 5939you know addresses of data in memory that is likely to be accessed soon. 5940If the target supports them, data prefetch instructions will be generated. 5941If the prefetch is done early enough before the access then the data will 5942be in the cache by the time it is accessed. 5943 5944The value of @var{addr} is the address of the memory to prefetch. 5945There are two optional arguments, @var{rw} and @var{locality}. 5946The value of @var{rw} is a compile-time constant one or zero; one 5947means that the prefetch is preparing for a write to the memory address 5948and zero, the default, means that the prefetch is preparing for a read. 5949The value @var{locality} must be a compile-time constant integer between 5950zero and three. A value of zero means that the data has no temporal 5951locality, so it need not be left in the cache after the access. A value 5952of three means that the data has a high degree of temporal locality and 5953should be left in all levels of cache possible. Values of one and two 5954mean, respectively, a low or moderate degree of temporal locality. The 5955default is three. 5956 5957@smallexample 5958for (i = 0; i < n; i++) 5959 @{ 5960 a[i] = a[i] + b[i]; 5961 __builtin_prefetch (&a[i+j], 1, 1); 5962 __builtin_prefetch (&b[i+j], 0, 1); 5963 /* @r{@dots{}} */ 5964 @} 5965@end smallexample 5966 5967Data prefetch does not generate faults if @var{addr} is invalid, but 5968the address expression itself must be valid. For example, a prefetch 5969of @code{p->next} will not fault if @code{p->next} is not a valid 5970address, but evaluation will fault if @code{p} is not a valid address. 5971 5972If the target does not support data prefetch, the address expression 5973is evaluated if it includes side effects but no other code is generated 5974and GCC does not issue a warning. 5975@end deftypefn 5976 5977@deftypefn {Built-in Function} double __builtin_huge_val (void) 5978Returns a positive infinity, if supported by the floating-point format, 5979else @code{DBL_MAX}. This function is suitable for implementing the 5980ISO C macro @code{HUGE_VAL}. 5981@end deftypefn 5982 5983@deftypefn {Built-in Function} float __builtin_huge_valf (void) 5984Similar to @code{__builtin_huge_val}, except the return type is @code{float}. 5985@end deftypefn 5986 5987@deftypefn {Built-in Function} {long double} __builtin_huge_vall (void) 5988Similar to @code{__builtin_huge_val}, except the return 5989type is @code{long double}. 5990@end deftypefn 5991 5992@deftypefn {Built-in Function} double __builtin_inf (void) 5993Similar to @code{__builtin_huge_val}, except a warning is generated 5994if the target floating-point format does not support infinities. 5995@end deftypefn 5996 5997@deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void) 5998Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}. 5999@end deftypefn 6000 6001@deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void) 6002Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}. 6003@end deftypefn 6004 6005@deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void) 6006Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}. 6007@end deftypefn 6008 6009@deftypefn {Built-in Function} float __builtin_inff (void) 6010Similar to @code{__builtin_inf}, except the return type is @code{float}. 6011This function is suitable for implementing the ISO C99 macro @code{INFINITY}. 6012@end deftypefn 6013 6014@deftypefn {Built-in Function} {long double} __builtin_infl (void) 6015Similar to @code{__builtin_inf}, except the return 6016type is @code{long double}. 6017@end deftypefn 6018 6019@deftypefn {Built-in Function} double __builtin_nan (const char *str) 6020This is an implementation of the ISO C99 function @code{nan}. 6021 6022Since ISO C99 defines this function in terms of @code{strtod}, which we 6023do not implement, a description of the parsing is in order. The string 6024is parsed as by @code{strtol}; that is, the base is recognized by 6025leading @samp{0} or @samp{0x} prefixes. The number parsed is placed 6026in the significand such that the least significant bit of the number 6027is at the least significant bit of the significand. The number is 6028truncated to fit the significand field provided. The significand is 6029forced to be a quiet NaN@. 6030 6031This function, if given a string literal all of which would have been 6032consumed by strtol, is evaluated early enough that it is considered a 6033compile-time constant. 6034@end deftypefn 6035 6036@deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str) 6037Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}. 6038@end deftypefn 6039 6040@deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str) 6041Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}. 6042@end deftypefn 6043 6044@deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str) 6045Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}. 6046@end deftypefn 6047 6048@deftypefn {Built-in Function} float __builtin_nanf (const char *str) 6049Similar to @code{__builtin_nan}, except the return type is @code{float}. 6050@end deftypefn 6051 6052@deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str) 6053Similar to @code{__builtin_nan}, except the return type is @code{long double}. 6054@end deftypefn 6055 6056@deftypefn {Built-in Function} double __builtin_nans (const char *str) 6057Similar to @code{__builtin_nan}, except the significand is forced 6058to be a signaling NaN@. The @code{nans} function is proposed by 6059@uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}. 6060@end deftypefn 6061 6062@deftypefn {Built-in Function} float __builtin_nansf (const char *str) 6063Similar to @code{__builtin_nans}, except the return type is @code{float}. 6064@end deftypefn 6065 6066@deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str) 6067Similar to @code{__builtin_nans}, except the return type is @code{long double}. 6068@end deftypefn 6069 6070@deftypefn {Built-in Function} int __builtin_ffs (unsigned int x) 6071Returns one plus the index of the least significant 1-bit of @var{x}, or 6072if @var{x} is zero, returns zero. 6073@end deftypefn 6074 6075@deftypefn {Built-in Function} int __builtin_clz (unsigned int x) 6076Returns the number of leading 0-bits in @var{x}, starting at the most 6077significant bit position. If @var{x} is 0, the result is undefined. 6078@end deftypefn 6079 6080@deftypefn {Built-in Function} int __builtin_ctz (unsigned int x) 6081Returns the number of trailing 0-bits in @var{x}, starting at the least 6082significant bit position. If @var{x} is 0, the result is undefined. 6083@end deftypefn 6084 6085@deftypefn {Built-in Function} int __builtin_popcount (unsigned int x) 6086Returns the number of 1-bits in @var{x}. 6087@end deftypefn 6088 6089@deftypefn {Built-in Function} int __builtin_parity (unsigned int x) 6090Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x} 6091modulo 2. 6092@end deftypefn 6093 6094@deftypefn {Built-in Function} int __builtin_ffsl (unsigned long) 6095Similar to @code{__builtin_ffs}, except the argument type is 6096@code{unsigned long}. 6097@end deftypefn 6098 6099@deftypefn {Built-in Function} int __builtin_clzl (unsigned long) 6100Similar to @code{__builtin_clz}, except the argument type is 6101@code{unsigned long}. 6102@end deftypefn 6103 6104@deftypefn {Built-in Function} int __builtin_ctzl (unsigned long) 6105Similar to @code{__builtin_ctz}, except the argument type is 6106@code{unsigned long}. 6107@end deftypefn 6108 6109@deftypefn {Built-in Function} int __builtin_popcountl (unsigned long) 6110Similar to @code{__builtin_popcount}, except the argument type is 6111@code{unsigned long}. 6112@end deftypefn 6113 6114@deftypefn {Built-in Function} int __builtin_parityl (unsigned long) 6115Similar to @code{__builtin_parity}, except the argument type is 6116@code{unsigned long}. 6117@end deftypefn 6118 6119@deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long) 6120Similar to @code{__builtin_ffs}, except the argument type is 6121@code{unsigned long long}. 6122@end deftypefn 6123 6124@deftypefn {Built-in Function} int __builtin_clzll (unsigned long long) 6125Similar to @code{__builtin_clz}, except the argument type is 6126@code{unsigned long long}. 6127@end deftypefn 6128 6129@deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long) 6130Similar to @code{__builtin_ctz}, except the argument type is 6131@code{unsigned long long}. 6132@end deftypefn 6133 6134@deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long) 6135Similar to @code{__builtin_popcount}, except the argument type is 6136@code{unsigned long long}. 6137@end deftypefn 6138 6139@deftypefn {Built-in Function} int __builtin_parityll (unsigned long long) 6140Similar to @code{__builtin_parity}, except the argument type is 6141@code{unsigned long long}. 6142@end deftypefn 6143 6144@deftypefn {Built-in Function} double __builtin_powi (double, int) 6145Returns the first argument raised to the power of the second. Unlike the 6146@code{pow} function no guarantees about precision and rounding are made. 6147@end deftypefn 6148 6149@deftypefn {Built-in Function} float __builtin_powif (float, int) 6150Similar to @code{__builtin_powi}, except the argument and return types 6151are @code{float}. 6152@end deftypefn 6153 6154@deftypefn {Built-in Function} {long double} __builtin_powil (long double, int) 6155Similar to @code{__builtin_powi}, except the argument and return types 6156are @code{long double}. 6157@end deftypefn 6158 6159 6160@node Target Builtins 6161@section Built-in Functions Specific to Particular Target Machines 6162 6163On some target machines, GCC supports many built-in functions specific 6164to those machines. Generally these generate calls to specific machine 6165instructions, but allow the compiler to schedule those calls. 6166 6167@menu 6168* Alpha Built-in Functions:: 6169* ARM Built-in Functions:: 6170* Blackfin Built-in Functions:: 6171* FR-V Built-in Functions:: 6172* X86 Built-in Functions:: 6173* MIPS DSP Built-in Functions:: 6174* MIPS Paired-Single Support:: 6175* PowerPC AltiVec Built-in Functions:: 6176* SPARC VIS Built-in Functions:: 6177@end menu 6178 6179@node Alpha Built-in Functions 6180@subsection Alpha Built-in Functions 6181 6182These built-in functions are available for the Alpha family of 6183processors, depending on the command-line switches used. 6184 6185The following built-in functions are always available. They 6186all generate the machine instruction that is part of the name. 6187 6188@smallexample 6189long __builtin_alpha_implver (void) 6190long __builtin_alpha_rpcc (void) 6191long __builtin_alpha_amask (long) 6192long __builtin_alpha_cmpbge (long, long) 6193long __builtin_alpha_extbl (long, long) 6194long __builtin_alpha_extwl (long, long) 6195long __builtin_alpha_extll (long, long) 6196long __builtin_alpha_extql (long, long) 6197long __builtin_alpha_extwh (long, long) 6198long __builtin_alpha_extlh (long, long) 6199long __builtin_alpha_extqh (long, long) 6200long __builtin_alpha_insbl (long, long) 6201long __builtin_alpha_inswl (long, long) 6202long __builtin_alpha_insll (long, long) 6203long __builtin_alpha_insql (long, long) 6204long __builtin_alpha_inswh (long, long) 6205long __builtin_alpha_inslh (long, long) 6206long __builtin_alpha_insqh (long, long) 6207long __builtin_alpha_mskbl (long, long) 6208long __builtin_alpha_mskwl (long, long) 6209long __builtin_alpha_mskll (long, long) 6210long __builtin_alpha_mskql (long, long) 6211long __builtin_alpha_mskwh (long, long) 6212long __builtin_alpha_msklh (long, long) 6213long __builtin_alpha_mskqh (long, long) 6214long __builtin_alpha_umulh (long, long) 6215long __builtin_alpha_zap (long, long) 6216long __builtin_alpha_zapnot (long, long) 6217@end smallexample 6218 6219The following built-in functions are always with @option{-mmax} 6220or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or 6221later. They all generate the machine instruction that is part 6222of the name. 6223 6224@smallexample 6225long __builtin_alpha_pklb (long) 6226long __builtin_alpha_pkwb (long) 6227long __builtin_alpha_unpkbl (long) 6228long __builtin_alpha_unpkbw (long) 6229long __builtin_alpha_minub8 (long, long) 6230long __builtin_alpha_minsb8 (long, long) 6231long __builtin_alpha_minuw4 (long, long) 6232long __builtin_alpha_minsw4 (long, long) 6233long __builtin_alpha_maxub8 (long, long) 6234long __builtin_alpha_maxsb8 (long, long) 6235long __builtin_alpha_maxuw4 (long, long) 6236long __builtin_alpha_maxsw4 (long, long) 6237long __builtin_alpha_perr (long, long) 6238@end smallexample 6239 6240The following built-in functions are always with @option{-mcix} 6241or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or 6242later. They all generate the machine instruction that is part 6243of the name. 6244 6245@smallexample 6246long __builtin_alpha_cttz (long) 6247long __builtin_alpha_ctlz (long) 6248long __builtin_alpha_ctpop (long) 6249@end smallexample 6250 6251The following builtins are available on systems that use the OSF/1 6252PALcode. Normally they invoke the @code{rduniq} and @code{wruniq} 6253PAL calls, but when invoked with @option{-mtls-kernel}, they invoke 6254@code{rdval} and @code{wrval}. 6255 6256@smallexample 6257void *__builtin_thread_pointer (void) 6258void __builtin_set_thread_pointer (void *) 6259@end smallexample 6260 6261@node ARM Built-in Functions 6262@subsection ARM Built-in Functions 6263 6264These built-in functions are available for the ARM family of 6265processors, when the @option{-mcpu=iwmmxt} switch is used: 6266 6267@smallexample 6268typedef int v2si __attribute__ ((vector_size (8))); 6269typedef short v4hi __attribute__ ((vector_size (8))); 6270typedef char v8qi __attribute__ ((vector_size (8))); 6271 6272int __builtin_arm_getwcx (int) 6273void __builtin_arm_setwcx (int, int) 6274int __builtin_arm_textrmsb (v8qi, int) 6275int __builtin_arm_textrmsh (v4hi, int) 6276int __builtin_arm_textrmsw (v2si, int) 6277int __builtin_arm_textrmub (v8qi, int) 6278int __builtin_arm_textrmuh (v4hi, int) 6279int __builtin_arm_textrmuw (v2si, int) 6280v8qi __builtin_arm_tinsrb (v8qi, int) 6281v4hi __builtin_arm_tinsrh (v4hi, int) 6282v2si __builtin_arm_tinsrw (v2si, int) 6283long long __builtin_arm_tmia (long long, int, int) 6284long long __builtin_arm_tmiabb (long long, int, int) 6285long long __builtin_arm_tmiabt (long long, int, int) 6286long long __builtin_arm_tmiaph (long long, int, int) 6287long long __builtin_arm_tmiatb (long long, int, int) 6288long long __builtin_arm_tmiatt (long long, int, int) 6289int __builtin_arm_tmovmskb (v8qi) 6290int __builtin_arm_tmovmskh (v4hi) 6291int __builtin_arm_tmovmskw (v2si) 6292long long __builtin_arm_waccb (v8qi) 6293long long __builtin_arm_wacch (v4hi) 6294long long __builtin_arm_waccw (v2si) 6295v8qi __builtin_arm_waddb (v8qi, v8qi) 6296v8qi __builtin_arm_waddbss (v8qi, v8qi) 6297v8qi __builtin_arm_waddbus (v8qi, v8qi) 6298v4hi __builtin_arm_waddh (v4hi, v4hi) 6299v4hi __builtin_arm_waddhss (v4hi, v4hi) 6300v4hi __builtin_arm_waddhus (v4hi, v4hi) 6301v2si __builtin_arm_waddw (v2si, v2si) 6302v2si __builtin_arm_waddwss (v2si, v2si) 6303v2si __builtin_arm_waddwus (v2si, v2si) 6304v8qi __builtin_arm_walign (v8qi, v8qi, int) 6305long long __builtin_arm_wand(long long, long long) 6306long long __builtin_arm_wandn (long long, long long) 6307v8qi __builtin_arm_wavg2b (v8qi, v8qi) 6308v8qi __builtin_arm_wavg2br (v8qi, v8qi) 6309v4hi __builtin_arm_wavg2h (v4hi, v4hi) 6310v4hi __builtin_arm_wavg2hr (v4hi, v4hi) 6311v8qi __builtin_arm_wcmpeqb (v8qi, v8qi) 6312v4hi __builtin_arm_wcmpeqh (v4hi, v4hi) 6313v2si __builtin_arm_wcmpeqw (v2si, v2si) 6314v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi) 6315v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi) 6316v2si __builtin_arm_wcmpgtsw (v2si, v2si) 6317v8qi __builtin_arm_wcmpgtub (v8qi, v8qi) 6318v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi) 6319v2si __builtin_arm_wcmpgtuw (v2si, v2si) 6320long long __builtin_arm_wmacs (long long, v4hi, v4hi) 6321long long __builtin_arm_wmacsz (v4hi, v4hi) 6322long long __builtin_arm_wmacu (long long, v4hi, v4hi) 6323long long __builtin_arm_wmacuz (v4hi, v4hi) 6324v4hi __builtin_arm_wmadds (v4hi, v4hi) 6325v4hi __builtin_arm_wmaddu (v4hi, v4hi) 6326v8qi __builtin_arm_wmaxsb (v8qi, v8qi) 6327v4hi __builtin_arm_wmaxsh (v4hi, v4hi) 6328v2si __builtin_arm_wmaxsw (v2si, v2si) 6329v8qi __builtin_arm_wmaxub (v8qi, v8qi) 6330v4hi __builtin_arm_wmaxuh (v4hi, v4hi) 6331v2si __builtin_arm_wmaxuw (v2si, v2si) 6332v8qi __builtin_arm_wminsb (v8qi, v8qi) 6333v4hi __builtin_arm_wminsh (v4hi, v4hi) 6334v2si __builtin_arm_wminsw (v2si, v2si) 6335v8qi __builtin_arm_wminub (v8qi, v8qi) 6336v4hi __builtin_arm_wminuh (v4hi, v4hi) 6337v2si __builtin_arm_wminuw (v2si, v2si) 6338v4hi __builtin_arm_wmulsm (v4hi, v4hi) 6339v4hi __builtin_arm_wmulul (v4hi, v4hi) 6340v4hi __builtin_arm_wmulum (v4hi, v4hi) 6341long long __builtin_arm_wor (long long, long long) 6342v2si __builtin_arm_wpackdss (long long, long long) 6343v2si __builtin_arm_wpackdus (long long, long long) 6344v8qi __builtin_arm_wpackhss (v4hi, v4hi) 6345v8qi __builtin_arm_wpackhus (v4hi, v4hi) 6346v4hi __builtin_arm_wpackwss (v2si, v2si) 6347v4hi __builtin_arm_wpackwus (v2si, v2si) 6348long long __builtin_arm_wrord (long long, long long) 6349long long __builtin_arm_wrordi (long long, int) 6350v4hi __builtin_arm_wrorh (v4hi, long long) 6351v4hi __builtin_arm_wrorhi (v4hi, int) 6352v2si __builtin_arm_wrorw (v2si, long long) 6353v2si __builtin_arm_wrorwi (v2si, int) 6354v2si __builtin_arm_wsadb (v8qi, v8qi) 6355v2si __builtin_arm_wsadbz (v8qi, v8qi) 6356v2si __builtin_arm_wsadh (v4hi, v4hi) 6357v2si __builtin_arm_wsadhz (v4hi, v4hi) 6358v4hi __builtin_arm_wshufh (v4hi, int) 6359long long __builtin_arm_wslld (long long, long long) 6360long long __builtin_arm_wslldi (long long, int) 6361v4hi __builtin_arm_wsllh (v4hi, long long) 6362v4hi __builtin_arm_wsllhi (v4hi, int) 6363v2si __builtin_arm_wsllw (v2si, long long) 6364v2si __builtin_arm_wsllwi (v2si, int) 6365long long __builtin_arm_wsrad (long long, long long) 6366long long __builtin_arm_wsradi (long long, int) 6367v4hi __builtin_arm_wsrah (v4hi, long long) 6368v4hi __builtin_arm_wsrahi (v4hi, int) 6369v2si __builtin_arm_wsraw (v2si, long long) 6370v2si __builtin_arm_wsrawi (v2si, int) 6371long long __builtin_arm_wsrld (long long, long long) 6372long long __builtin_arm_wsrldi (long long, int) 6373v4hi __builtin_arm_wsrlh (v4hi, long long) 6374v4hi __builtin_arm_wsrlhi (v4hi, int) 6375v2si __builtin_arm_wsrlw (v2si, long long) 6376v2si __builtin_arm_wsrlwi (v2si, int) 6377v8qi __builtin_arm_wsubb (v8qi, v8qi) 6378v8qi __builtin_arm_wsubbss (v8qi, v8qi) 6379v8qi __builtin_arm_wsubbus (v8qi, v8qi) 6380v4hi __builtin_arm_wsubh (v4hi, v4hi) 6381v4hi __builtin_arm_wsubhss (v4hi, v4hi) 6382v4hi __builtin_arm_wsubhus (v4hi, v4hi) 6383v2si __builtin_arm_wsubw (v2si, v2si) 6384v2si __builtin_arm_wsubwss (v2si, v2si) 6385v2si __builtin_arm_wsubwus (v2si, v2si) 6386v4hi __builtin_arm_wunpckehsb (v8qi) 6387v2si __builtin_arm_wunpckehsh (v4hi) 6388long long __builtin_arm_wunpckehsw (v2si) 6389v4hi __builtin_arm_wunpckehub (v8qi) 6390v2si __builtin_arm_wunpckehuh (v4hi) 6391long long __builtin_arm_wunpckehuw (v2si) 6392v4hi __builtin_arm_wunpckelsb (v8qi) 6393v2si __builtin_arm_wunpckelsh (v4hi) 6394long long __builtin_arm_wunpckelsw (v2si) 6395v4hi __builtin_arm_wunpckelub (v8qi) 6396v2si __builtin_arm_wunpckeluh (v4hi) 6397long long __builtin_arm_wunpckeluw (v2si) 6398v8qi __builtin_arm_wunpckihb (v8qi, v8qi) 6399v4hi __builtin_arm_wunpckihh (v4hi, v4hi) 6400v2si __builtin_arm_wunpckihw (v2si, v2si) 6401v8qi __builtin_arm_wunpckilb (v8qi, v8qi) 6402v4hi __builtin_arm_wunpckilh (v4hi, v4hi) 6403v2si __builtin_arm_wunpckilw (v2si, v2si) 6404long long __builtin_arm_wxor (long long, long long) 6405long long __builtin_arm_wzero () 6406@end smallexample 6407 6408@node Blackfin Built-in Functions 6409@subsection Blackfin Built-in Functions 6410 6411Currently, there are two Blackfin-specific built-in functions. These are 6412used for generating @code{CSYNC} and @code{SSYNC} machine insns without 6413using inline assembly; by using these built-in functions the compiler can 6414automatically add workarounds for hardware errata involving these 6415instructions. These functions are named as follows: 6416 6417@smallexample 6418void __builtin_bfin_csync (void) 6419void __builtin_bfin_ssync (void) 6420@end smallexample 6421 6422@node FR-V Built-in Functions 6423@subsection FR-V Built-in Functions 6424 6425GCC provides many FR-V-specific built-in functions. In general, 6426these functions are intended to be compatible with those described 6427by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu 6428Semiconductor}. The two exceptions are @code{__MDUNPACKH} and 6429@code{__MBTOHE}, the gcc forms of which pass 128-bit values by 6430pointer rather than by value. 6431 6432Most of the functions are named after specific FR-V instructions. 6433Such functions are said to be ``directly mapped'' and are summarized 6434here in tabular form. 6435 6436@menu 6437* Argument Types:: 6438* Directly-mapped Integer Functions:: 6439* Directly-mapped Media Functions:: 6440* Raw read/write Functions:: 6441* Other Built-in Functions:: 6442@end menu 6443 6444@node Argument Types 6445@subsubsection Argument Types 6446 6447The arguments to the built-in functions can be divided into three groups: 6448register numbers, compile-time constants and run-time values. In order 6449to make this classification clear at a glance, the arguments and return 6450values are given the following pseudo types: 6451 6452@multitable @columnfractions .20 .30 .15 .35 6453@item Pseudo type @tab Real C type @tab Constant? @tab Description 6454@item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword 6455@item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word 6456@item @code{sw1} @tab @code{int} @tab No @tab a signed word 6457@item @code{uw2} @tab @code{unsigned long long} @tab No 6458@tab an unsigned doubleword 6459@item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword 6460@item @code{const} @tab @code{int} @tab Yes @tab an integer constant 6461@item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number 6462@item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number 6463@end multitable 6464 6465These pseudo types are not defined by GCC, they are simply a notational 6466convenience used in this manual. 6467 6468Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2} 6469and @code{sw2} are evaluated at run time. They correspond to 6470register operands in the underlying FR-V instructions. 6471 6472@code{const} arguments represent immediate operands in the underlying 6473FR-V instructions. They must be compile-time constants. 6474 6475@code{acc} arguments are evaluated at compile time and specify the number 6476of an accumulator register. For example, an @code{acc} argument of 2 6477will select the ACC2 register. 6478 6479@code{iacc} arguments are similar to @code{acc} arguments but specify the 6480number of an IACC register. See @pxref{Other Built-in Functions} 6481for more details. 6482 6483@node Directly-mapped Integer Functions 6484@subsubsection Directly-mapped Integer Functions 6485 6486The functions listed below map directly to FR-V I-type instructions. 6487 6488@multitable @columnfractions .45 .32 .23 6489@item Function prototype @tab Example usage @tab Assembly output 6490@item @code{sw1 __ADDSS (sw1, sw1)} 6491@tab @code{@var{c} = __ADDSS (@var{a}, @var{b})} 6492@tab @code{ADDSS @var{a},@var{b},@var{c}} 6493@item @code{sw1 __SCAN (sw1, sw1)} 6494@tab @code{@var{c} = __SCAN (@var{a}, @var{b})} 6495@tab @code{SCAN @var{a},@var{b},@var{c}} 6496@item @code{sw1 __SCUTSS (sw1)} 6497@tab @code{@var{b} = __SCUTSS (@var{a})} 6498@tab @code{SCUTSS @var{a},@var{b}} 6499@item @code{sw1 __SLASS (sw1, sw1)} 6500@tab @code{@var{c} = __SLASS (@var{a}, @var{b})} 6501@tab @code{SLASS @var{a},@var{b},@var{c}} 6502@item @code{void __SMASS (sw1, sw1)} 6503@tab @code{__SMASS (@var{a}, @var{b})} 6504@tab @code{SMASS @var{a},@var{b}} 6505@item @code{void __SMSSS (sw1, sw1)} 6506@tab @code{__SMSSS (@var{a}, @var{b})} 6507@tab @code{SMSSS @var{a},@var{b}} 6508@item @code{void __SMU (sw1, sw1)} 6509@tab @code{__SMU (@var{a}, @var{b})} 6510@tab @code{SMU @var{a},@var{b}} 6511@item @code{sw2 __SMUL (sw1, sw1)} 6512@tab @code{@var{c} = __SMUL (@var{a}, @var{b})} 6513@tab @code{SMUL @var{a},@var{b},@var{c}} 6514@item @code{sw1 __SUBSS (sw1, sw1)} 6515@tab @code{@var{c} = __SUBSS (@var{a}, @var{b})} 6516@tab @code{SUBSS @var{a},@var{b},@var{c}} 6517@item @code{uw2 __UMUL (uw1, uw1)} 6518@tab @code{@var{c} = __UMUL (@var{a}, @var{b})} 6519@tab @code{UMUL @var{a},@var{b},@var{c}} 6520@end multitable 6521 6522@node Directly-mapped Media Functions 6523@subsubsection Directly-mapped Media Functions 6524 6525The functions listed below map directly to FR-V M-type instructions. 6526 6527@multitable @columnfractions .45 .32 .23 6528@item Function prototype @tab Example usage @tab Assembly output 6529@item @code{uw1 __MABSHS (sw1)} 6530@tab @code{@var{b} = __MABSHS (@var{a})} 6531@tab @code{MABSHS @var{a},@var{b}} 6532@item @code{void __MADDACCS (acc, acc)} 6533@tab @code{__MADDACCS (@var{b}, @var{a})} 6534@tab @code{MADDACCS @var{a},@var{b}} 6535@item @code{sw1 __MADDHSS (sw1, sw1)} 6536@tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})} 6537@tab @code{MADDHSS @var{a},@var{b},@var{c}} 6538@item @code{uw1 __MADDHUS (uw1, uw1)} 6539@tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})} 6540@tab @code{MADDHUS @var{a},@var{b},@var{c}} 6541@item @code{uw1 __MAND (uw1, uw1)} 6542@tab @code{@var{c} = __MAND (@var{a}, @var{b})} 6543@tab @code{MAND @var{a},@var{b},@var{c}} 6544@item @code{void __MASACCS (acc, acc)} 6545@tab @code{__MASACCS (@var{b}, @var{a})} 6546@tab @code{MASACCS @var{a},@var{b}} 6547@item @code{uw1 __MAVEH (uw1, uw1)} 6548@tab @code{@var{c} = __MAVEH (@var{a}, @var{b})} 6549@tab @code{MAVEH @var{a},@var{b},@var{c}} 6550@item @code{uw2 __MBTOH (uw1)} 6551@tab @code{@var{b} = __MBTOH (@var{a})} 6552@tab @code{MBTOH @var{a},@var{b}} 6553@item @code{void __MBTOHE (uw1 *, uw1)} 6554@tab @code{__MBTOHE (&@var{b}, @var{a})} 6555@tab @code{MBTOHE @var{a},@var{b}} 6556@item @code{void __MCLRACC (acc)} 6557@tab @code{__MCLRACC (@var{a})} 6558@tab @code{MCLRACC @var{a}} 6559@item @code{void __MCLRACCA (void)} 6560@tab @code{__MCLRACCA ()} 6561@tab @code{MCLRACCA} 6562@item @code{uw1 __Mcop1 (uw1, uw1)} 6563@tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})} 6564@tab @code{Mcop1 @var{a},@var{b},@var{c}} 6565@item @code{uw1 __Mcop2 (uw1, uw1)} 6566@tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})} 6567@tab @code{Mcop2 @var{a},@var{b},@var{c}} 6568@item @code{uw1 __MCPLHI (uw2, const)} 6569@tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})} 6570@tab @code{MCPLHI @var{a},#@var{b},@var{c}} 6571@item @code{uw1 __MCPLI (uw2, const)} 6572@tab @code{@var{c} = __MCPLI (@var{a}, @var{b})} 6573@tab @code{MCPLI @var{a},#@var{b},@var{c}} 6574@item @code{void __MCPXIS (acc, sw1, sw1)} 6575@tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})} 6576@tab @code{MCPXIS @var{a},@var{b},@var{c}} 6577@item @code{void __MCPXIU (acc, uw1, uw1)} 6578@tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})} 6579@tab @code{MCPXIU @var{a},@var{b},@var{c}} 6580@item @code{void __MCPXRS (acc, sw1, sw1)} 6581@tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})} 6582@tab @code{MCPXRS @var{a},@var{b},@var{c}} 6583@item @code{void __MCPXRU (acc, uw1, uw1)} 6584@tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})} 6585@tab @code{MCPXRU @var{a},@var{b},@var{c}} 6586@item @code{uw1 __MCUT (acc, uw1)} 6587@tab @code{@var{c} = __MCUT (@var{a}, @var{b})} 6588@tab @code{MCUT @var{a},@var{b},@var{c}} 6589@item @code{uw1 __MCUTSS (acc, sw1)} 6590@tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})} 6591@tab @code{MCUTSS @var{a},@var{b},@var{c}} 6592@item @code{void __MDADDACCS (acc, acc)} 6593@tab @code{__MDADDACCS (@var{b}, @var{a})} 6594@tab @code{MDADDACCS @var{a},@var{b}} 6595@item @code{void __MDASACCS (acc, acc)} 6596@tab @code{__MDASACCS (@var{b}, @var{a})} 6597@tab @code{MDASACCS @var{a},@var{b}} 6598@item @code{uw2 __MDCUTSSI (acc, const)} 6599@tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})} 6600@tab @code{MDCUTSSI @var{a},#@var{b},@var{c}} 6601@item @code{uw2 __MDPACKH (uw2, uw2)} 6602@tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})} 6603@tab @code{MDPACKH @var{a},@var{b},@var{c}} 6604@item @code{uw2 __MDROTLI (uw2, const)} 6605@tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})} 6606@tab @code{MDROTLI @var{a},#@var{b},@var{c}} 6607@item @code{void __MDSUBACCS (acc, acc)} 6608@tab @code{__MDSUBACCS (@var{b}, @var{a})} 6609@tab @code{MDSUBACCS @var{a},@var{b}} 6610@item @code{void __MDUNPACKH (uw1 *, uw2)} 6611@tab @code{__MDUNPACKH (&@var{b}, @var{a})} 6612@tab @code{MDUNPACKH @var{a},@var{b}} 6613@item @code{uw2 __MEXPDHD (uw1, const)} 6614@tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})} 6615@tab @code{MEXPDHD @var{a},#@var{b},@var{c}} 6616@item @code{uw1 __MEXPDHW (uw1, const)} 6617@tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})} 6618@tab @code{MEXPDHW @var{a},#@var{b},@var{c}} 6619@item @code{uw1 __MHDSETH (uw1, const)} 6620@tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})} 6621@tab @code{MHDSETH @var{a},#@var{b},@var{c}} 6622@item @code{sw1 __MHDSETS (const)} 6623@tab @code{@var{b} = __MHDSETS (@var{a})} 6624@tab @code{MHDSETS #@var{a},@var{b}} 6625@item @code{uw1 __MHSETHIH (uw1, const)} 6626@tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})} 6627@tab @code{MHSETHIH #@var{a},@var{b}} 6628@item @code{sw1 __MHSETHIS (sw1, const)} 6629@tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})} 6630@tab @code{MHSETHIS #@var{a},@var{b}} 6631@item @code{uw1 __MHSETLOH (uw1, const)} 6632@tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})} 6633@tab @code{MHSETLOH #@var{a},@var{b}} 6634@item @code{sw1 __MHSETLOS (sw1, const)} 6635@tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})} 6636@tab @code{MHSETLOS #@var{a},@var{b}} 6637@item @code{uw1 __MHTOB (uw2)} 6638@tab @code{@var{b} = __MHTOB (@var{a})} 6639@tab @code{MHTOB @var{a},@var{b}} 6640@item @code{void __MMACHS (acc, sw1, sw1)} 6641@tab @code{__MMACHS (@var{c}, @var{a}, @var{b})} 6642@tab @code{MMACHS @var{a},@var{b},@var{c}} 6643@item @code{void __MMACHU (acc, uw1, uw1)} 6644@tab @code{__MMACHU (@var{c}, @var{a}, @var{b})} 6645@tab @code{MMACHU @var{a},@var{b},@var{c}} 6646@item @code{void __MMRDHS (acc, sw1, sw1)} 6647@tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})} 6648@tab @code{MMRDHS @var{a},@var{b},@var{c}} 6649@item @code{void __MMRDHU (acc, uw1, uw1)} 6650@tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})} 6651@tab @code{MMRDHU @var{a},@var{b},@var{c}} 6652@item @code{void __MMULHS (acc, sw1, sw1)} 6653@tab @code{__MMULHS (@var{c}, @var{a}, @var{b})} 6654@tab @code{MMULHS @var{a},@var{b},@var{c}} 6655@item @code{void __MMULHU (acc, uw1, uw1)} 6656@tab @code{__MMULHU (@var{c}, @var{a}, @var{b})} 6657@tab @code{MMULHU @var{a},@var{b},@var{c}} 6658@item @code{void __MMULXHS (acc, sw1, sw1)} 6659@tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})} 6660@tab @code{MMULXHS @var{a},@var{b},@var{c}} 6661@item @code{void __MMULXHU (acc, uw1, uw1)} 6662@tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})} 6663@tab @code{MMULXHU @var{a},@var{b},@var{c}} 6664@item @code{uw1 __MNOT (uw1)} 6665@tab @code{@var{b} = __MNOT (@var{a})} 6666@tab @code{MNOT @var{a},@var{b}} 6667@item @code{uw1 __MOR (uw1, uw1)} 6668@tab @code{@var{c} = __MOR (@var{a}, @var{b})} 6669@tab @code{MOR @var{a},@var{b},@var{c}} 6670@item @code{uw1 __MPACKH (uh, uh)} 6671@tab @code{@var{c} = __MPACKH (@var{a}, @var{b})} 6672@tab @code{MPACKH @var{a},@var{b},@var{c}} 6673@item @code{sw2 __MQADDHSS (sw2, sw2)} 6674@tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})} 6675@tab @code{MQADDHSS @var{a},@var{b},@var{c}} 6676@item @code{uw2 __MQADDHUS (uw2, uw2)} 6677@tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})} 6678@tab @code{MQADDHUS @var{a},@var{b},@var{c}} 6679@item @code{void __MQCPXIS (acc, sw2, sw2)} 6680@tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})} 6681@tab @code{MQCPXIS @var{a},@var{b},@var{c}} 6682@item @code{void __MQCPXIU (acc, uw2, uw2)} 6683@tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})} 6684@tab @code{MQCPXIU @var{a},@var{b},@var{c}} 6685@item @code{void __MQCPXRS (acc, sw2, sw2)} 6686@tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})} 6687@tab @code{MQCPXRS @var{a},@var{b},@var{c}} 6688@item @code{void __MQCPXRU (acc, uw2, uw2)} 6689@tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})} 6690@tab @code{MQCPXRU @var{a},@var{b},@var{c}} 6691@item @code{sw2 __MQLCLRHS (sw2, sw2)} 6692@tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})} 6693@tab @code{MQLCLRHS @var{a},@var{b},@var{c}} 6694@item @code{sw2 __MQLMTHS (sw2, sw2)} 6695@tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})} 6696@tab @code{MQLMTHS @var{a},@var{b},@var{c}} 6697@item @code{void __MQMACHS (acc, sw2, sw2)} 6698@tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})} 6699@tab @code{MQMACHS @var{a},@var{b},@var{c}} 6700@item @code{void __MQMACHU (acc, uw2, uw2)} 6701@tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})} 6702@tab @code{MQMACHU @var{a},@var{b},@var{c}} 6703@item @code{void __MQMACXHS (acc, sw2, sw2)} 6704@tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})} 6705@tab @code{MQMACXHS @var{a},@var{b},@var{c}} 6706@item @code{void __MQMULHS (acc, sw2, sw2)} 6707@tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})} 6708@tab @code{MQMULHS @var{a},@var{b},@var{c}} 6709@item @code{void __MQMULHU (acc, uw2, uw2)} 6710@tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})} 6711@tab @code{MQMULHU @var{a},@var{b},@var{c}} 6712@item @code{void __MQMULXHS (acc, sw2, sw2)} 6713@tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})} 6714@tab @code{MQMULXHS @var{a},@var{b},@var{c}} 6715@item @code{void __MQMULXHU (acc, uw2, uw2)} 6716@tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})} 6717@tab @code{MQMULXHU @var{a},@var{b},@var{c}} 6718@item @code{sw2 __MQSATHS (sw2, sw2)} 6719@tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})} 6720@tab @code{MQSATHS @var{a},@var{b},@var{c}} 6721@item @code{uw2 __MQSLLHI (uw2, int)} 6722@tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})} 6723@tab @code{MQSLLHI @var{a},@var{b},@var{c}} 6724@item @code{sw2 __MQSRAHI (sw2, int)} 6725@tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})} 6726@tab @code{MQSRAHI @var{a},@var{b},@var{c}} 6727@item @code{sw2 __MQSUBHSS (sw2, sw2)} 6728@tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})} 6729@tab @code{MQSUBHSS @var{a},@var{b},@var{c}} 6730@item @code{uw2 __MQSUBHUS (uw2, uw2)} 6731@tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})} 6732@tab @code{MQSUBHUS @var{a},@var{b},@var{c}} 6733@item @code{void __MQXMACHS (acc, sw2, sw2)} 6734@tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})} 6735@tab @code{MQXMACHS @var{a},@var{b},@var{c}} 6736@item @code{void __MQXMACXHS (acc, sw2, sw2)} 6737@tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})} 6738@tab @code{MQXMACXHS @var{a},@var{b},@var{c}} 6739@item @code{uw1 __MRDACC (acc)} 6740@tab @code{@var{b} = __MRDACC (@var{a})} 6741@tab @code{MRDACC @var{a},@var{b}} 6742@item @code{uw1 __MRDACCG (acc)} 6743@tab @code{@var{b} = __MRDACCG (@var{a})} 6744@tab @code{MRDACCG @var{a},@var{b}} 6745@item @code{uw1 __MROTLI (uw1, const)} 6746@tab @code{@var{c} = __MROTLI (@var{a}, @var{b})} 6747@tab @code{MROTLI @var{a},#@var{b},@var{c}} 6748@item @code{uw1 __MROTRI (uw1, const)} 6749@tab @code{@var{c} = __MROTRI (@var{a}, @var{b})} 6750@tab @code{MROTRI @var{a},#@var{b},@var{c}} 6751@item @code{sw1 __MSATHS (sw1, sw1)} 6752@tab @code{@var{c} = __MSATHS (@var{a}, @var{b})} 6753@tab @code{MSATHS @var{a},@var{b},@var{c}} 6754@item @code{uw1 __MSATHU (uw1, uw1)} 6755@tab @code{@var{c} = __MSATHU (@var{a}, @var{b})} 6756@tab @code{MSATHU @var{a},@var{b},@var{c}} 6757@item @code{uw1 __MSLLHI (uw1, const)} 6758@tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})} 6759@tab @code{MSLLHI @var{a},#@var{b},@var{c}} 6760@item @code{sw1 __MSRAHI (sw1, const)} 6761@tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})} 6762@tab @code{MSRAHI @var{a},#@var{b},@var{c}} 6763@item @code{uw1 __MSRLHI (uw1, const)} 6764@tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})} 6765@tab @code{MSRLHI @var{a},#@var{b},@var{c}} 6766@item @code{void __MSUBACCS (acc, acc)} 6767@tab @code{__MSUBACCS (@var{b}, @var{a})} 6768@tab @code{MSUBACCS @var{a},@var{b}} 6769@item @code{sw1 __MSUBHSS (sw1, sw1)} 6770@tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})} 6771@tab @code{MSUBHSS @var{a},@var{b},@var{c}} 6772@item @code{uw1 __MSUBHUS (uw1, uw1)} 6773@tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})} 6774@tab @code{MSUBHUS @var{a},@var{b},@var{c}} 6775@item @code{void __MTRAP (void)} 6776@tab @code{__MTRAP ()} 6777@tab @code{MTRAP} 6778@item @code{uw2 __MUNPACKH (uw1)} 6779@tab @code{@var{b} = __MUNPACKH (@var{a})} 6780@tab @code{MUNPACKH @var{a},@var{b}} 6781@item @code{uw1 __MWCUT (uw2, uw1)} 6782@tab @code{@var{c} = __MWCUT (@var{a}, @var{b})} 6783@tab @code{MWCUT @var{a},@var{b},@var{c}} 6784@item @code{void __MWTACC (acc, uw1)} 6785@tab @code{__MWTACC (@var{b}, @var{a})} 6786@tab @code{MWTACC @var{a},@var{b}} 6787@item @code{void __MWTACCG (acc, uw1)} 6788@tab @code{__MWTACCG (@var{b}, @var{a})} 6789@tab @code{MWTACCG @var{a},@var{b}} 6790@item @code{uw1 __MXOR (uw1, uw1)} 6791@tab @code{@var{c} = __MXOR (@var{a}, @var{b})} 6792@tab @code{MXOR @var{a},@var{b},@var{c}} 6793@end multitable 6794 6795@node Raw read/write Functions 6796@subsubsection Raw read/write Functions 6797 6798This sections describes built-in functions related to read and write 6799instructions to access memory. These functions generate 6800@code{membar} instructions to flush the I/O load and stores where 6801appropriate, as described in Fujitsu's manual described above. 6802 6803@table @code 6804 6805@item unsigned char __builtin_read8 (void *@var{data}) 6806@item unsigned short __builtin_read16 (void *@var{data}) 6807@item unsigned long __builtin_read32 (void *@var{data}) 6808@item unsigned long long __builtin_read64 (void *@var{data}) 6809 6810@item void __builtin_write8 (void *@var{data}, unsigned char @var{datum}) 6811@item void __builtin_write16 (void *@var{data}, unsigned short @var{datum}) 6812@item void __builtin_write32 (void *@var{data}, unsigned long @var{datum}) 6813@item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum}) 6814@end table 6815 6816@node Other Built-in Functions 6817@subsubsection Other Built-in Functions 6818 6819This section describes built-in functions that are not named after 6820a specific FR-V instruction. 6821 6822@table @code 6823@item sw2 __IACCreadll (iacc @var{reg}) 6824Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved 6825for future expansion and must be 0. 6826 6827@item sw1 __IACCreadl (iacc @var{reg}) 6828Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1. 6829Other values of @var{reg} are rejected as invalid. 6830 6831@item void __IACCsetll (iacc @var{reg}, sw2 @var{x}) 6832Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument 6833is reserved for future expansion and must be 0. 6834 6835@item void __IACCsetl (iacc @var{reg}, sw1 @var{x}) 6836Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg} 6837is 1. Other values of @var{reg} are rejected as invalid. 6838 6839@item void __data_prefetch0 (const void *@var{x}) 6840Use the @code{dcpl} instruction to load the contents of address @var{x} 6841into the data cache. 6842 6843@item void __data_prefetch (const void *@var{x}) 6844Use the @code{nldub} instruction to load the contents of address @var{x} 6845into the data cache. The instruction will be issued in slot I1@. 6846@end table 6847 6848@node X86 Built-in Functions 6849@subsection X86 Built-in Functions 6850 6851These built-in functions are available for the i386 and x86-64 family 6852of computers, depending on the command-line switches used. 6853 6854Note that, if you specify command-line switches such as @option{-msse}, 6855the compiler could use the extended instruction sets even if the built-ins 6856are not used explicitly in the program. For this reason, applications 6857which perform runtime CPU detection must compile separate files for each 6858supported architecture, using the appropriate flags. In particular, 6859the file containing the CPU detection code should be compiled without 6860these options. 6861 6862The following machine modes are available for use with MMX built-in functions 6863(@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers, 6864@code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a 6865vector of eight 8-bit integers. Some of the built-in functions operate on 6866MMX registers as a whole 64-bit entity, these use @code{DI} as their mode. 6867 6868If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector 6869of two 32-bit floating point values. 6870 6871If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit 6872floating point values. Some instructions use a vector of four 32-bit 6873integers, these use @code{V4SI}. Finally, some instructions operate on an 6874entire vector register, interpreting it as a 128-bit integer, these use mode 6875@code{TI}. 6876 6877The following built-in functions are made available by @option{-mmmx}. 6878All of them generate the machine instruction that is part of the name. 6879 6880@smallexample 6881v8qi __builtin_ia32_paddb (v8qi, v8qi) 6882v4hi __builtin_ia32_paddw (v4hi, v4hi) 6883v2si __builtin_ia32_paddd (v2si, v2si) 6884v8qi __builtin_ia32_psubb (v8qi, v8qi) 6885v4hi __builtin_ia32_psubw (v4hi, v4hi) 6886v2si __builtin_ia32_psubd (v2si, v2si) 6887v8qi __builtin_ia32_paddsb (v8qi, v8qi) 6888v4hi __builtin_ia32_paddsw (v4hi, v4hi) 6889v8qi __builtin_ia32_psubsb (v8qi, v8qi) 6890v4hi __builtin_ia32_psubsw (v4hi, v4hi) 6891v8qi __builtin_ia32_paddusb (v8qi, v8qi) 6892v4hi __builtin_ia32_paddusw (v4hi, v4hi) 6893v8qi __builtin_ia32_psubusb (v8qi, v8qi) 6894v4hi __builtin_ia32_psubusw (v4hi, v4hi) 6895v4hi __builtin_ia32_pmullw (v4hi, v4hi) 6896v4hi __builtin_ia32_pmulhw (v4hi, v4hi) 6897di __builtin_ia32_pand (di, di) 6898di __builtin_ia32_pandn (di,di) 6899di __builtin_ia32_por (di, di) 6900di __builtin_ia32_pxor (di, di) 6901v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi) 6902v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi) 6903v2si __builtin_ia32_pcmpeqd (v2si, v2si) 6904v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi) 6905v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi) 6906v2si __builtin_ia32_pcmpgtd (v2si, v2si) 6907v8qi __builtin_ia32_punpckhbw (v8qi, v8qi) 6908v4hi __builtin_ia32_punpckhwd (v4hi, v4hi) 6909v2si __builtin_ia32_punpckhdq (v2si, v2si) 6910v8qi __builtin_ia32_punpcklbw (v8qi, v8qi) 6911v4hi __builtin_ia32_punpcklwd (v4hi, v4hi) 6912v2si __builtin_ia32_punpckldq (v2si, v2si) 6913v8qi __builtin_ia32_packsswb (v4hi, v4hi) 6914v4hi __builtin_ia32_packssdw (v2si, v2si) 6915v8qi __builtin_ia32_packuswb (v4hi, v4hi) 6916@end smallexample 6917 6918The following built-in functions are made available either with 6919@option{-msse}, or with a combination of @option{-m3dnow} and 6920@option{-march=athlon}. All of them generate the machine 6921instruction that is part of the name. 6922 6923@smallexample 6924v4hi __builtin_ia32_pmulhuw (v4hi, v4hi) 6925v8qi __builtin_ia32_pavgb (v8qi, v8qi) 6926v4hi __builtin_ia32_pavgw (v4hi, v4hi) 6927v4hi __builtin_ia32_psadbw (v8qi, v8qi) 6928v8qi __builtin_ia32_pmaxub (v8qi, v8qi) 6929v4hi __builtin_ia32_pmaxsw (v4hi, v4hi) 6930v8qi __builtin_ia32_pminub (v8qi, v8qi) 6931v4hi __builtin_ia32_pminsw (v4hi, v4hi) 6932int __builtin_ia32_pextrw (v4hi, int) 6933v4hi __builtin_ia32_pinsrw (v4hi, int, int) 6934int __builtin_ia32_pmovmskb (v8qi) 6935void __builtin_ia32_maskmovq (v8qi, v8qi, char *) 6936void __builtin_ia32_movntq (di *, di) 6937void __builtin_ia32_sfence (void) 6938@end smallexample 6939 6940The following built-in functions are available when @option{-msse} is used. 6941All of them generate the machine instruction that is part of the name. 6942 6943@smallexample 6944int __builtin_ia32_comieq (v4sf, v4sf) 6945int __builtin_ia32_comineq (v4sf, v4sf) 6946int __builtin_ia32_comilt (v4sf, v4sf) 6947int __builtin_ia32_comile (v4sf, v4sf) 6948int __builtin_ia32_comigt (v4sf, v4sf) 6949int __builtin_ia32_comige (v4sf, v4sf) 6950int __builtin_ia32_ucomieq (v4sf, v4sf) 6951int __builtin_ia32_ucomineq (v4sf, v4sf) 6952int __builtin_ia32_ucomilt (v4sf, v4sf) 6953int __builtin_ia32_ucomile (v4sf, v4sf) 6954int __builtin_ia32_ucomigt (v4sf, v4sf) 6955int __builtin_ia32_ucomige (v4sf, v4sf) 6956v4sf __builtin_ia32_addps (v4sf, v4sf) 6957v4sf __builtin_ia32_subps (v4sf, v4sf) 6958v4sf __builtin_ia32_mulps (v4sf, v4sf) 6959v4sf __builtin_ia32_divps (v4sf, v4sf) 6960v4sf __builtin_ia32_addss (v4sf, v4sf) 6961v4sf __builtin_ia32_subss (v4sf, v4sf) 6962v4sf __builtin_ia32_mulss (v4sf, v4sf) 6963v4sf __builtin_ia32_divss (v4sf, v4sf) 6964v4si __builtin_ia32_cmpeqps (v4sf, v4sf) 6965v4si __builtin_ia32_cmpltps (v4sf, v4sf) 6966v4si __builtin_ia32_cmpleps (v4sf, v4sf) 6967v4si __builtin_ia32_cmpgtps (v4sf, v4sf) 6968v4si __builtin_ia32_cmpgeps (v4sf, v4sf) 6969v4si __builtin_ia32_cmpunordps (v4sf, v4sf) 6970v4si __builtin_ia32_cmpneqps (v4sf, v4sf) 6971v4si __builtin_ia32_cmpnltps (v4sf, v4sf) 6972v4si __builtin_ia32_cmpnleps (v4sf, v4sf) 6973v4si __builtin_ia32_cmpngtps (v4sf, v4sf) 6974v4si __builtin_ia32_cmpngeps (v4sf, v4sf) 6975v4si __builtin_ia32_cmpordps (v4sf, v4sf) 6976v4si __builtin_ia32_cmpeqss (v4sf, v4sf) 6977v4si __builtin_ia32_cmpltss (v4sf, v4sf) 6978v4si __builtin_ia32_cmpless (v4sf, v4sf) 6979v4si __builtin_ia32_cmpunordss (v4sf, v4sf) 6980v4si __builtin_ia32_cmpneqss (v4sf, v4sf) 6981v4si __builtin_ia32_cmpnlts (v4sf, v4sf) 6982v4si __builtin_ia32_cmpnless (v4sf, v4sf) 6983v4si __builtin_ia32_cmpordss (v4sf, v4sf) 6984v4sf __builtin_ia32_maxps (v4sf, v4sf) 6985v4sf __builtin_ia32_maxss (v4sf, v4sf) 6986v4sf __builtin_ia32_minps (v4sf, v4sf) 6987v4sf __builtin_ia32_minss (v4sf, v4sf) 6988v4sf __builtin_ia32_andps (v4sf, v4sf) 6989v4sf __builtin_ia32_andnps (v4sf, v4sf) 6990v4sf __builtin_ia32_orps (v4sf, v4sf) 6991v4sf __builtin_ia32_xorps (v4sf, v4sf) 6992v4sf __builtin_ia32_movss (v4sf, v4sf) 6993v4sf __builtin_ia32_movhlps (v4sf, v4sf) 6994v4sf __builtin_ia32_movlhps (v4sf, v4sf) 6995v4sf __builtin_ia32_unpckhps (v4sf, v4sf) 6996v4sf __builtin_ia32_unpcklps (v4sf, v4sf) 6997v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si) 6998v4sf __builtin_ia32_cvtsi2ss (v4sf, int) 6999v2si __builtin_ia32_cvtps2pi (v4sf) 7000int __builtin_ia32_cvtss2si (v4sf) 7001v2si __builtin_ia32_cvttps2pi (v4sf) 7002int __builtin_ia32_cvttss2si (v4sf) 7003v4sf __builtin_ia32_rcpps (v4sf) 7004v4sf __builtin_ia32_rsqrtps (v4sf) 7005v4sf __builtin_ia32_sqrtps (v4sf) 7006v4sf __builtin_ia32_rcpss (v4sf) 7007v4sf __builtin_ia32_rsqrtss (v4sf) 7008v4sf __builtin_ia32_sqrtss (v4sf) 7009v4sf __builtin_ia32_shufps (v4sf, v4sf, int) 7010void __builtin_ia32_movntps (float *, v4sf) 7011int __builtin_ia32_movmskps (v4sf) 7012@end smallexample 7013 7014The following built-in functions are available when @option{-msse} is used. 7015 7016@table @code 7017@item v4sf __builtin_ia32_loadaps (float *) 7018Generates the @code{movaps} machine instruction as a load from memory. 7019@item void __builtin_ia32_storeaps (float *, v4sf) 7020Generates the @code{movaps} machine instruction as a store to memory. 7021@item v4sf __builtin_ia32_loadups (float *) 7022Generates the @code{movups} machine instruction as a load from memory. 7023@item void __builtin_ia32_storeups (float *, v4sf) 7024Generates the @code{movups} machine instruction as a store to memory. 7025@item v4sf __builtin_ia32_loadsss (float *) 7026Generates the @code{movss} machine instruction as a load from memory. 7027@item void __builtin_ia32_storess (float *, v4sf) 7028Generates the @code{movss} machine instruction as a store to memory. 7029@item v4sf __builtin_ia32_loadhps (v4sf, v2si *) 7030Generates the @code{movhps} machine instruction as a load from memory. 7031@item v4sf __builtin_ia32_loadlps (v4sf, v2si *) 7032Generates the @code{movlps} machine instruction as a load from memory 7033@item void __builtin_ia32_storehps (v4sf, v2si *) 7034Generates the @code{movhps} machine instruction as a store to memory. 7035@item void __builtin_ia32_storelps (v4sf, v2si *) 7036Generates the @code{movlps} machine instruction as a store to memory. 7037@end table 7038 7039The following built-in functions are available when @option{-msse2} is used. 7040All of them generate the machine instruction that is part of the name. 7041 7042@smallexample 7043int __builtin_ia32_comisdeq (v2df, v2df) 7044int __builtin_ia32_comisdlt (v2df, v2df) 7045int __builtin_ia32_comisdle (v2df, v2df) 7046int __builtin_ia32_comisdgt (v2df, v2df) 7047int __builtin_ia32_comisdge (v2df, v2df) 7048int __builtin_ia32_comisdneq (v2df, v2df) 7049int __builtin_ia32_ucomisdeq (v2df, v2df) 7050int __builtin_ia32_ucomisdlt (v2df, v2df) 7051int __builtin_ia32_ucomisdle (v2df, v2df) 7052int __builtin_ia32_ucomisdgt (v2df, v2df) 7053int __builtin_ia32_ucomisdge (v2df, v2df) 7054int __builtin_ia32_ucomisdneq (v2df, v2df) 7055v2df __builtin_ia32_cmpeqpd (v2df, v2df) 7056v2df __builtin_ia32_cmpltpd (v2df, v2df) 7057v2df __builtin_ia32_cmplepd (v2df, v2df) 7058v2df __builtin_ia32_cmpgtpd (v2df, v2df) 7059v2df __builtin_ia32_cmpgepd (v2df, v2df) 7060v2df __builtin_ia32_cmpunordpd (v2df, v2df) 7061v2df __builtin_ia32_cmpneqpd (v2df, v2df) 7062v2df __builtin_ia32_cmpnltpd (v2df, v2df) 7063v2df __builtin_ia32_cmpnlepd (v2df, v2df) 7064v2df __builtin_ia32_cmpngtpd (v2df, v2df) 7065v2df __builtin_ia32_cmpngepd (v2df, v2df) 7066v2df __builtin_ia32_cmpordpd (v2df, v2df) 7067v2df __builtin_ia32_cmpeqsd (v2df, v2df) 7068v2df __builtin_ia32_cmpltsd (v2df, v2df) 7069v2df __builtin_ia32_cmplesd (v2df, v2df) 7070v2df __builtin_ia32_cmpunordsd (v2df, v2df) 7071v2df __builtin_ia32_cmpneqsd (v2df, v2df) 7072v2df __builtin_ia32_cmpnltsd (v2df, v2df) 7073v2df __builtin_ia32_cmpnlesd (v2df, v2df) 7074v2df __builtin_ia32_cmpordsd (v2df, v2df) 7075v2di __builtin_ia32_paddq (v2di, v2di) 7076v2di __builtin_ia32_psubq (v2di, v2di) 7077v2df __builtin_ia32_addpd (v2df, v2df) 7078v2df __builtin_ia32_subpd (v2df, v2df) 7079v2df __builtin_ia32_mulpd (v2df, v2df) 7080v2df __builtin_ia32_divpd (v2df, v2df) 7081v2df __builtin_ia32_addsd (v2df, v2df) 7082v2df __builtin_ia32_subsd (v2df, v2df) 7083v2df __builtin_ia32_mulsd (v2df, v2df) 7084v2df __builtin_ia32_divsd (v2df, v2df) 7085v2df __builtin_ia32_minpd (v2df, v2df) 7086v2df __builtin_ia32_maxpd (v2df, v2df) 7087v2df __builtin_ia32_minsd (v2df, v2df) 7088v2df __builtin_ia32_maxsd (v2df, v2df) 7089v2df __builtin_ia32_andpd (v2df, v2df) 7090v2df __builtin_ia32_andnpd (v2df, v2df) 7091v2df __builtin_ia32_orpd (v2df, v2df) 7092v2df __builtin_ia32_xorpd (v2df, v2df) 7093v2df __builtin_ia32_movsd (v2df, v2df) 7094v2df __builtin_ia32_unpckhpd (v2df, v2df) 7095v2df __builtin_ia32_unpcklpd (v2df, v2df) 7096v16qi __builtin_ia32_paddb128 (v16qi, v16qi) 7097v8hi __builtin_ia32_paddw128 (v8hi, v8hi) 7098v4si __builtin_ia32_paddd128 (v4si, v4si) 7099v2di __builtin_ia32_paddq128 (v2di, v2di) 7100v16qi __builtin_ia32_psubb128 (v16qi, v16qi) 7101v8hi __builtin_ia32_psubw128 (v8hi, v8hi) 7102v4si __builtin_ia32_psubd128 (v4si, v4si) 7103v2di __builtin_ia32_psubq128 (v2di, v2di) 7104v8hi __builtin_ia32_pmullw128 (v8hi, v8hi) 7105v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi) 7106v2di __builtin_ia32_pand128 (v2di, v2di) 7107v2di __builtin_ia32_pandn128 (v2di, v2di) 7108v2di __builtin_ia32_por128 (v2di, v2di) 7109v2di __builtin_ia32_pxor128 (v2di, v2di) 7110v16qi __builtin_ia32_pavgb128 (v16qi, v16qi) 7111v8hi __builtin_ia32_pavgw128 (v8hi, v8hi) 7112v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi) 7113v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi) 7114v4si __builtin_ia32_pcmpeqd128 (v4si, v4si) 7115v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi) 7116v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi) 7117v4si __builtin_ia32_pcmpgtd128 (v4si, v4si) 7118v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi) 7119v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi) 7120v16qi __builtin_ia32_pminub128 (v16qi, v16qi) 7121v8hi __builtin_ia32_pminsw128 (v8hi, v8hi) 7122v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi) 7123v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi) 7124v4si __builtin_ia32_punpckhdq128 (v4si, v4si) 7125v2di __builtin_ia32_punpckhqdq128 (v2di, v2di) 7126v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi) 7127v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi) 7128v4si __builtin_ia32_punpckldq128 (v4si, v4si) 7129v2di __builtin_ia32_punpcklqdq128 (v2di, v2di) 7130v16qi __builtin_ia32_packsswb128 (v16qi, v16qi) 7131v8hi __builtin_ia32_packssdw128 (v8hi, v8hi) 7132v16qi __builtin_ia32_packuswb128 (v16qi, v16qi) 7133v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi) 7134void __builtin_ia32_maskmovdqu (v16qi, v16qi) 7135v2df __builtin_ia32_loadupd (double *) 7136void __builtin_ia32_storeupd (double *, v2df) 7137v2df __builtin_ia32_loadhpd (v2df, double *) 7138v2df __builtin_ia32_loadlpd (v2df, double *) 7139int __builtin_ia32_movmskpd (v2df) 7140int __builtin_ia32_pmovmskb128 (v16qi) 7141void __builtin_ia32_movnti (int *, int) 7142void __builtin_ia32_movntpd (double *, v2df) 7143void __builtin_ia32_movntdq (v2df *, v2df) 7144v4si __builtin_ia32_pshufd (v4si, int) 7145v8hi __builtin_ia32_pshuflw (v8hi, int) 7146v8hi __builtin_ia32_pshufhw (v8hi, int) 7147v2di __builtin_ia32_psadbw128 (v16qi, v16qi) 7148v2df __builtin_ia32_sqrtpd (v2df) 7149v2df __builtin_ia32_sqrtsd (v2df) 7150v2df __builtin_ia32_shufpd (v2df, v2df, int) 7151v2df __builtin_ia32_cvtdq2pd (v4si) 7152v4sf __builtin_ia32_cvtdq2ps (v4si) 7153v4si __builtin_ia32_cvtpd2dq (v2df) 7154v2si __builtin_ia32_cvtpd2pi (v2df) 7155v4sf __builtin_ia32_cvtpd2ps (v2df) 7156v4si __builtin_ia32_cvttpd2dq (v2df) 7157v2si __builtin_ia32_cvttpd2pi (v2df) 7158v2df __builtin_ia32_cvtpi2pd (v2si) 7159int __builtin_ia32_cvtsd2si (v2df) 7160int __builtin_ia32_cvttsd2si (v2df) 7161long long __builtin_ia32_cvtsd2si64 (v2df) 7162long long __builtin_ia32_cvttsd2si64 (v2df) 7163v4si __builtin_ia32_cvtps2dq (v4sf) 7164v2df __builtin_ia32_cvtps2pd (v4sf) 7165v4si __builtin_ia32_cvttps2dq (v4sf) 7166v2df __builtin_ia32_cvtsi2sd (v2df, int) 7167v2df __builtin_ia32_cvtsi642sd (v2df, long long) 7168v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df) 7169v2df __builtin_ia32_cvtss2sd (v2df, v4sf) 7170void __builtin_ia32_clflush (const void *) 7171void __builtin_ia32_lfence (void) 7172void __builtin_ia32_mfence (void) 7173v16qi __builtin_ia32_loaddqu (const char *) 7174void __builtin_ia32_storedqu (char *, v16qi) 7175unsigned long long __builtin_ia32_pmuludq (v2si, v2si) 7176v2di __builtin_ia32_pmuludq128 (v4si, v4si) 7177v8hi __builtin_ia32_psllw128 (v8hi, v2di) 7178v4si __builtin_ia32_pslld128 (v4si, v2di) 7179v2di __builtin_ia32_psllq128 (v4si, v2di) 7180v8hi __builtin_ia32_psrlw128 (v8hi, v2di) 7181v4si __builtin_ia32_psrld128 (v4si, v2di) 7182v2di __builtin_ia32_psrlq128 (v2di, v2di) 7183v8hi __builtin_ia32_psraw128 (v8hi, v2di) 7184v4si __builtin_ia32_psrad128 (v4si, v2di) 7185v2di __builtin_ia32_pslldqi128 (v2di, int) 7186v8hi __builtin_ia32_psllwi128 (v8hi, int) 7187v4si __builtin_ia32_pslldi128 (v4si, int) 7188v2di __builtin_ia32_psllqi128 (v2di, int) 7189v2di __builtin_ia32_psrldqi128 (v2di, int) 7190v8hi __builtin_ia32_psrlwi128 (v8hi, int) 7191v4si __builtin_ia32_psrldi128 (v4si, int) 7192v2di __builtin_ia32_psrlqi128 (v2di, int) 7193v8hi __builtin_ia32_psrawi128 (v8hi, int) 7194v4si __builtin_ia32_psradi128 (v4si, int) 7195v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi) 7196@end smallexample 7197 7198The following built-in functions are available when @option{-msse3} is used. 7199All of them generate the machine instruction that is part of the name. 7200 7201@smallexample 7202v2df __builtin_ia32_addsubpd (v2df, v2df) 7203v4sf __builtin_ia32_addsubps (v4sf, v4sf) 7204v2df __builtin_ia32_haddpd (v2df, v2df) 7205v4sf __builtin_ia32_haddps (v4sf, v4sf) 7206v2df __builtin_ia32_hsubpd (v2df, v2df) 7207v4sf __builtin_ia32_hsubps (v4sf, v4sf) 7208v16qi __builtin_ia32_lddqu (char const *) 7209void __builtin_ia32_monitor (void *, unsigned int, unsigned int) 7210v2df __builtin_ia32_movddup (v2df) 7211v4sf __builtin_ia32_movshdup (v4sf) 7212v4sf __builtin_ia32_movsldup (v4sf) 7213void __builtin_ia32_mwait (unsigned int, unsigned int) 7214@end smallexample 7215 7216The following built-in functions are available when @option{-msse3} is used. 7217 7218@table @code 7219@item v2df __builtin_ia32_loadddup (double const *) 7220Generates the @code{movddup} machine instruction as a load from memory. 7221@end table 7222 7223The following built-in functions are available when @option{-m3dnow} is used. 7224All of them generate the machine instruction that is part of the name. 7225 7226@smallexample 7227void __builtin_ia32_femms (void) 7228v8qi __builtin_ia32_pavgusb (v8qi, v8qi) 7229v2si __builtin_ia32_pf2id (v2sf) 7230v2sf __builtin_ia32_pfacc (v2sf, v2sf) 7231v2sf __builtin_ia32_pfadd (v2sf, v2sf) 7232v2si __builtin_ia32_pfcmpeq (v2sf, v2sf) 7233v2si __builtin_ia32_pfcmpge (v2sf, v2sf) 7234v2si __builtin_ia32_pfcmpgt (v2sf, v2sf) 7235v2sf __builtin_ia32_pfmax (v2sf, v2sf) 7236v2sf __builtin_ia32_pfmin (v2sf, v2sf) 7237v2sf __builtin_ia32_pfmul (v2sf, v2sf) 7238v2sf __builtin_ia32_pfrcp (v2sf) 7239v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf) 7240v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf) 7241v2sf __builtin_ia32_pfrsqrt (v2sf) 7242v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf) 7243v2sf __builtin_ia32_pfsub (v2sf, v2sf) 7244v2sf __builtin_ia32_pfsubr (v2sf, v2sf) 7245v2sf __builtin_ia32_pi2fd (v2si) 7246v4hi __builtin_ia32_pmulhrw (v4hi, v4hi) 7247@end smallexample 7248 7249The following built-in functions are available when both @option{-m3dnow} 7250and @option{-march=athlon} are used. All of them generate the machine 7251instruction that is part of the name. 7252 7253@smallexample 7254v2si __builtin_ia32_pf2iw (v2sf) 7255v2sf __builtin_ia32_pfnacc (v2sf, v2sf) 7256v2sf __builtin_ia32_pfpnacc (v2sf, v2sf) 7257v2sf __builtin_ia32_pi2fw (v2si) 7258v2sf __builtin_ia32_pswapdsf (v2sf) 7259v2si __builtin_ia32_pswapdsi (v2si) 7260@end smallexample 7261 7262@node MIPS DSP Built-in Functions 7263@subsection MIPS DSP Built-in Functions 7264 7265The MIPS DSP Application-Specific Extension (ASE) includes new 7266instructions that are designed to improve the performance of DSP and 7267media applications. It provides instructions that operate on packed 72688-bit integer data, Q15 fractional data and Q31 fractional data. 7269 7270GCC supports MIPS DSP operations using both the generic 7271vector extensions (@pxref{Vector Extensions}) and a collection of 7272MIPS-specific built-in functions. Both kinds of support are 7273enabled by the @option{-mdsp} command-line option. 7274 7275At present, GCC only provides support for operations on 32-bit 7276vectors. The vector type associated with 8-bit integer data is 7277usually called @code{v4i8} and the vector type associated with Q15 is 7278usually called @code{v2q15}. They can be defined in C as follows: 7279 7280@smallexample 7281typedef char v4i8 __attribute__ ((vector_size(4))); 7282typedef short v2q15 __attribute__ ((vector_size(4))); 7283@end smallexample 7284 7285@code{v4i8} and @code{v2q15} values are initialized in the same way as 7286aggregates. For example: 7287 7288@smallexample 7289v4i8 a = @{1, 2, 3, 4@}; 7290v4i8 b; 7291b = (v4i8) @{5, 6, 7, 8@}; 7292 7293v2q15 c = @{0x0fcb, 0x3a75@}; 7294v2q15 d; 7295d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@}; 7296@end smallexample 7297 7298@emph{Note:} The CPU's endianness determines the order in which values 7299are packed. On little-endian targets, the first value is the least 7300significant and the last value is the most significant. The opposite 7301order applies to big-endian targets. For example, the code above will 7302set the lowest byte of @code{a} to @code{1} on little-endian targets 7303and @code{4} on big-endian targets. 7304 7305@emph{Note:} Q15 and Q31 values must be initialized with their integer 7306representation. As shown in this example, the integer representation 7307of a Q15 value can be obtained by multiplying the fractional value by 7308@code{0x1.0p15}. The equivalent for Q31 values is to multiply by 7309@code{0x1.0p31}. 7310 7311The table below lists the @code{v4i8} and @code{v2q15} operations for which 7312hardware support exists. @code{a} and @code{b} are @code{v4i8} values, 7313and @code{c} and @code{d} are @code{v2q15} values. 7314 7315@multitable @columnfractions .50 .50 7316@item C code @tab MIPS instruction 7317@item @code{a + b} @tab @code{addu.qb} 7318@item @code{c + d} @tab @code{addq.ph} 7319@item @code{a - b} @tab @code{subu.qb} 7320@item @code{c - d} @tab @code{subq.ph} 7321@end multitable 7322 7323It is easier to describe the DSP built-in functions if we first define 7324the following types: 7325 7326@smallexample 7327typedef int q31; 7328typedef int i32; 7329typedef long long a64; 7330@end smallexample 7331 7332@code{q31} and @code{i32} are actually the same as @code{int}, but we 7333use @code{q31} to indicate a Q31 fractional value and @code{i32} to 7334indicate a 32-bit integer value. Similarly, @code{a64} is the same as 7335@code{long long}, but we use @code{a64} to indicate values that will 7336be placed in one of the four DSP accumulators (@code{$ac0}, 7337@code{$ac1}, @code{$ac2} or @code{$ac3}). 7338 7339Also, some built-in functions prefer or require immediate numbers as 7340parameters, because the corresponding DSP instructions accept both immediate 7341numbers and register operands, or accept immediate numbers only. The 7342immediate parameters are listed as follows. 7343 7344@smallexample 7345imm0_7: 0 to 7. 7346imm0_15: 0 to 15. 7347imm0_31: 0 to 31. 7348imm0_63: 0 to 63. 7349imm0_255: 0 to 255. 7350imm_n32_31: -32 to 31. 7351imm_n512_511: -512 to 511. 7352@end smallexample 7353 7354The following built-in functions map directly to a particular MIPS DSP 7355instruction. Please refer to the architecture specification 7356for details on what each instruction does. 7357 7358@smallexample 7359v2q15 __builtin_mips_addq_ph (v2q15, v2q15) 7360v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15) 7361q31 __builtin_mips_addq_s_w (q31, q31) 7362v4i8 __builtin_mips_addu_qb (v4i8, v4i8) 7363v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8) 7364v2q15 __builtin_mips_subq_ph (v2q15, v2q15) 7365v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15) 7366q31 __builtin_mips_subq_s_w (q31, q31) 7367v4i8 __builtin_mips_subu_qb (v4i8, v4i8) 7368v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8) 7369i32 __builtin_mips_addsc (i32, i32) 7370i32 __builtin_mips_addwc (i32, i32) 7371i32 __builtin_mips_modsub (i32, i32) 7372i32 __builtin_mips_raddu_w_qb (v4i8) 7373v2q15 __builtin_mips_absq_s_ph (v2q15) 7374q31 __builtin_mips_absq_s_w (q31) 7375v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15) 7376v2q15 __builtin_mips_precrq_ph_w (q31, q31) 7377v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31) 7378v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15) 7379q31 __builtin_mips_preceq_w_phl (v2q15) 7380q31 __builtin_mips_preceq_w_phr (v2q15) 7381v2q15 __builtin_mips_precequ_ph_qbl (v4i8) 7382v2q15 __builtin_mips_precequ_ph_qbr (v4i8) 7383v2q15 __builtin_mips_precequ_ph_qbla (v4i8) 7384v2q15 __builtin_mips_precequ_ph_qbra (v4i8) 7385v2q15 __builtin_mips_preceu_ph_qbl (v4i8) 7386v2q15 __builtin_mips_preceu_ph_qbr (v4i8) 7387v2q15 __builtin_mips_preceu_ph_qbla (v4i8) 7388v2q15 __builtin_mips_preceu_ph_qbra (v4i8) 7389v4i8 __builtin_mips_shll_qb (v4i8, imm0_7) 7390v4i8 __builtin_mips_shll_qb (v4i8, i32) 7391v2q15 __builtin_mips_shll_ph (v2q15, imm0_15) 7392v2q15 __builtin_mips_shll_ph (v2q15, i32) 7393v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15) 7394v2q15 __builtin_mips_shll_s_ph (v2q15, i32) 7395q31 __builtin_mips_shll_s_w (q31, imm0_31) 7396q31 __builtin_mips_shll_s_w (q31, i32) 7397v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7) 7398v4i8 __builtin_mips_shrl_qb (v4i8, i32) 7399v2q15 __builtin_mips_shra_ph (v2q15, imm0_15) 7400v2q15 __builtin_mips_shra_ph (v2q15, i32) 7401v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15) 7402v2q15 __builtin_mips_shra_r_ph (v2q15, i32) 7403q31 __builtin_mips_shra_r_w (q31, imm0_31) 7404q31 __builtin_mips_shra_r_w (q31, i32) 7405v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15) 7406v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15) 7407v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15) 7408q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15) 7409q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15) 7410a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8) 7411a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8) 7412a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8) 7413a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8) 7414a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15) 7415a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31) 7416a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15) 7417a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31) 7418a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15) 7419a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15) 7420a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15) 7421a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15) 7422a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15) 7423i32 __builtin_mips_bitrev (i32) 7424i32 __builtin_mips_insv (i32, i32) 7425v4i8 __builtin_mips_repl_qb (imm0_255) 7426v4i8 __builtin_mips_repl_qb (i32) 7427v2q15 __builtin_mips_repl_ph (imm_n512_511) 7428v2q15 __builtin_mips_repl_ph (i32) 7429void __builtin_mips_cmpu_eq_qb (v4i8, v4i8) 7430void __builtin_mips_cmpu_lt_qb (v4i8, v4i8) 7431void __builtin_mips_cmpu_le_qb (v4i8, v4i8) 7432i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8) 7433i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8) 7434i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8) 7435void __builtin_mips_cmp_eq_ph (v2q15, v2q15) 7436void __builtin_mips_cmp_lt_ph (v2q15, v2q15) 7437void __builtin_mips_cmp_le_ph (v2q15, v2q15) 7438v4i8 __builtin_mips_pick_qb (v4i8, v4i8) 7439v2q15 __builtin_mips_pick_ph (v2q15, v2q15) 7440v2q15 __builtin_mips_packrl_ph (v2q15, v2q15) 7441i32 __builtin_mips_extr_w (a64, imm0_31) 7442i32 __builtin_mips_extr_w (a64, i32) 7443i32 __builtin_mips_extr_r_w (a64, imm0_31) 7444i32 __builtin_mips_extr_s_h (a64, i32) 7445i32 __builtin_mips_extr_rs_w (a64, imm0_31) 7446i32 __builtin_mips_extr_rs_w (a64, i32) 7447i32 __builtin_mips_extr_s_h (a64, imm0_31) 7448i32 __builtin_mips_extr_r_w (a64, i32) 7449i32 __builtin_mips_extp (a64, imm0_31) 7450i32 __builtin_mips_extp (a64, i32) 7451i32 __builtin_mips_extpdp (a64, imm0_31) 7452i32 __builtin_mips_extpdp (a64, i32) 7453a64 __builtin_mips_shilo (a64, imm_n32_31) 7454a64 __builtin_mips_shilo (a64, i32) 7455a64 __builtin_mips_mthlip (a64, i32) 7456void __builtin_mips_wrdsp (i32, imm0_63) 7457i32 __builtin_mips_rddsp (imm0_63) 7458i32 __builtin_mips_lbux (void *, i32) 7459i32 __builtin_mips_lhx (void *, i32) 7460i32 __builtin_mips_lwx (void *, i32) 7461i32 __builtin_mips_bposge32 (void) 7462@end smallexample 7463 7464@node MIPS Paired-Single Support 7465@subsection MIPS Paired-Single Support 7466 7467The MIPS64 architecture includes a number of instructions that 7468operate on pairs of single-precision floating-point values. 7469Each pair is packed into a 64-bit floating-point register, 7470with one element being designated the ``upper half'' and 7471the other being designated the ``lower half''. 7472 7473GCC supports paired-single operations using both the generic 7474vector extensions (@pxref{Vector Extensions}) and a collection of 7475MIPS-specific built-in functions. Both kinds of support are 7476enabled by the @option{-mpaired-single} command-line option. 7477 7478The vector type associated with paired-single values is usually 7479called @code{v2sf}. It can be defined in C as follows: 7480 7481@smallexample 7482typedef float v2sf __attribute__ ((vector_size (8))); 7483@end smallexample 7484 7485@code{v2sf} values are initialized in the same way as aggregates. 7486For example: 7487 7488@smallexample 7489v2sf a = @{1.5, 9.1@}; 7490v2sf b; 7491float e, f; 7492b = (v2sf) @{e, f@}; 7493@end smallexample 7494 7495@emph{Note:} The CPU's endianness determines which value is stored in 7496the upper half of a register and which value is stored in the lower half. 7497On little-endian targets, the first value is the lower one and the second 7498value is the upper one. The opposite order applies to big-endian targets. 7499For example, the code above will set the lower half of @code{a} to 7500@code{1.5} on little-endian targets and @code{9.1} on big-endian targets. 7501 7502@menu 7503* Paired-Single Arithmetic:: 7504* Paired-Single Built-in Functions:: 7505* MIPS-3D Built-in Functions:: 7506@end menu 7507 7508@node Paired-Single Arithmetic 7509@subsubsection Paired-Single Arithmetic 7510 7511The table below lists the @code{v2sf} operations for which hardware 7512support exists. @code{a}, @code{b} and @code{c} are @code{v2sf} 7513values and @code{x} is an integral value. 7514 7515@multitable @columnfractions .50 .50 7516@item C code @tab MIPS instruction 7517@item @code{a + b} @tab @code{add.ps} 7518@item @code{a - b} @tab @code{sub.ps} 7519@item @code{-a} @tab @code{neg.ps} 7520@item @code{a * b} @tab @code{mul.ps} 7521@item @code{a * b + c} @tab @code{madd.ps} 7522@item @code{a * b - c} @tab @code{msub.ps} 7523@item @code{-(a * b + c)} @tab @code{nmadd.ps} 7524@item @code{-(a * b - c)} @tab @code{nmsub.ps} 7525@item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps} 7526@end multitable 7527 7528Note that the multiply-accumulate instructions can be disabled 7529using the command-line option @code{-mno-fused-madd}. 7530 7531@node Paired-Single Built-in Functions 7532@subsubsection Paired-Single Built-in Functions 7533 7534The following paired-single functions map directly to a particular 7535MIPS instruction. Please refer to the architecture specification 7536for details on what each instruction does. 7537 7538@table @code 7539@item v2sf __builtin_mips_pll_ps (v2sf, v2sf) 7540Pair lower lower (@code{pll.ps}). 7541 7542@item v2sf __builtin_mips_pul_ps (v2sf, v2sf) 7543Pair upper lower (@code{pul.ps}). 7544 7545@item v2sf __builtin_mips_plu_ps (v2sf, v2sf) 7546Pair lower upper (@code{plu.ps}). 7547 7548@item v2sf __builtin_mips_puu_ps (v2sf, v2sf) 7549Pair upper upper (@code{puu.ps}). 7550 7551@item v2sf __builtin_mips_cvt_ps_s (float, float) 7552Convert pair to paired single (@code{cvt.ps.s}). 7553 7554@item float __builtin_mips_cvt_s_pl (v2sf) 7555Convert pair lower to single (@code{cvt.s.pl}). 7556 7557@item float __builtin_mips_cvt_s_pu (v2sf) 7558Convert pair upper to single (@code{cvt.s.pu}). 7559 7560@item v2sf __builtin_mips_abs_ps (v2sf) 7561Absolute value (@code{abs.ps}). 7562 7563@item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int) 7564Align variable (@code{alnv.ps}). 7565 7566@emph{Note:} The value of the third parameter must be 0 or 4 7567modulo 8, otherwise the result will be unpredictable. Please read the 7568instruction description for details. 7569@end table 7570 7571The following multi-instruction functions are also available. 7572In each case, @var{cond} can be any of the 16 floating-point conditions: 7573@code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult}, 7574@code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl}, 7575@code{lt}, @code{nge}, @code{le} or @code{ngt}. 7576 7577@table @code 7578@item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7579@itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7580Conditional move based on floating point comparison (@code{c.@var{cond}.ps}, 7581@code{movt.ps}/@code{movf.ps}). 7582 7583The @code{movt} functions return the value @var{x} computed by: 7584 7585@smallexample 7586c.@var{cond}.ps @var{cc},@var{a},@var{b} 7587mov.ps @var{x},@var{c} 7588movt.ps @var{x},@var{d},@var{cc} 7589@end smallexample 7590 7591The @code{movf} functions are similar but use @code{movf.ps} instead 7592of @code{movt.ps}. 7593 7594@item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7595@itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7596Comparison of two paired-single values (@code{c.@var{cond}.ps}, 7597@code{bc1t}/@code{bc1f}). 7598 7599These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps} 7600and return either the upper or lower half of the result. For example: 7601 7602@smallexample 7603v2sf a, b; 7604if (__builtin_mips_upper_c_eq_ps (a, b)) 7605 upper_halves_are_equal (); 7606else 7607 upper_halves_are_unequal (); 7608 7609if (__builtin_mips_lower_c_eq_ps (a, b)) 7610 lower_halves_are_equal (); 7611else 7612 lower_halves_are_unequal (); 7613@end smallexample 7614@end table 7615 7616@node MIPS-3D Built-in Functions 7617@subsubsection MIPS-3D Built-in Functions 7618 7619The MIPS-3D Application-Specific Extension (ASE) includes additional 7620paired-single instructions that are designed to improve the performance 7621of 3D graphics operations. Support for these instructions is controlled 7622by the @option{-mips3d} command-line option. 7623 7624The functions listed below map directly to a particular MIPS-3D 7625instruction. Please refer to the architecture specification for 7626more details on what each instruction does. 7627 7628@table @code 7629@item v2sf __builtin_mips_addr_ps (v2sf, v2sf) 7630Reduction add (@code{addr.ps}). 7631 7632@item v2sf __builtin_mips_mulr_ps (v2sf, v2sf) 7633Reduction multiply (@code{mulr.ps}). 7634 7635@item v2sf __builtin_mips_cvt_pw_ps (v2sf) 7636Convert paired single to paired word (@code{cvt.pw.ps}). 7637 7638@item v2sf __builtin_mips_cvt_ps_pw (v2sf) 7639Convert paired word to paired single (@code{cvt.ps.pw}). 7640 7641@item float __builtin_mips_recip1_s (float) 7642@itemx double __builtin_mips_recip1_d (double) 7643@itemx v2sf __builtin_mips_recip1_ps (v2sf) 7644Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}). 7645 7646@item float __builtin_mips_recip2_s (float, float) 7647@itemx double __builtin_mips_recip2_d (double, double) 7648@itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf) 7649Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}). 7650 7651@item float __builtin_mips_rsqrt1_s (float) 7652@itemx double __builtin_mips_rsqrt1_d (double) 7653@itemx v2sf __builtin_mips_rsqrt1_ps (v2sf) 7654Reduced precision reciprocal square root (sequence step 1) 7655(@code{rsqrt1.@var{fmt}}). 7656 7657@item float __builtin_mips_rsqrt2_s (float, float) 7658@itemx double __builtin_mips_rsqrt2_d (double, double) 7659@itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf) 7660Reduced precision reciprocal square root (sequence step 2) 7661(@code{rsqrt2.@var{fmt}}). 7662@end table 7663 7664The following multi-instruction functions are also available. 7665In each case, @var{cond} can be any of the 16 floating-point conditions: 7666@code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult}, 7667@code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, 7668@code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}. 7669 7670@table @code 7671@item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b}) 7672@itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b}) 7673Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}}, 7674@code{bc1t}/@code{bc1f}). 7675 7676These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s} 7677or @code{cabs.@var{cond}.d} and return the result as a boolean value. 7678For example: 7679 7680@smallexample 7681float a, b; 7682if (__builtin_mips_cabs_eq_s (a, b)) 7683 true (); 7684else 7685 false (); 7686@end smallexample 7687 7688@item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7689@itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7690Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps}, 7691@code{bc1t}/@code{bc1f}). 7692 7693These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps} 7694and return either the upper or lower half of the result. For example: 7695 7696@smallexample 7697v2sf a, b; 7698if (__builtin_mips_upper_cabs_eq_ps (a, b)) 7699 upper_halves_are_equal (); 7700else 7701 upper_halves_are_unequal (); 7702 7703if (__builtin_mips_lower_cabs_eq_ps (a, b)) 7704 lower_halves_are_equal (); 7705else 7706 lower_halves_are_unequal (); 7707@end smallexample 7708 7709@item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7710@itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7711Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps}, 7712@code{movt.ps}/@code{movf.ps}). 7713 7714The @code{movt} functions return the value @var{x} computed by: 7715 7716@smallexample 7717cabs.@var{cond}.ps @var{cc},@var{a},@var{b} 7718mov.ps @var{x},@var{c} 7719movt.ps @var{x},@var{d},@var{cc} 7720@end smallexample 7721 7722The @code{movf} functions are similar but use @code{movf.ps} instead 7723of @code{movt.ps}. 7724 7725@item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7726@itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7727@itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7728@itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 7729Comparison of two paired-single values 7730(@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps}, 7731@code{bc1any2t}/@code{bc1any2f}). 7732 7733These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps} 7734or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either 7735result is true and the @code{all} forms return true if both results are true. 7736For example: 7737 7738@smallexample 7739v2sf a, b; 7740if (__builtin_mips_any_c_eq_ps (a, b)) 7741 one_is_true (); 7742else 7743 both_are_false (); 7744 7745if (__builtin_mips_all_c_eq_ps (a, b)) 7746 both_are_true (); 7747else 7748 one_is_false (); 7749@end smallexample 7750 7751@item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7752@itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7753@itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7754@itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 7755Comparison of four paired-single values 7756(@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps}, 7757@code{bc1any4t}/@code{bc1any4f}). 7758 7759These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps} 7760to compare @var{a} with @var{b} and to compare @var{c} with @var{d}. 7761The @code{any} forms return true if any of the four results are true 7762and the @code{all} forms return true if all four results are true. 7763For example: 7764 7765@smallexample 7766v2sf a, b, c, d; 7767if (__builtin_mips_any_c_eq_4s (a, b, c, d)) 7768 some_are_true (); 7769else 7770 all_are_false (); 7771 7772if (__builtin_mips_all_c_eq_4s (a, b, c, d)) 7773 all_are_true (); 7774else 7775 some_are_false (); 7776@end smallexample 7777@end table 7778 7779@node PowerPC AltiVec Built-in Functions 7780@subsection PowerPC AltiVec Built-in Functions 7781 7782GCC provides an interface for the PowerPC family of processors to access 7783the AltiVec operations described in Motorola's AltiVec Programming 7784Interface Manual. The interface is made available by including 7785@code{<altivec.h>} and using @option{-maltivec} and 7786@option{-mabi=altivec}. The interface supports the following vector 7787types. 7788 7789@smallexample 7790vector unsigned char 7791vector signed char 7792vector bool char 7793 7794vector unsigned short 7795vector signed short 7796vector bool short 7797vector pixel 7798 7799vector unsigned int 7800vector signed int 7801vector bool int 7802vector float 7803@end smallexample 7804 7805GCC's implementation of the high-level language interface available from 7806C and C++ code differs from Motorola's documentation in several ways. 7807 7808@itemize @bullet 7809 7810@item 7811A vector constant is a list of constant expressions within curly braces. 7812 7813@item 7814A vector initializer requires no cast if the vector constant is of the 7815same type as the variable it is initializing. 7816 7817@item 7818If @code{signed} or @code{unsigned} is omitted, the signedness of the 7819vector type is the default signedness of the base type. The default 7820varies depending on the operating system, so a portable program should 7821always specify the signedness. 7822 7823@item 7824Compiling with @option{-maltivec} adds keywords @code{__vector}, 7825@code{__pixel}, and @code{__bool}. Macros @option{vector}, 7826@code{pixel}, and @code{bool} are defined in @code{<altivec.h>} and can 7827be undefined. 7828 7829@item 7830GCC allows using a @code{typedef} name as the type specifier for a 7831vector type. 7832 7833@item 7834For C, overloaded functions are implemented with macros so the following 7835does not work: 7836 7837@smallexample 7838 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo); 7839@end smallexample 7840 7841Since @code{vec_add} is a macro, the vector constant in the example 7842is treated as four separate arguments. Wrap the entire argument in 7843parentheses for this to work. 7844@end itemize 7845 7846@emph{Note:} Only the @code{<altivec.h>} interface is supported. 7847Internally, GCC uses built-in functions to achieve the functionality in 7848the aforementioned header file, but they are not supported and are 7849subject to change without notice. 7850 7851The following interfaces are supported for the generic and specific 7852AltiVec operations and the AltiVec predicates. In cases where there 7853is a direct mapping between generic and specific operations, only the 7854generic names are shown here, although the specific operations can also 7855be used. 7856 7857Arguments that are documented as @code{const int} require literal 7858integral values within the range required for that operation. 7859 7860@smallexample 7861vector signed char vec_abs (vector signed char); 7862vector signed short vec_abs (vector signed short); 7863vector signed int vec_abs (vector signed int); 7864vector float vec_abs (vector float); 7865 7866vector signed char vec_abss (vector signed char); 7867vector signed short vec_abss (vector signed short); 7868vector signed int vec_abss (vector signed int); 7869 7870vector signed char vec_add (vector bool char, vector signed char); 7871vector signed char vec_add (vector signed char, vector bool char); 7872vector signed char vec_add (vector signed char, vector signed char); 7873vector unsigned char vec_add (vector bool char, vector unsigned char); 7874vector unsigned char vec_add (vector unsigned char, vector bool char); 7875vector unsigned char vec_add (vector unsigned char, 7876 vector unsigned char); 7877vector signed short vec_add (vector bool short, vector signed short); 7878vector signed short vec_add (vector signed short, vector bool short); 7879vector signed short vec_add (vector signed short, vector signed short); 7880vector unsigned short vec_add (vector bool short, 7881 vector unsigned short); 7882vector unsigned short vec_add (vector unsigned short, 7883 vector bool short); 7884vector unsigned short vec_add (vector unsigned short, 7885 vector unsigned short); 7886vector signed int vec_add (vector bool int, vector signed int); 7887vector signed int vec_add (vector signed int, vector bool int); 7888vector signed int vec_add (vector signed int, vector signed int); 7889vector unsigned int vec_add (vector bool int, vector unsigned int); 7890vector unsigned int vec_add (vector unsigned int, vector bool int); 7891vector unsigned int vec_add (vector unsigned int, vector unsigned int); 7892vector float vec_add (vector float, vector float); 7893 7894vector float vec_vaddfp (vector float, vector float); 7895 7896vector signed int vec_vadduwm (vector bool int, vector signed int); 7897vector signed int vec_vadduwm (vector signed int, vector bool int); 7898vector signed int vec_vadduwm (vector signed int, vector signed int); 7899vector unsigned int vec_vadduwm (vector bool int, vector unsigned int); 7900vector unsigned int vec_vadduwm (vector unsigned int, vector bool int); 7901vector unsigned int vec_vadduwm (vector unsigned int, 7902 vector unsigned int); 7903 7904vector signed short vec_vadduhm (vector bool short, 7905 vector signed short); 7906vector signed short vec_vadduhm (vector signed short, 7907 vector bool short); 7908vector signed short vec_vadduhm (vector signed short, 7909 vector signed short); 7910vector unsigned short vec_vadduhm (vector bool short, 7911 vector unsigned short); 7912vector unsigned short vec_vadduhm (vector unsigned short, 7913 vector bool short); 7914vector unsigned short vec_vadduhm (vector unsigned short, 7915 vector unsigned short); 7916 7917vector signed char vec_vaddubm (vector bool char, vector signed char); 7918vector signed char vec_vaddubm (vector signed char, vector bool char); 7919vector signed char vec_vaddubm (vector signed char, vector signed char); 7920vector unsigned char vec_vaddubm (vector bool char, 7921 vector unsigned char); 7922vector unsigned char vec_vaddubm (vector unsigned char, 7923 vector bool char); 7924vector unsigned char vec_vaddubm (vector unsigned char, 7925 vector unsigned char); 7926 7927vector unsigned int vec_addc (vector unsigned int, vector unsigned int); 7928 7929vector unsigned char vec_adds (vector bool char, vector unsigned char); 7930vector unsigned char vec_adds (vector unsigned char, vector bool char); 7931vector unsigned char vec_adds (vector unsigned char, 7932 vector unsigned char); 7933vector signed char vec_adds (vector bool char, vector signed char); 7934vector signed char vec_adds (vector signed char, vector bool char); 7935vector signed char vec_adds (vector signed char, vector signed char); 7936vector unsigned short vec_adds (vector bool short, 7937 vector unsigned short); 7938vector unsigned short vec_adds (vector unsigned short, 7939 vector bool short); 7940vector unsigned short vec_adds (vector unsigned short, 7941 vector unsigned short); 7942vector signed short vec_adds (vector bool short, vector signed short); 7943vector signed short vec_adds (vector signed short, vector bool short); 7944vector signed short vec_adds (vector signed short, vector signed short); 7945vector unsigned int vec_adds (vector bool int, vector unsigned int); 7946vector unsigned int vec_adds (vector unsigned int, vector bool int); 7947vector unsigned int vec_adds (vector unsigned int, vector unsigned int); 7948vector signed int vec_adds (vector bool int, vector signed int); 7949vector signed int vec_adds (vector signed int, vector bool int); 7950vector signed int vec_adds (vector signed int, vector signed int); 7951 7952vector signed int vec_vaddsws (vector bool int, vector signed int); 7953vector signed int vec_vaddsws (vector signed int, vector bool int); 7954vector signed int vec_vaddsws (vector signed int, vector signed int); 7955 7956vector unsigned int vec_vadduws (vector bool int, vector unsigned int); 7957vector unsigned int vec_vadduws (vector unsigned int, vector bool int); 7958vector unsigned int vec_vadduws (vector unsigned int, 7959 vector unsigned int); 7960 7961vector signed short vec_vaddshs (vector bool short, 7962 vector signed short); 7963vector signed short vec_vaddshs (vector signed short, 7964 vector bool short); 7965vector signed short vec_vaddshs (vector signed short, 7966 vector signed short); 7967 7968vector unsigned short vec_vadduhs (vector bool short, 7969 vector unsigned short); 7970vector unsigned short vec_vadduhs (vector unsigned short, 7971 vector bool short); 7972vector unsigned short vec_vadduhs (vector unsigned short, 7973 vector unsigned short); 7974 7975vector signed char vec_vaddsbs (vector bool char, vector signed char); 7976vector signed char vec_vaddsbs (vector signed char, vector bool char); 7977vector signed char vec_vaddsbs (vector signed char, vector signed char); 7978 7979vector unsigned char vec_vaddubs (vector bool char, 7980 vector unsigned char); 7981vector unsigned char vec_vaddubs (vector unsigned char, 7982 vector bool char); 7983vector unsigned char vec_vaddubs (vector unsigned char, 7984 vector unsigned char); 7985 7986vector float vec_and (vector float, vector float); 7987vector float vec_and (vector float, vector bool int); 7988vector float vec_and (vector bool int, vector float); 7989vector bool int vec_and (vector bool int, vector bool int); 7990vector signed int vec_and (vector bool int, vector signed int); 7991vector signed int vec_and (vector signed int, vector bool int); 7992vector signed int vec_and (vector signed int, vector signed int); 7993vector unsigned int vec_and (vector bool int, vector unsigned int); 7994vector unsigned int vec_and (vector unsigned int, vector bool int); 7995vector unsigned int vec_and (vector unsigned int, vector unsigned int); 7996vector bool short vec_and (vector bool short, vector bool short); 7997vector signed short vec_and (vector bool short, vector signed short); 7998vector signed short vec_and (vector signed short, vector bool short); 7999vector signed short vec_and (vector signed short, vector signed short); 8000vector unsigned short vec_and (vector bool short, 8001 vector unsigned short); 8002vector unsigned short vec_and (vector unsigned short, 8003 vector bool short); 8004vector unsigned short vec_and (vector unsigned short, 8005 vector unsigned short); 8006vector signed char vec_and (vector bool char, vector signed char); 8007vector bool char vec_and (vector bool char, vector bool char); 8008vector signed char vec_and (vector signed char, vector bool char); 8009vector signed char vec_and (vector signed char, vector signed char); 8010vector unsigned char vec_and (vector bool char, vector unsigned char); 8011vector unsigned char vec_and (vector unsigned char, vector bool char); 8012vector unsigned char vec_and (vector unsigned char, 8013 vector unsigned char); 8014 8015vector float vec_andc (vector float, vector float); 8016vector float vec_andc (vector float, vector bool int); 8017vector float vec_andc (vector bool int, vector float); 8018vector bool int vec_andc (vector bool int, vector bool int); 8019vector signed int vec_andc (vector bool int, vector signed int); 8020vector signed int vec_andc (vector signed int, vector bool int); 8021vector signed int vec_andc (vector signed int, vector signed int); 8022vector unsigned int vec_andc (vector bool int, vector unsigned int); 8023vector unsigned int vec_andc (vector unsigned int, vector bool int); 8024vector unsigned int vec_andc (vector unsigned int, vector unsigned int); 8025vector bool short vec_andc (vector bool short, vector bool short); 8026vector signed short vec_andc (vector bool short, vector signed short); 8027vector signed short vec_andc (vector signed short, vector bool short); 8028vector signed short vec_andc (vector signed short, vector signed short); 8029vector unsigned short vec_andc (vector bool short, 8030 vector unsigned short); 8031vector unsigned short vec_andc (vector unsigned short, 8032 vector bool short); 8033vector unsigned short vec_andc (vector unsigned short, 8034 vector unsigned short); 8035vector signed char vec_andc (vector bool char, vector signed char); 8036vector bool char vec_andc (vector bool char, vector bool char); 8037vector signed char vec_andc (vector signed char, vector bool char); 8038vector signed char vec_andc (vector signed char, vector signed char); 8039vector unsigned char vec_andc (vector bool char, vector unsigned char); 8040vector unsigned char vec_andc (vector unsigned char, vector bool char); 8041vector unsigned char vec_andc (vector unsigned char, 8042 vector unsigned char); 8043 8044vector unsigned char vec_avg (vector unsigned char, 8045 vector unsigned char); 8046vector signed char vec_avg (vector signed char, vector signed char); 8047vector unsigned short vec_avg (vector unsigned short, 8048 vector unsigned short); 8049vector signed short vec_avg (vector signed short, vector signed short); 8050vector unsigned int vec_avg (vector unsigned int, vector unsigned int); 8051vector signed int vec_avg (vector signed int, vector signed int); 8052 8053vector signed int vec_vavgsw (vector signed int, vector signed int); 8054 8055vector unsigned int vec_vavguw (vector unsigned int, 8056 vector unsigned int); 8057 8058vector signed short vec_vavgsh (vector signed short, 8059 vector signed short); 8060 8061vector unsigned short vec_vavguh (vector unsigned short, 8062 vector unsigned short); 8063 8064vector signed char vec_vavgsb (vector signed char, vector signed char); 8065 8066vector unsigned char vec_vavgub (vector unsigned char, 8067 vector unsigned char); 8068 8069vector float vec_ceil (vector float); 8070 8071vector signed int vec_cmpb (vector float, vector float); 8072 8073vector bool char vec_cmpeq (vector signed char, vector signed char); 8074vector bool char vec_cmpeq (vector unsigned char, vector unsigned char); 8075vector bool short vec_cmpeq (vector signed short, vector signed short); 8076vector bool short vec_cmpeq (vector unsigned short, 8077 vector unsigned short); 8078vector bool int vec_cmpeq (vector signed int, vector signed int); 8079vector bool int vec_cmpeq (vector unsigned int, vector unsigned int); 8080vector bool int vec_cmpeq (vector float, vector float); 8081 8082vector bool int vec_vcmpeqfp (vector float, vector float); 8083 8084vector bool int vec_vcmpequw (vector signed int, vector signed int); 8085vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int); 8086 8087vector bool short vec_vcmpequh (vector signed short, 8088 vector signed short); 8089vector bool short vec_vcmpequh (vector unsigned short, 8090 vector unsigned short); 8091 8092vector bool char vec_vcmpequb (vector signed char, vector signed char); 8093vector bool char vec_vcmpequb (vector unsigned char, 8094 vector unsigned char); 8095 8096vector bool int vec_cmpge (vector float, vector float); 8097 8098vector bool char vec_cmpgt (vector unsigned char, vector unsigned char); 8099vector bool char vec_cmpgt (vector signed char, vector signed char); 8100vector bool short vec_cmpgt (vector unsigned short, 8101 vector unsigned short); 8102vector bool short vec_cmpgt (vector signed short, vector signed short); 8103vector bool int vec_cmpgt (vector unsigned int, vector unsigned int); 8104vector bool int vec_cmpgt (vector signed int, vector signed int); 8105vector bool int vec_cmpgt (vector float, vector float); 8106 8107vector bool int vec_vcmpgtfp (vector float, vector float); 8108 8109vector bool int vec_vcmpgtsw (vector signed int, vector signed int); 8110 8111vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int); 8112 8113vector bool short vec_vcmpgtsh (vector signed short, 8114 vector signed short); 8115 8116vector bool short vec_vcmpgtuh (vector unsigned short, 8117 vector unsigned short); 8118 8119vector bool char vec_vcmpgtsb (vector signed char, vector signed char); 8120 8121vector bool char vec_vcmpgtub (vector unsigned char, 8122 vector unsigned char); 8123 8124vector bool int vec_cmple (vector float, vector float); 8125 8126vector bool char vec_cmplt (vector unsigned char, vector unsigned char); 8127vector bool char vec_cmplt (vector signed char, vector signed char); 8128vector bool short vec_cmplt (vector unsigned short, 8129 vector unsigned short); 8130vector bool short vec_cmplt (vector signed short, vector signed short); 8131vector bool int vec_cmplt (vector unsigned int, vector unsigned int); 8132vector bool int vec_cmplt (vector signed int, vector signed int); 8133vector bool int vec_cmplt (vector float, vector float); 8134 8135vector float vec_ctf (vector unsigned int, const int); 8136vector float vec_ctf (vector signed int, const int); 8137 8138vector float vec_vcfsx (vector signed int, const int); 8139 8140vector float vec_vcfux (vector unsigned int, const int); 8141 8142vector signed int vec_cts (vector float, const int); 8143 8144vector unsigned int vec_ctu (vector float, const int); 8145 8146void vec_dss (const int); 8147 8148void vec_dssall (void); 8149 8150void vec_dst (const vector unsigned char *, int, const int); 8151void vec_dst (const vector signed char *, int, const int); 8152void vec_dst (const vector bool char *, int, const int); 8153void vec_dst (const vector unsigned short *, int, const int); 8154void vec_dst (const vector signed short *, int, const int); 8155void vec_dst (const vector bool short *, int, const int); 8156void vec_dst (const vector pixel *, int, const int); 8157void vec_dst (const vector unsigned int *, int, const int); 8158void vec_dst (const vector signed int *, int, const int); 8159void vec_dst (const vector bool int *, int, const int); 8160void vec_dst (const vector float *, int, const int); 8161void vec_dst (const unsigned char *, int, const int); 8162void vec_dst (const signed char *, int, const int); 8163void vec_dst (const unsigned short *, int, const int); 8164void vec_dst (const short *, int, const int); 8165void vec_dst (const unsigned int *, int, const int); 8166void vec_dst (const int *, int, const int); 8167void vec_dst (const unsigned long *, int, const int); 8168void vec_dst (const long *, int, const int); 8169void vec_dst (const float *, int, const int); 8170 8171void vec_dstst (const vector unsigned char *, int, const int); 8172void vec_dstst (const vector signed char *, int, const int); 8173void vec_dstst (const vector bool char *, int, const int); 8174void vec_dstst (const vector unsigned short *, int, const int); 8175void vec_dstst (const vector signed short *, int, const int); 8176void vec_dstst (const vector bool short *, int, const int); 8177void vec_dstst (const vector pixel *, int, const int); 8178void vec_dstst (const vector unsigned int *, int, const int); 8179void vec_dstst (const vector signed int *, int, const int); 8180void vec_dstst (const vector bool int *, int, const int); 8181void vec_dstst (const vector float *, int, const int); 8182void vec_dstst (const unsigned char *, int, const int); 8183void vec_dstst (const signed char *, int, const int); 8184void vec_dstst (const unsigned short *, int, const int); 8185void vec_dstst (const short *, int, const int); 8186void vec_dstst (const unsigned int *, int, const int); 8187void vec_dstst (const int *, int, const int); 8188void vec_dstst (const unsigned long *, int, const int); 8189void vec_dstst (const long *, int, const int); 8190void vec_dstst (const float *, int, const int); 8191 8192void vec_dststt (const vector unsigned char *, int, const int); 8193void vec_dststt (const vector signed char *, int, const int); 8194void vec_dststt (const vector bool char *, int, const int); 8195void vec_dststt (const vector unsigned short *, int, const int); 8196void vec_dststt (const vector signed short *, int, const int); 8197void vec_dststt (const vector bool short *, int, const int); 8198void vec_dststt (const vector pixel *, int, const int); 8199void vec_dststt (const vector unsigned int *, int, const int); 8200void vec_dststt (const vector signed int *, int, const int); 8201void vec_dststt (const vector bool int *, int, const int); 8202void vec_dststt (const vector float *, int, const int); 8203void vec_dststt (const unsigned char *, int, const int); 8204void vec_dststt (const signed char *, int, const int); 8205void vec_dststt (const unsigned short *, int, const int); 8206void vec_dststt (const short *, int, const int); 8207void vec_dststt (const unsigned int *, int, const int); 8208void vec_dststt (const int *, int, const int); 8209void vec_dststt (const unsigned long *, int, const int); 8210void vec_dststt (const long *, int, const int); 8211void vec_dststt (const float *, int, const int); 8212 8213void vec_dstt (const vector unsigned char *, int, const int); 8214void vec_dstt (const vector signed char *, int, const int); 8215void vec_dstt (const vector bool char *, int, const int); 8216void vec_dstt (const vector unsigned short *, int, const int); 8217void vec_dstt (const vector signed short *, int, const int); 8218void vec_dstt (const vector bool short *, int, const int); 8219void vec_dstt (const vector pixel *, int, const int); 8220void vec_dstt (const vector unsigned int *, int, const int); 8221void vec_dstt (const vector signed int *, int, const int); 8222void vec_dstt (const vector bool int *, int, const int); 8223void vec_dstt (const vector float *, int, const int); 8224void vec_dstt (const unsigned char *, int, const int); 8225void vec_dstt (const signed char *, int, const int); 8226void vec_dstt (const unsigned short *, int, const int); 8227void vec_dstt (const short *, int, const int); 8228void vec_dstt (const unsigned int *, int, const int); 8229void vec_dstt (const int *, int, const int); 8230void vec_dstt (const unsigned long *, int, const int); 8231void vec_dstt (const long *, int, const int); 8232void vec_dstt (const float *, int, const int); 8233 8234vector float vec_expte (vector float); 8235 8236vector float vec_floor (vector float); 8237 8238vector float vec_ld (int, const vector float *); 8239vector float vec_ld (int, const float *); 8240vector bool int vec_ld (int, const vector bool int *); 8241vector signed int vec_ld (int, const vector signed int *); 8242vector signed int vec_ld (int, const int *); 8243vector signed int vec_ld (int, const long *); 8244vector unsigned int vec_ld (int, const vector unsigned int *); 8245vector unsigned int vec_ld (int, const unsigned int *); 8246vector unsigned int vec_ld (int, const unsigned long *); 8247vector bool short vec_ld (int, const vector bool short *); 8248vector pixel vec_ld (int, const vector pixel *); 8249vector signed short vec_ld (int, const vector signed short *); 8250vector signed short vec_ld (int, const short *); 8251vector unsigned short vec_ld (int, const vector unsigned short *); 8252vector unsigned short vec_ld (int, const unsigned short *); 8253vector bool char vec_ld (int, const vector bool char *); 8254vector signed char vec_ld (int, const vector signed char *); 8255vector signed char vec_ld (int, const signed char *); 8256vector unsigned char vec_ld (int, const vector unsigned char *); 8257vector unsigned char vec_ld (int, const unsigned char *); 8258 8259vector signed char vec_lde (int, const signed char *); 8260vector unsigned char vec_lde (int, const unsigned char *); 8261vector signed short vec_lde (int, const short *); 8262vector unsigned short vec_lde (int, const unsigned short *); 8263vector float vec_lde (int, const float *); 8264vector signed int vec_lde (int, const int *); 8265vector unsigned int vec_lde (int, const unsigned int *); 8266vector signed int vec_lde (int, const long *); 8267vector unsigned int vec_lde (int, const unsigned long *); 8268 8269vector float vec_lvewx (int, float *); 8270vector signed int vec_lvewx (int, int *); 8271vector unsigned int vec_lvewx (int, unsigned int *); 8272vector signed int vec_lvewx (int, long *); 8273vector unsigned int vec_lvewx (int, unsigned long *); 8274 8275vector signed short vec_lvehx (int, short *); 8276vector unsigned short vec_lvehx (int, unsigned short *); 8277 8278vector signed char vec_lvebx (int, char *); 8279vector unsigned char vec_lvebx (int, unsigned char *); 8280 8281vector float vec_ldl (int, const vector float *); 8282vector float vec_ldl (int, const float *); 8283vector bool int vec_ldl (int, const vector bool int *); 8284vector signed int vec_ldl (int, const vector signed int *); 8285vector signed int vec_ldl (int, const int *); 8286vector signed int vec_ldl (int, const long *); 8287vector unsigned int vec_ldl (int, const vector unsigned int *); 8288vector unsigned int vec_ldl (int, const unsigned int *); 8289vector unsigned int vec_ldl (int, const unsigned long *); 8290vector bool short vec_ldl (int, const vector bool short *); 8291vector pixel vec_ldl (int, const vector pixel *); 8292vector signed short vec_ldl (int, const vector signed short *); 8293vector signed short vec_ldl (int, const short *); 8294vector unsigned short vec_ldl (int, const vector unsigned short *); 8295vector unsigned short vec_ldl (int, const unsigned short *); 8296vector bool char vec_ldl (int, const vector bool char *); 8297vector signed char vec_ldl (int, const vector signed char *); 8298vector signed char vec_ldl (int, const signed char *); 8299vector unsigned char vec_ldl (int, const vector unsigned char *); 8300vector unsigned char vec_ldl (int, const unsigned char *); 8301 8302vector float vec_loge (vector float); 8303 8304vector unsigned char vec_lvsl (int, const volatile unsigned char *); 8305vector unsigned char vec_lvsl (int, const volatile signed char *); 8306vector unsigned char vec_lvsl (int, const volatile unsigned short *); 8307vector unsigned char vec_lvsl (int, const volatile short *); 8308vector unsigned char vec_lvsl (int, const volatile unsigned int *); 8309vector unsigned char vec_lvsl (int, const volatile int *); 8310vector unsigned char vec_lvsl (int, const volatile unsigned long *); 8311vector unsigned char vec_lvsl (int, const volatile long *); 8312vector unsigned char vec_lvsl (int, const volatile float *); 8313 8314vector unsigned char vec_lvsr (int, const volatile unsigned char *); 8315vector unsigned char vec_lvsr (int, const volatile signed char *); 8316vector unsigned char vec_lvsr (int, const volatile unsigned short *); 8317vector unsigned char vec_lvsr (int, const volatile short *); 8318vector unsigned char vec_lvsr (int, const volatile unsigned int *); 8319vector unsigned char vec_lvsr (int, const volatile int *); 8320vector unsigned char vec_lvsr (int, const volatile unsigned long *); 8321vector unsigned char vec_lvsr (int, const volatile long *); 8322vector unsigned char vec_lvsr (int, const volatile float *); 8323 8324vector float vec_madd (vector float, vector float, vector float); 8325 8326vector signed short vec_madds (vector signed short, 8327 vector signed short, 8328 vector signed short); 8329 8330vector unsigned char vec_max (vector bool char, vector unsigned char); 8331vector unsigned char vec_max (vector unsigned char, vector bool char); 8332vector unsigned char vec_max (vector unsigned char, 8333 vector unsigned char); 8334vector signed char vec_max (vector bool char, vector signed char); 8335vector signed char vec_max (vector signed char, vector bool char); 8336vector signed char vec_max (vector signed char, vector signed char); 8337vector unsigned short vec_max (vector bool short, 8338 vector unsigned short); 8339vector unsigned short vec_max (vector unsigned short, 8340 vector bool short); 8341vector unsigned short vec_max (vector unsigned short, 8342 vector unsigned short); 8343vector signed short vec_max (vector bool short, vector signed short); 8344vector signed short vec_max (vector signed short, vector bool short); 8345vector signed short vec_max (vector signed short, vector signed short); 8346vector unsigned int vec_max (vector bool int, vector unsigned int); 8347vector unsigned int vec_max (vector unsigned int, vector bool int); 8348vector unsigned int vec_max (vector unsigned int, vector unsigned int); 8349vector signed int vec_max (vector bool int, vector signed int); 8350vector signed int vec_max (vector signed int, vector bool int); 8351vector signed int vec_max (vector signed int, vector signed int); 8352vector float vec_max (vector float, vector float); 8353 8354vector float vec_vmaxfp (vector float, vector float); 8355 8356vector signed int vec_vmaxsw (vector bool int, vector signed int); 8357vector signed int vec_vmaxsw (vector signed int, vector bool int); 8358vector signed int vec_vmaxsw (vector signed int, vector signed int); 8359 8360vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int); 8361vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int); 8362vector unsigned int vec_vmaxuw (vector unsigned int, 8363 vector unsigned int); 8364 8365vector signed short vec_vmaxsh (vector bool short, vector signed short); 8366vector signed short vec_vmaxsh (vector signed short, vector bool short); 8367vector signed short vec_vmaxsh (vector signed short, 8368 vector signed short); 8369 8370vector unsigned short vec_vmaxuh (vector bool short, 8371 vector unsigned short); 8372vector unsigned short vec_vmaxuh (vector unsigned short, 8373 vector bool short); 8374vector unsigned short vec_vmaxuh (vector unsigned short, 8375 vector unsigned short); 8376 8377vector signed char vec_vmaxsb (vector bool char, vector signed char); 8378vector signed char vec_vmaxsb (vector signed char, vector bool char); 8379vector signed char vec_vmaxsb (vector signed char, vector signed char); 8380 8381vector unsigned char vec_vmaxub (vector bool char, 8382 vector unsigned char); 8383vector unsigned char vec_vmaxub (vector unsigned char, 8384 vector bool char); 8385vector unsigned char vec_vmaxub (vector unsigned char, 8386 vector unsigned char); 8387 8388vector bool char vec_mergeh (vector bool char, vector bool char); 8389vector signed char vec_mergeh (vector signed char, vector signed char); 8390vector unsigned char vec_mergeh (vector unsigned char, 8391 vector unsigned char); 8392vector bool short vec_mergeh (vector bool short, vector bool short); 8393vector pixel vec_mergeh (vector pixel, vector pixel); 8394vector signed short vec_mergeh (vector signed short, 8395 vector signed short); 8396vector unsigned short vec_mergeh (vector unsigned short, 8397 vector unsigned short); 8398vector float vec_mergeh (vector float, vector float); 8399vector bool int vec_mergeh (vector bool int, vector bool int); 8400vector signed int vec_mergeh (vector signed int, vector signed int); 8401vector unsigned int vec_mergeh (vector unsigned int, 8402 vector unsigned int); 8403 8404vector float vec_vmrghw (vector float, vector float); 8405vector bool int vec_vmrghw (vector bool int, vector bool int); 8406vector signed int vec_vmrghw (vector signed int, vector signed int); 8407vector unsigned int vec_vmrghw (vector unsigned int, 8408 vector unsigned int); 8409 8410vector bool short vec_vmrghh (vector bool short, vector bool short); 8411vector signed short vec_vmrghh (vector signed short, 8412 vector signed short); 8413vector unsigned short vec_vmrghh (vector unsigned short, 8414 vector unsigned short); 8415vector pixel vec_vmrghh (vector pixel, vector pixel); 8416 8417vector bool char vec_vmrghb (vector bool char, vector bool char); 8418vector signed char vec_vmrghb (vector signed char, vector signed char); 8419vector unsigned char vec_vmrghb (vector unsigned char, 8420 vector unsigned char); 8421 8422vector bool char vec_mergel (vector bool char, vector bool char); 8423vector signed char vec_mergel (vector signed char, vector signed char); 8424vector unsigned char vec_mergel (vector unsigned char, 8425 vector unsigned char); 8426vector bool short vec_mergel (vector bool short, vector bool short); 8427vector pixel vec_mergel (vector pixel, vector pixel); 8428vector signed short vec_mergel (vector signed short, 8429 vector signed short); 8430vector unsigned short vec_mergel (vector unsigned short, 8431 vector unsigned short); 8432vector float vec_mergel (vector float, vector float); 8433vector bool int vec_mergel (vector bool int, vector bool int); 8434vector signed int vec_mergel (vector signed int, vector signed int); 8435vector unsigned int vec_mergel (vector unsigned int, 8436 vector unsigned int); 8437 8438vector float vec_vmrglw (vector float, vector float); 8439vector signed int vec_vmrglw (vector signed int, vector signed int); 8440vector unsigned int vec_vmrglw (vector unsigned int, 8441 vector unsigned int); 8442vector bool int vec_vmrglw (vector bool int, vector bool int); 8443 8444vector bool short vec_vmrglh (vector bool short, vector bool short); 8445vector signed short vec_vmrglh (vector signed short, 8446 vector signed short); 8447vector unsigned short vec_vmrglh (vector unsigned short, 8448 vector unsigned short); 8449vector pixel vec_vmrglh (vector pixel, vector pixel); 8450 8451vector bool char vec_vmrglb (vector bool char, vector bool char); 8452vector signed char vec_vmrglb (vector signed char, vector signed char); 8453vector unsigned char vec_vmrglb (vector unsigned char, 8454 vector unsigned char); 8455 8456vector unsigned short vec_mfvscr (void); 8457 8458vector unsigned char vec_min (vector bool char, vector unsigned char); 8459vector unsigned char vec_min (vector unsigned char, vector bool char); 8460vector unsigned char vec_min (vector unsigned char, 8461 vector unsigned char); 8462vector signed char vec_min (vector bool char, vector signed char); 8463vector signed char vec_min (vector signed char, vector bool char); 8464vector signed char vec_min (vector signed char, vector signed char); 8465vector unsigned short vec_min (vector bool short, 8466 vector unsigned short); 8467vector unsigned short vec_min (vector unsigned short, 8468 vector bool short); 8469vector unsigned short vec_min (vector unsigned short, 8470 vector unsigned short); 8471vector signed short vec_min (vector bool short, vector signed short); 8472vector signed short vec_min (vector signed short, vector bool short); 8473vector signed short vec_min (vector signed short, vector signed short); 8474vector unsigned int vec_min (vector bool int, vector unsigned int); 8475vector unsigned int vec_min (vector unsigned int, vector bool int); 8476vector unsigned int vec_min (vector unsigned int, vector unsigned int); 8477vector signed int vec_min (vector bool int, vector signed int); 8478vector signed int vec_min (vector signed int, vector bool int); 8479vector signed int vec_min (vector signed int, vector signed int); 8480vector float vec_min (vector float, vector float); 8481 8482vector float vec_vminfp (vector float, vector float); 8483 8484vector signed int vec_vminsw (vector bool int, vector signed int); 8485vector signed int vec_vminsw (vector signed int, vector bool int); 8486vector signed int vec_vminsw (vector signed int, vector signed int); 8487 8488vector unsigned int vec_vminuw (vector bool int, vector unsigned int); 8489vector unsigned int vec_vminuw (vector unsigned int, vector bool int); 8490vector unsigned int vec_vminuw (vector unsigned int, 8491 vector unsigned int); 8492 8493vector signed short vec_vminsh (vector bool short, vector signed short); 8494vector signed short vec_vminsh (vector signed short, vector bool short); 8495vector signed short vec_vminsh (vector signed short, 8496 vector signed short); 8497 8498vector unsigned short vec_vminuh (vector bool short, 8499 vector unsigned short); 8500vector unsigned short vec_vminuh (vector unsigned short, 8501 vector bool short); 8502vector unsigned short vec_vminuh (vector unsigned short, 8503 vector unsigned short); 8504 8505vector signed char vec_vminsb (vector bool char, vector signed char); 8506vector signed char vec_vminsb (vector signed char, vector bool char); 8507vector signed char vec_vminsb (vector signed char, vector signed char); 8508 8509vector unsigned char vec_vminub (vector bool char, 8510 vector unsigned char); 8511vector unsigned char vec_vminub (vector unsigned char, 8512 vector bool char); 8513vector unsigned char vec_vminub (vector unsigned char, 8514 vector unsigned char); 8515 8516vector signed short vec_mladd (vector signed short, 8517 vector signed short, 8518 vector signed short); 8519vector signed short vec_mladd (vector signed short, 8520 vector unsigned short, 8521 vector unsigned short); 8522vector signed short vec_mladd (vector unsigned short, 8523 vector signed short, 8524 vector signed short); 8525vector unsigned short vec_mladd (vector unsigned short, 8526 vector unsigned short, 8527 vector unsigned short); 8528 8529vector signed short vec_mradds (vector signed short, 8530 vector signed short, 8531 vector signed short); 8532 8533vector unsigned int vec_msum (vector unsigned char, 8534 vector unsigned char, 8535 vector unsigned int); 8536vector signed int vec_msum (vector signed char, 8537 vector unsigned char, 8538 vector signed int); 8539vector unsigned int vec_msum (vector unsigned short, 8540 vector unsigned short, 8541 vector unsigned int); 8542vector signed int vec_msum (vector signed short, 8543 vector signed short, 8544 vector signed int); 8545 8546vector signed int vec_vmsumshm (vector signed short, 8547 vector signed short, 8548 vector signed int); 8549 8550vector unsigned int vec_vmsumuhm (vector unsigned short, 8551 vector unsigned short, 8552 vector unsigned int); 8553 8554vector signed int vec_vmsummbm (vector signed char, 8555 vector unsigned char, 8556 vector signed int); 8557 8558vector unsigned int vec_vmsumubm (vector unsigned char, 8559 vector unsigned char, 8560 vector unsigned int); 8561 8562vector unsigned int vec_msums (vector unsigned short, 8563 vector unsigned short, 8564 vector unsigned int); 8565vector signed int vec_msums (vector signed short, 8566 vector signed short, 8567 vector signed int); 8568 8569vector signed int vec_vmsumshs (vector signed short, 8570 vector signed short, 8571 vector signed int); 8572 8573vector unsigned int vec_vmsumuhs (vector unsigned short, 8574 vector unsigned short, 8575 vector unsigned int); 8576 8577void vec_mtvscr (vector signed int); 8578void vec_mtvscr (vector unsigned int); 8579void vec_mtvscr (vector bool int); 8580void vec_mtvscr (vector signed short); 8581void vec_mtvscr (vector unsigned short); 8582void vec_mtvscr (vector bool short); 8583void vec_mtvscr (vector pixel); 8584void vec_mtvscr (vector signed char); 8585void vec_mtvscr (vector unsigned char); 8586void vec_mtvscr (vector bool char); 8587 8588vector unsigned short vec_mule (vector unsigned char, 8589 vector unsigned char); 8590vector signed short vec_mule (vector signed char, 8591 vector signed char); 8592vector unsigned int vec_mule (vector unsigned short, 8593 vector unsigned short); 8594vector signed int vec_mule (vector signed short, vector signed short); 8595 8596vector signed int vec_vmulesh (vector signed short, 8597 vector signed short); 8598 8599vector unsigned int vec_vmuleuh (vector unsigned short, 8600 vector unsigned short); 8601 8602vector signed short vec_vmulesb (vector signed char, 8603 vector signed char); 8604 8605vector unsigned short vec_vmuleub (vector unsigned char, 8606 vector unsigned char); 8607 8608vector unsigned short vec_mulo (vector unsigned char, 8609 vector unsigned char); 8610vector signed short vec_mulo (vector signed char, vector signed char); 8611vector unsigned int vec_mulo (vector unsigned short, 8612 vector unsigned short); 8613vector signed int vec_mulo (vector signed short, vector signed short); 8614 8615vector signed int vec_vmulosh (vector signed short, 8616 vector signed short); 8617 8618vector unsigned int vec_vmulouh (vector unsigned short, 8619 vector unsigned short); 8620 8621vector signed short vec_vmulosb (vector signed char, 8622 vector signed char); 8623 8624vector unsigned short vec_vmuloub (vector unsigned char, 8625 vector unsigned char); 8626 8627vector float vec_nmsub (vector float, vector float, vector float); 8628 8629vector float vec_nor (vector float, vector float); 8630vector signed int vec_nor (vector signed int, vector signed int); 8631vector unsigned int vec_nor (vector unsigned int, vector unsigned int); 8632vector bool int vec_nor (vector bool int, vector bool int); 8633vector signed short vec_nor (vector signed short, vector signed short); 8634vector unsigned short vec_nor (vector unsigned short, 8635 vector unsigned short); 8636vector bool short vec_nor (vector bool short, vector bool short); 8637vector signed char vec_nor (vector signed char, vector signed char); 8638vector unsigned char vec_nor (vector unsigned char, 8639 vector unsigned char); 8640vector bool char vec_nor (vector bool char, vector bool char); 8641 8642vector float vec_or (vector float, vector float); 8643vector float vec_or (vector float, vector bool int); 8644vector float vec_or (vector bool int, vector float); 8645vector bool int vec_or (vector bool int, vector bool int); 8646vector signed int vec_or (vector bool int, vector signed int); 8647vector signed int vec_or (vector signed int, vector bool int); 8648vector signed int vec_or (vector signed int, vector signed int); 8649vector unsigned int vec_or (vector bool int, vector unsigned int); 8650vector unsigned int vec_or (vector unsigned int, vector bool int); 8651vector unsigned int vec_or (vector unsigned int, vector unsigned int); 8652vector bool short vec_or (vector bool short, vector bool short); 8653vector signed short vec_or (vector bool short, vector signed short); 8654vector signed short vec_or (vector signed short, vector bool short); 8655vector signed short vec_or (vector signed short, vector signed short); 8656vector unsigned short vec_or (vector bool short, vector unsigned short); 8657vector unsigned short vec_or (vector unsigned short, vector bool short); 8658vector unsigned short vec_or (vector unsigned short, 8659 vector unsigned short); 8660vector signed char vec_or (vector bool char, vector signed char); 8661vector bool char vec_or (vector bool char, vector bool char); 8662vector signed char vec_or (vector signed char, vector bool char); 8663vector signed char vec_or (vector signed char, vector signed char); 8664vector unsigned char vec_or (vector bool char, vector unsigned char); 8665vector unsigned char vec_or (vector unsigned char, vector bool char); 8666vector unsigned char vec_or (vector unsigned char, 8667 vector unsigned char); 8668 8669vector signed char vec_pack (vector signed short, vector signed short); 8670vector unsigned char vec_pack (vector unsigned short, 8671 vector unsigned short); 8672vector bool char vec_pack (vector bool short, vector bool short); 8673vector signed short vec_pack (vector signed int, vector signed int); 8674vector unsigned short vec_pack (vector unsigned int, 8675 vector unsigned int); 8676vector bool short vec_pack (vector bool int, vector bool int); 8677 8678vector bool short vec_vpkuwum (vector bool int, vector bool int); 8679vector signed short vec_vpkuwum (vector signed int, vector signed int); 8680vector unsigned short vec_vpkuwum (vector unsigned int, 8681 vector unsigned int); 8682 8683vector bool char vec_vpkuhum (vector bool short, vector bool short); 8684vector signed char vec_vpkuhum (vector signed short, 8685 vector signed short); 8686vector unsigned char vec_vpkuhum (vector unsigned short, 8687 vector unsigned short); 8688 8689vector pixel vec_packpx (vector unsigned int, vector unsigned int); 8690 8691vector unsigned char vec_packs (vector unsigned short, 8692 vector unsigned short); 8693vector signed char vec_packs (vector signed short, vector signed short); 8694vector unsigned short vec_packs (vector unsigned int, 8695 vector unsigned int); 8696vector signed short vec_packs (vector signed int, vector signed int); 8697 8698vector signed short vec_vpkswss (vector signed int, vector signed int); 8699 8700vector unsigned short vec_vpkuwus (vector unsigned int, 8701 vector unsigned int); 8702 8703vector signed char vec_vpkshss (vector signed short, 8704 vector signed short); 8705 8706vector unsigned char vec_vpkuhus (vector unsigned short, 8707 vector unsigned short); 8708 8709vector unsigned char vec_packsu (vector unsigned short, 8710 vector unsigned short); 8711vector unsigned char vec_packsu (vector signed short, 8712 vector signed short); 8713vector unsigned short vec_packsu (vector unsigned int, 8714 vector unsigned int); 8715vector unsigned short vec_packsu (vector signed int, vector signed int); 8716 8717vector unsigned short vec_vpkswus (vector signed int, 8718 vector signed int); 8719 8720vector unsigned char vec_vpkshus (vector signed short, 8721 vector signed short); 8722 8723vector float vec_perm (vector float, 8724 vector float, 8725 vector unsigned char); 8726vector signed int vec_perm (vector signed int, 8727 vector signed int, 8728 vector unsigned char); 8729vector unsigned int vec_perm (vector unsigned int, 8730 vector unsigned int, 8731 vector unsigned char); 8732vector bool int vec_perm (vector bool int, 8733 vector bool int, 8734 vector unsigned char); 8735vector signed short vec_perm (vector signed short, 8736 vector signed short, 8737 vector unsigned char); 8738vector unsigned short vec_perm (vector unsigned short, 8739 vector unsigned short, 8740 vector unsigned char); 8741vector bool short vec_perm (vector bool short, 8742 vector bool short, 8743 vector unsigned char); 8744vector pixel vec_perm (vector pixel, 8745 vector pixel, 8746 vector unsigned char); 8747vector signed char vec_perm (vector signed char, 8748 vector signed char, 8749 vector unsigned char); 8750vector unsigned char vec_perm (vector unsigned char, 8751 vector unsigned char, 8752 vector unsigned char); 8753vector bool char vec_perm (vector bool char, 8754 vector bool char, 8755 vector unsigned char); 8756 8757vector float vec_re (vector float); 8758 8759vector signed char vec_rl (vector signed char, 8760 vector unsigned char); 8761vector unsigned char vec_rl (vector unsigned char, 8762 vector unsigned char); 8763vector signed short vec_rl (vector signed short, vector unsigned short); 8764vector unsigned short vec_rl (vector unsigned short, 8765 vector unsigned short); 8766vector signed int vec_rl (vector signed int, vector unsigned int); 8767vector unsigned int vec_rl (vector unsigned int, vector unsigned int); 8768 8769vector signed int vec_vrlw (vector signed int, vector unsigned int); 8770vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int); 8771 8772vector signed short vec_vrlh (vector signed short, 8773 vector unsigned short); 8774vector unsigned short vec_vrlh (vector unsigned short, 8775 vector unsigned short); 8776 8777vector signed char vec_vrlb (vector signed char, vector unsigned char); 8778vector unsigned char vec_vrlb (vector unsigned char, 8779 vector unsigned char); 8780 8781vector float vec_round (vector float); 8782 8783vector float vec_rsqrte (vector float); 8784 8785vector float vec_sel (vector float, vector float, vector bool int); 8786vector float vec_sel (vector float, vector float, vector unsigned int); 8787vector signed int vec_sel (vector signed int, 8788 vector signed int, 8789 vector bool int); 8790vector signed int vec_sel (vector signed int, 8791 vector signed int, 8792 vector unsigned int); 8793vector unsigned int vec_sel (vector unsigned int, 8794 vector unsigned int, 8795 vector bool int); 8796vector unsigned int vec_sel (vector unsigned int, 8797 vector unsigned int, 8798 vector unsigned int); 8799vector bool int vec_sel (vector bool int, 8800 vector bool int, 8801 vector bool int); 8802vector bool int vec_sel (vector bool int, 8803 vector bool int, 8804 vector unsigned int); 8805vector signed short vec_sel (vector signed short, 8806 vector signed short, 8807 vector bool short); 8808vector signed short vec_sel (vector signed short, 8809 vector signed short, 8810 vector unsigned short); 8811vector unsigned short vec_sel (vector unsigned short, 8812 vector unsigned short, 8813 vector bool short); 8814vector unsigned short vec_sel (vector unsigned short, 8815 vector unsigned short, 8816 vector unsigned short); 8817vector bool short vec_sel (vector bool short, 8818 vector bool short, 8819 vector bool short); 8820vector bool short vec_sel (vector bool short, 8821 vector bool short, 8822 vector unsigned short); 8823vector signed char vec_sel (vector signed char, 8824 vector signed char, 8825 vector bool char); 8826vector signed char vec_sel (vector signed char, 8827 vector signed char, 8828 vector unsigned char); 8829vector unsigned char vec_sel (vector unsigned char, 8830 vector unsigned char, 8831 vector bool char); 8832vector unsigned char vec_sel (vector unsigned char, 8833 vector unsigned char, 8834 vector unsigned char); 8835vector bool char vec_sel (vector bool char, 8836 vector bool char, 8837 vector bool char); 8838vector bool char vec_sel (vector bool char, 8839 vector bool char, 8840 vector unsigned char); 8841 8842vector signed char vec_sl (vector signed char, 8843 vector unsigned char); 8844vector unsigned char vec_sl (vector unsigned char, 8845 vector unsigned char); 8846vector signed short vec_sl (vector signed short, vector unsigned short); 8847vector unsigned short vec_sl (vector unsigned short, 8848 vector unsigned short); 8849vector signed int vec_sl (vector signed int, vector unsigned int); 8850vector unsigned int vec_sl (vector unsigned int, vector unsigned int); 8851 8852vector signed int vec_vslw (vector signed int, vector unsigned int); 8853vector unsigned int vec_vslw (vector unsigned int, vector unsigned int); 8854 8855vector signed short vec_vslh (vector signed short, 8856 vector unsigned short); 8857vector unsigned short vec_vslh (vector unsigned short, 8858 vector unsigned short); 8859 8860vector signed char vec_vslb (vector signed char, vector unsigned char); 8861vector unsigned char vec_vslb (vector unsigned char, 8862 vector unsigned char); 8863 8864vector float vec_sld (vector float, vector float, const int); 8865vector signed int vec_sld (vector signed int, 8866 vector signed int, 8867 const int); 8868vector unsigned int vec_sld (vector unsigned int, 8869 vector unsigned int, 8870 const int); 8871vector bool int vec_sld (vector bool int, 8872 vector bool int, 8873 const int); 8874vector signed short vec_sld (vector signed short, 8875 vector signed short, 8876 const int); 8877vector unsigned short vec_sld (vector unsigned short, 8878 vector unsigned short, 8879 const int); 8880vector bool short vec_sld (vector bool short, 8881 vector bool short, 8882 const int); 8883vector pixel vec_sld (vector pixel, 8884 vector pixel, 8885 const int); 8886vector signed char vec_sld (vector signed char, 8887 vector signed char, 8888 const int); 8889vector unsigned char vec_sld (vector unsigned char, 8890 vector unsigned char, 8891 const int); 8892vector bool char vec_sld (vector bool char, 8893 vector bool char, 8894 const int); 8895 8896vector signed int vec_sll (vector signed int, 8897 vector unsigned int); 8898vector signed int vec_sll (vector signed int, 8899 vector unsigned short); 8900vector signed int vec_sll (vector signed int, 8901 vector unsigned char); 8902vector unsigned int vec_sll (vector unsigned int, 8903 vector unsigned int); 8904vector unsigned int vec_sll (vector unsigned int, 8905 vector unsigned short); 8906vector unsigned int vec_sll (vector unsigned int, 8907 vector unsigned char); 8908vector bool int vec_sll (vector bool int, 8909 vector unsigned int); 8910vector bool int vec_sll (vector bool int, 8911 vector unsigned short); 8912vector bool int vec_sll (vector bool int, 8913 vector unsigned char); 8914vector signed short vec_sll (vector signed short, 8915 vector unsigned int); 8916vector signed short vec_sll (vector signed short, 8917 vector unsigned short); 8918vector signed short vec_sll (vector signed short, 8919 vector unsigned char); 8920vector unsigned short vec_sll (vector unsigned short, 8921 vector unsigned int); 8922vector unsigned short vec_sll (vector unsigned short, 8923 vector unsigned short); 8924vector unsigned short vec_sll (vector unsigned short, 8925 vector unsigned char); 8926vector bool short vec_sll (vector bool short, vector unsigned int); 8927vector bool short vec_sll (vector bool short, vector unsigned short); 8928vector bool short vec_sll (vector bool short, vector unsigned char); 8929vector pixel vec_sll (vector pixel, vector unsigned int); 8930vector pixel vec_sll (vector pixel, vector unsigned short); 8931vector pixel vec_sll (vector pixel, vector unsigned char); 8932vector signed char vec_sll (vector signed char, vector unsigned int); 8933vector signed char vec_sll (vector signed char, vector unsigned short); 8934vector signed char vec_sll (vector signed char, vector unsigned char); 8935vector unsigned char vec_sll (vector unsigned char, 8936 vector unsigned int); 8937vector unsigned char vec_sll (vector unsigned char, 8938 vector unsigned short); 8939vector unsigned char vec_sll (vector unsigned char, 8940 vector unsigned char); 8941vector bool char vec_sll (vector bool char, vector unsigned int); 8942vector bool char vec_sll (vector bool char, vector unsigned short); 8943vector bool char vec_sll (vector bool char, vector unsigned char); 8944 8945vector float vec_slo (vector float, vector signed char); 8946vector float vec_slo (vector float, vector unsigned char); 8947vector signed int vec_slo (vector signed int, vector signed char); 8948vector signed int vec_slo (vector signed int, vector unsigned char); 8949vector unsigned int vec_slo (vector unsigned int, vector signed char); 8950vector unsigned int vec_slo (vector unsigned int, vector unsigned char); 8951vector signed short vec_slo (vector signed short, vector signed char); 8952vector signed short vec_slo (vector signed short, vector unsigned char); 8953vector unsigned short vec_slo (vector unsigned short, 8954 vector signed char); 8955vector unsigned short vec_slo (vector unsigned short, 8956 vector unsigned char); 8957vector pixel vec_slo (vector pixel, vector signed char); 8958vector pixel vec_slo (vector pixel, vector unsigned char); 8959vector signed char vec_slo (vector signed char, vector signed char); 8960vector signed char vec_slo (vector signed char, vector unsigned char); 8961vector unsigned char vec_slo (vector unsigned char, vector signed char); 8962vector unsigned char vec_slo (vector unsigned char, 8963 vector unsigned char); 8964 8965vector signed char vec_splat (vector signed char, const int); 8966vector unsigned char vec_splat (vector unsigned char, const int); 8967vector bool char vec_splat (vector bool char, const int); 8968vector signed short vec_splat (vector signed short, const int); 8969vector unsigned short vec_splat (vector unsigned short, const int); 8970vector bool short vec_splat (vector bool short, const int); 8971vector pixel vec_splat (vector pixel, const int); 8972vector float vec_splat (vector float, const int); 8973vector signed int vec_splat (vector signed int, const int); 8974vector unsigned int vec_splat (vector unsigned int, const int); 8975vector bool int vec_splat (vector bool int, const int); 8976 8977vector float vec_vspltw (vector float, const int); 8978vector signed int vec_vspltw (vector signed int, const int); 8979vector unsigned int vec_vspltw (vector unsigned int, const int); 8980vector bool int vec_vspltw (vector bool int, const int); 8981 8982vector bool short vec_vsplth (vector bool short, const int); 8983vector signed short vec_vsplth (vector signed short, const int); 8984vector unsigned short vec_vsplth (vector unsigned short, const int); 8985vector pixel vec_vsplth (vector pixel, const int); 8986 8987vector signed char vec_vspltb (vector signed char, const int); 8988vector unsigned char vec_vspltb (vector unsigned char, const int); 8989vector bool char vec_vspltb (vector bool char, const int); 8990 8991vector signed char vec_splat_s8 (const int); 8992 8993vector signed short vec_splat_s16 (const int); 8994 8995vector signed int vec_splat_s32 (const int); 8996 8997vector unsigned char vec_splat_u8 (const int); 8998 8999vector unsigned short vec_splat_u16 (const int); 9000 9001vector unsigned int vec_splat_u32 (const int); 9002 9003vector signed char vec_sr (vector signed char, vector unsigned char); 9004vector unsigned char vec_sr (vector unsigned char, 9005 vector unsigned char); 9006vector signed short vec_sr (vector signed short, 9007 vector unsigned short); 9008vector unsigned short vec_sr (vector unsigned short, 9009 vector unsigned short); 9010vector signed int vec_sr (vector signed int, vector unsigned int); 9011vector unsigned int vec_sr (vector unsigned int, vector unsigned int); 9012 9013vector signed int vec_vsrw (vector signed int, vector unsigned int); 9014vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int); 9015 9016vector signed short vec_vsrh (vector signed short, 9017 vector unsigned short); 9018vector unsigned short vec_vsrh (vector unsigned short, 9019 vector unsigned short); 9020 9021vector signed char vec_vsrb (vector signed char, vector unsigned char); 9022vector unsigned char vec_vsrb (vector unsigned char, 9023 vector unsigned char); 9024 9025vector signed char vec_sra (vector signed char, vector unsigned char); 9026vector unsigned char vec_sra (vector unsigned char, 9027 vector unsigned char); 9028vector signed short vec_sra (vector signed short, 9029 vector unsigned short); 9030vector unsigned short vec_sra (vector unsigned short, 9031 vector unsigned short); 9032vector signed int vec_sra (vector signed int, vector unsigned int); 9033vector unsigned int vec_sra (vector unsigned int, vector unsigned int); 9034 9035vector signed int vec_vsraw (vector signed int, vector unsigned int); 9036vector unsigned int vec_vsraw (vector unsigned int, 9037 vector unsigned int); 9038 9039vector signed short vec_vsrah (vector signed short, 9040 vector unsigned short); 9041vector unsigned short vec_vsrah (vector unsigned short, 9042 vector unsigned short); 9043 9044vector signed char vec_vsrab (vector signed char, vector unsigned char); 9045vector unsigned char vec_vsrab (vector unsigned char, 9046 vector unsigned char); 9047 9048vector signed int vec_srl (vector signed int, vector unsigned int); 9049vector signed int vec_srl (vector signed int, vector unsigned short); 9050vector signed int vec_srl (vector signed int, vector unsigned char); 9051vector unsigned int vec_srl (vector unsigned int, vector unsigned int); 9052vector unsigned int vec_srl (vector unsigned int, 9053 vector unsigned short); 9054vector unsigned int vec_srl (vector unsigned int, vector unsigned char); 9055vector bool int vec_srl (vector bool int, vector unsigned int); 9056vector bool int vec_srl (vector bool int, vector unsigned short); 9057vector bool int vec_srl (vector bool int, vector unsigned char); 9058vector signed short vec_srl (vector signed short, vector unsigned int); 9059vector signed short vec_srl (vector signed short, 9060 vector unsigned short); 9061vector signed short vec_srl (vector signed short, vector unsigned char); 9062vector unsigned short vec_srl (vector unsigned short, 9063 vector unsigned int); 9064vector unsigned short vec_srl (vector unsigned short, 9065 vector unsigned short); 9066vector unsigned short vec_srl (vector unsigned short, 9067 vector unsigned char); 9068vector bool short vec_srl (vector bool short, vector unsigned int); 9069vector bool short vec_srl (vector bool short, vector unsigned short); 9070vector bool short vec_srl (vector bool short, vector unsigned char); 9071vector pixel vec_srl (vector pixel, vector unsigned int); 9072vector pixel vec_srl (vector pixel, vector unsigned short); 9073vector pixel vec_srl (vector pixel, vector unsigned char); 9074vector signed char vec_srl (vector signed char, vector unsigned int); 9075vector signed char vec_srl (vector signed char, vector unsigned short); 9076vector signed char vec_srl (vector signed char, vector unsigned char); 9077vector unsigned char vec_srl (vector unsigned char, 9078 vector unsigned int); 9079vector unsigned char vec_srl (vector unsigned char, 9080 vector unsigned short); 9081vector unsigned char vec_srl (vector unsigned char, 9082 vector unsigned char); 9083vector bool char vec_srl (vector bool char, vector unsigned int); 9084vector bool char vec_srl (vector bool char, vector unsigned short); 9085vector bool char vec_srl (vector bool char, vector unsigned char); 9086 9087vector float vec_sro (vector float, vector signed char); 9088vector float vec_sro (vector float, vector unsigned char); 9089vector signed int vec_sro (vector signed int, vector signed char); 9090vector signed int vec_sro (vector signed int, vector unsigned char); 9091vector unsigned int vec_sro (vector unsigned int, vector signed char); 9092vector unsigned int vec_sro (vector unsigned int, vector unsigned char); 9093vector signed short vec_sro (vector signed short, vector signed char); 9094vector signed short vec_sro (vector signed short, vector unsigned char); 9095vector unsigned short vec_sro (vector unsigned short, 9096 vector signed char); 9097vector unsigned short vec_sro (vector unsigned short, 9098 vector unsigned char); 9099vector pixel vec_sro (vector pixel, vector signed char); 9100vector pixel vec_sro (vector pixel, vector unsigned char); 9101vector signed char vec_sro (vector signed char, vector signed char); 9102vector signed char vec_sro (vector signed char, vector unsigned char); 9103vector unsigned char vec_sro (vector unsigned char, vector signed char); 9104vector unsigned char vec_sro (vector unsigned char, 9105 vector unsigned char); 9106 9107void vec_st (vector float, int, vector float *); 9108void vec_st (vector float, int, float *); 9109void vec_st (vector signed int, int, vector signed int *); 9110void vec_st (vector signed int, int, int *); 9111void vec_st (vector unsigned int, int, vector unsigned int *); 9112void vec_st (vector unsigned int, int, unsigned int *); 9113void vec_st (vector bool int, int, vector bool int *); 9114void vec_st (vector bool int, int, unsigned int *); 9115void vec_st (vector bool int, int, int *); 9116void vec_st (vector signed short, int, vector signed short *); 9117void vec_st (vector signed short, int, short *); 9118void vec_st (vector unsigned short, int, vector unsigned short *); 9119void vec_st (vector unsigned short, int, unsigned short *); 9120void vec_st (vector bool short, int, vector bool short *); 9121void vec_st (vector bool short, int, unsigned short *); 9122void vec_st (vector pixel, int, vector pixel *); 9123void vec_st (vector pixel, int, unsigned short *); 9124void vec_st (vector pixel, int, short *); 9125void vec_st (vector bool short, int, short *); 9126void vec_st (vector signed char, int, vector signed char *); 9127void vec_st (vector signed char, int, signed char *); 9128void vec_st (vector unsigned char, int, vector unsigned char *); 9129void vec_st (vector unsigned char, int, unsigned char *); 9130void vec_st (vector bool char, int, vector bool char *); 9131void vec_st (vector bool char, int, unsigned char *); 9132void vec_st (vector bool char, int, signed char *); 9133 9134void vec_ste (vector signed char, int, signed char *); 9135void vec_ste (vector unsigned char, int, unsigned char *); 9136void vec_ste (vector bool char, int, signed char *); 9137void vec_ste (vector bool char, int, unsigned char *); 9138void vec_ste (vector signed short, int, short *); 9139void vec_ste (vector unsigned short, int, unsigned short *); 9140void vec_ste (vector bool short, int, short *); 9141void vec_ste (vector bool short, int, unsigned short *); 9142void vec_ste (vector pixel, int, short *); 9143void vec_ste (vector pixel, int, unsigned short *); 9144void vec_ste (vector float, int, float *); 9145void vec_ste (vector signed int, int, int *); 9146void vec_ste (vector unsigned int, int, unsigned int *); 9147void vec_ste (vector bool int, int, int *); 9148void vec_ste (vector bool int, int, unsigned int *); 9149 9150void vec_stvewx (vector float, int, float *); 9151void vec_stvewx (vector signed int, int, int *); 9152void vec_stvewx (vector unsigned int, int, unsigned int *); 9153void vec_stvewx (vector bool int, int, int *); 9154void vec_stvewx (vector bool int, int, unsigned int *); 9155 9156void vec_stvehx (vector signed short, int, short *); 9157void vec_stvehx (vector unsigned short, int, unsigned short *); 9158void vec_stvehx (vector bool short, int, short *); 9159void vec_stvehx (vector bool short, int, unsigned short *); 9160void vec_stvehx (vector pixel, int, short *); 9161void vec_stvehx (vector pixel, int, unsigned short *); 9162 9163void vec_stvebx (vector signed char, int, signed char *); 9164void vec_stvebx (vector unsigned char, int, unsigned char *); 9165void vec_stvebx (vector bool char, int, signed char *); 9166void vec_stvebx (vector bool char, int, unsigned char *); 9167 9168void vec_stl (vector float, int, vector float *); 9169void vec_stl (vector float, int, float *); 9170void vec_stl (vector signed int, int, vector signed int *); 9171void vec_stl (vector signed int, int, int *); 9172void vec_stl (vector unsigned int, int, vector unsigned int *); 9173void vec_stl (vector unsigned int, int, unsigned int *); 9174void vec_stl (vector bool int, int, vector bool int *); 9175void vec_stl (vector bool int, int, unsigned int *); 9176void vec_stl (vector bool int, int, int *); 9177void vec_stl (vector signed short, int, vector signed short *); 9178void vec_stl (vector signed short, int, short *); 9179void vec_stl (vector unsigned short, int, vector unsigned short *); 9180void vec_stl (vector unsigned short, int, unsigned short *); 9181void vec_stl (vector bool short, int, vector bool short *); 9182void vec_stl (vector bool short, int, unsigned short *); 9183void vec_stl (vector bool short, int, short *); 9184void vec_stl (vector pixel, int, vector pixel *); 9185void vec_stl (vector pixel, int, unsigned short *); 9186void vec_stl (vector pixel, int, short *); 9187void vec_stl (vector signed char, int, vector signed char *); 9188void vec_stl (vector signed char, int, signed char *); 9189void vec_stl (vector unsigned char, int, vector unsigned char *); 9190void vec_stl (vector unsigned char, int, unsigned char *); 9191void vec_stl (vector bool char, int, vector bool char *); 9192void vec_stl (vector bool char, int, unsigned char *); 9193void vec_stl (vector bool char, int, signed char *); 9194 9195vector signed char vec_sub (vector bool char, vector signed char); 9196vector signed char vec_sub (vector signed char, vector bool char); 9197vector signed char vec_sub (vector signed char, vector signed char); 9198vector unsigned char vec_sub (vector bool char, vector unsigned char); 9199vector unsigned char vec_sub (vector unsigned char, vector bool char); 9200vector unsigned char vec_sub (vector unsigned char, 9201 vector unsigned char); 9202vector signed short vec_sub (vector bool short, vector signed short); 9203vector signed short vec_sub (vector signed short, vector bool short); 9204vector signed short vec_sub (vector signed short, vector signed short); 9205vector unsigned short vec_sub (vector bool short, 9206 vector unsigned short); 9207vector unsigned short vec_sub (vector unsigned short, 9208 vector bool short); 9209vector unsigned short vec_sub (vector unsigned short, 9210 vector unsigned short); 9211vector signed int vec_sub (vector bool int, vector signed int); 9212vector signed int vec_sub (vector signed int, vector bool int); 9213vector signed int vec_sub (vector signed int, vector signed int); 9214vector unsigned int vec_sub (vector bool int, vector unsigned int); 9215vector unsigned int vec_sub (vector unsigned int, vector bool int); 9216vector unsigned int vec_sub (vector unsigned int, vector unsigned int); 9217vector float vec_sub (vector float, vector float); 9218 9219vector float vec_vsubfp (vector float, vector float); 9220 9221vector signed int vec_vsubuwm (vector bool int, vector signed int); 9222vector signed int vec_vsubuwm (vector signed int, vector bool int); 9223vector signed int vec_vsubuwm (vector signed int, vector signed int); 9224vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int); 9225vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int); 9226vector unsigned int vec_vsubuwm (vector unsigned int, 9227 vector unsigned int); 9228 9229vector signed short vec_vsubuhm (vector bool short, 9230 vector signed short); 9231vector signed short vec_vsubuhm (vector signed short, 9232 vector bool short); 9233vector signed short vec_vsubuhm (vector signed short, 9234 vector signed short); 9235vector unsigned short vec_vsubuhm (vector bool short, 9236 vector unsigned short); 9237vector unsigned short vec_vsubuhm (vector unsigned short, 9238 vector bool short); 9239vector unsigned short vec_vsubuhm (vector unsigned short, 9240 vector unsigned short); 9241 9242vector signed char vec_vsububm (vector bool char, vector signed char); 9243vector signed char vec_vsububm (vector signed char, vector bool char); 9244vector signed char vec_vsububm (vector signed char, vector signed char); 9245vector unsigned char vec_vsububm (vector bool char, 9246 vector unsigned char); 9247vector unsigned char vec_vsububm (vector unsigned char, 9248 vector bool char); 9249vector unsigned char vec_vsububm (vector unsigned char, 9250 vector unsigned char); 9251 9252vector unsigned int vec_subc (vector unsigned int, vector unsigned int); 9253 9254vector unsigned char vec_subs (vector bool char, vector unsigned char); 9255vector unsigned char vec_subs (vector unsigned char, vector bool char); 9256vector unsigned char vec_subs (vector unsigned char, 9257 vector unsigned char); 9258vector signed char vec_subs (vector bool char, vector signed char); 9259vector signed char vec_subs (vector signed char, vector bool char); 9260vector signed char vec_subs (vector signed char, vector signed char); 9261vector unsigned short vec_subs (vector bool short, 9262 vector unsigned short); 9263vector unsigned short vec_subs (vector unsigned short, 9264 vector bool short); 9265vector unsigned short vec_subs (vector unsigned short, 9266 vector unsigned short); 9267vector signed short vec_subs (vector bool short, vector signed short); 9268vector signed short vec_subs (vector signed short, vector bool short); 9269vector signed short vec_subs (vector signed short, vector signed short); 9270vector unsigned int vec_subs (vector bool int, vector unsigned int); 9271vector unsigned int vec_subs (vector unsigned int, vector bool int); 9272vector unsigned int vec_subs (vector unsigned int, vector unsigned int); 9273vector signed int vec_subs (vector bool int, vector signed int); 9274vector signed int vec_subs (vector signed int, vector bool int); 9275vector signed int vec_subs (vector signed int, vector signed int); 9276 9277vector signed int vec_vsubsws (vector bool int, vector signed int); 9278vector signed int vec_vsubsws (vector signed int, vector bool int); 9279vector signed int vec_vsubsws (vector signed int, vector signed int); 9280 9281vector unsigned int vec_vsubuws (vector bool int, vector unsigned int); 9282vector unsigned int vec_vsubuws (vector unsigned int, vector bool int); 9283vector unsigned int vec_vsubuws (vector unsigned int, 9284 vector unsigned int); 9285 9286vector signed short vec_vsubshs (vector bool short, 9287 vector signed short); 9288vector signed short vec_vsubshs (vector signed short, 9289 vector bool short); 9290vector signed short vec_vsubshs (vector signed short, 9291 vector signed short); 9292 9293vector unsigned short vec_vsubuhs (vector bool short, 9294 vector unsigned short); 9295vector unsigned short vec_vsubuhs (vector unsigned short, 9296 vector bool short); 9297vector unsigned short vec_vsubuhs (vector unsigned short, 9298 vector unsigned short); 9299 9300vector signed char vec_vsubsbs (vector bool char, vector signed char); 9301vector signed char vec_vsubsbs (vector signed char, vector bool char); 9302vector signed char vec_vsubsbs (vector signed char, vector signed char); 9303 9304vector unsigned char vec_vsububs (vector bool char, 9305 vector unsigned char); 9306vector unsigned char vec_vsububs (vector unsigned char, 9307 vector bool char); 9308vector unsigned char vec_vsububs (vector unsigned char, 9309 vector unsigned char); 9310 9311vector unsigned int vec_sum4s (vector unsigned char, 9312 vector unsigned int); 9313vector signed int vec_sum4s (vector signed char, vector signed int); 9314vector signed int vec_sum4s (vector signed short, vector signed int); 9315 9316vector signed int vec_vsum4shs (vector signed short, vector signed int); 9317 9318vector signed int vec_vsum4sbs (vector signed char, vector signed int); 9319 9320vector unsigned int vec_vsum4ubs (vector unsigned char, 9321 vector unsigned int); 9322 9323vector signed int vec_sum2s (vector signed int, vector signed int); 9324 9325vector signed int vec_sums (vector signed int, vector signed int); 9326 9327vector float vec_trunc (vector float); 9328 9329vector signed short vec_unpackh (vector signed char); 9330vector bool short vec_unpackh (vector bool char); 9331vector signed int vec_unpackh (vector signed short); 9332vector bool int vec_unpackh (vector bool short); 9333vector unsigned int vec_unpackh (vector pixel); 9334 9335vector bool int vec_vupkhsh (vector bool short); 9336vector signed int vec_vupkhsh (vector signed short); 9337 9338vector unsigned int vec_vupkhpx (vector pixel); 9339 9340vector bool short vec_vupkhsb (vector bool char); 9341vector signed short vec_vupkhsb (vector signed char); 9342 9343vector signed short vec_unpackl (vector signed char); 9344vector bool short vec_unpackl (vector bool char); 9345vector unsigned int vec_unpackl (vector pixel); 9346vector signed int vec_unpackl (vector signed short); 9347vector bool int vec_unpackl (vector bool short); 9348 9349vector unsigned int vec_vupklpx (vector pixel); 9350 9351vector bool int vec_vupklsh (vector bool short); 9352vector signed int vec_vupklsh (vector signed short); 9353 9354vector bool short vec_vupklsb (vector bool char); 9355vector signed short vec_vupklsb (vector signed char); 9356 9357vector float vec_xor (vector float, vector float); 9358vector float vec_xor (vector float, vector bool int); 9359vector float vec_xor (vector bool int, vector float); 9360vector bool int vec_xor (vector bool int, vector bool int); 9361vector signed int vec_xor (vector bool int, vector signed int); 9362vector signed int vec_xor (vector signed int, vector bool int); 9363vector signed int vec_xor (vector signed int, vector signed int); 9364vector unsigned int vec_xor (vector bool int, vector unsigned int); 9365vector unsigned int vec_xor (vector unsigned int, vector bool int); 9366vector unsigned int vec_xor (vector unsigned int, vector unsigned int); 9367vector bool short vec_xor (vector bool short, vector bool short); 9368vector signed short vec_xor (vector bool short, vector signed short); 9369vector signed short vec_xor (vector signed short, vector bool short); 9370vector signed short vec_xor (vector signed short, vector signed short); 9371vector unsigned short vec_xor (vector bool short, 9372 vector unsigned short); 9373vector unsigned short vec_xor (vector unsigned short, 9374 vector bool short); 9375vector unsigned short vec_xor (vector unsigned short, 9376 vector unsigned short); 9377vector signed char vec_xor (vector bool char, vector signed char); 9378vector bool char vec_xor (vector bool char, vector bool char); 9379vector signed char vec_xor (vector signed char, vector bool char); 9380vector signed char vec_xor (vector signed char, vector signed char); 9381vector unsigned char vec_xor (vector bool char, vector unsigned char); 9382vector unsigned char vec_xor (vector unsigned char, vector bool char); 9383vector unsigned char vec_xor (vector unsigned char, 9384 vector unsigned char); 9385 9386int vec_all_eq (vector signed char, vector bool char); 9387int vec_all_eq (vector signed char, vector signed char); 9388int vec_all_eq (vector unsigned char, vector bool char); 9389int vec_all_eq (vector unsigned char, vector unsigned char); 9390int vec_all_eq (vector bool char, vector bool char); 9391int vec_all_eq (vector bool char, vector unsigned char); 9392int vec_all_eq (vector bool char, vector signed char); 9393int vec_all_eq (vector signed short, vector bool short); 9394int vec_all_eq (vector signed short, vector signed short); 9395int vec_all_eq (vector unsigned short, vector bool short); 9396int vec_all_eq (vector unsigned short, vector unsigned short); 9397int vec_all_eq (vector bool short, vector bool short); 9398int vec_all_eq (vector bool short, vector unsigned short); 9399int vec_all_eq (vector bool short, vector signed short); 9400int vec_all_eq (vector pixel, vector pixel); 9401int vec_all_eq (vector signed int, vector bool int); 9402int vec_all_eq (vector signed int, vector signed int); 9403int vec_all_eq (vector unsigned int, vector bool int); 9404int vec_all_eq (vector unsigned int, vector unsigned int); 9405int vec_all_eq (vector bool int, vector bool int); 9406int vec_all_eq (vector bool int, vector unsigned int); 9407int vec_all_eq (vector bool int, vector signed int); 9408int vec_all_eq (vector float, vector float); 9409 9410int vec_all_ge (vector bool char, vector unsigned char); 9411int vec_all_ge (vector unsigned char, vector bool char); 9412int vec_all_ge (vector unsigned char, vector unsigned char); 9413int vec_all_ge (vector bool char, vector signed char); 9414int vec_all_ge (vector signed char, vector bool char); 9415int vec_all_ge (vector signed char, vector signed char); 9416int vec_all_ge (vector bool short, vector unsigned short); 9417int vec_all_ge (vector unsigned short, vector bool short); 9418int vec_all_ge (vector unsigned short, vector unsigned short); 9419int vec_all_ge (vector signed short, vector signed short); 9420int vec_all_ge (vector bool short, vector signed short); 9421int vec_all_ge (vector signed short, vector bool short); 9422int vec_all_ge (vector bool int, vector unsigned int); 9423int vec_all_ge (vector unsigned int, vector bool int); 9424int vec_all_ge (vector unsigned int, vector unsigned int); 9425int vec_all_ge (vector bool int, vector signed int); 9426int vec_all_ge (vector signed int, vector bool int); 9427int vec_all_ge (vector signed int, vector signed int); 9428int vec_all_ge (vector float, vector float); 9429 9430int vec_all_gt (vector bool char, vector unsigned char); 9431int vec_all_gt (vector unsigned char, vector bool char); 9432int vec_all_gt (vector unsigned char, vector unsigned char); 9433int vec_all_gt (vector bool char, vector signed char); 9434int vec_all_gt (vector signed char, vector bool char); 9435int vec_all_gt (vector signed char, vector signed char); 9436int vec_all_gt (vector bool short, vector unsigned short); 9437int vec_all_gt (vector unsigned short, vector bool short); 9438int vec_all_gt (vector unsigned short, vector unsigned short); 9439int vec_all_gt (vector bool short, vector signed short); 9440int vec_all_gt (vector signed short, vector bool short); 9441int vec_all_gt (vector signed short, vector signed short); 9442int vec_all_gt (vector bool int, vector unsigned int); 9443int vec_all_gt (vector unsigned int, vector bool int); 9444int vec_all_gt (vector unsigned int, vector unsigned int); 9445int vec_all_gt (vector bool int, vector signed int); 9446int vec_all_gt (vector signed int, vector bool int); 9447int vec_all_gt (vector signed int, vector signed int); 9448int vec_all_gt (vector float, vector float); 9449 9450int vec_all_in (vector float, vector float); 9451 9452int vec_all_le (vector bool char, vector unsigned char); 9453int vec_all_le (vector unsigned char, vector bool char); 9454int vec_all_le (vector unsigned char, vector unsigned char); 9455int vec_all_le (vector bool char, vector signed char); 9456int vec_all_le (vector signed char, vector bool char); 9457int vec_all_le (vector signed char, vector signed char); 9458int vec_all_le (vector bool short, vector unsigned short); 9459int vec_all_le (vector unsigned short, vector bool short); 9460int vec_all_le (vector unsigned short, vector unsigned short); 9461int vec_all_le (vector bool short, vector signed short); 9462int vec_all_le (vector signed short, vector bool short); 9463int vec_all_le (vector signed short, vector signed short); 9464int vec_all_le (vector bool int, vector unsigned int); 9465int vec_all_le (vector unsigned int, vector bool int); 9466int vec_all_le (vector unsigned int, vector unsigned int); 9467int vec_all_le (vector bool int, vector signed int); 9468int vec_all_le (vector signed int, vector bool int); 9469int vec_all_le (vector signed int, vector signed int); 9470int vec_all_le (vector float, vector float); 9471 9472int vec_all_lt (vector bool char, vector unsigned char); 9473int vec_all_lt (vector unsigned char, vector bool char); 9474int vec_all_lt (vector unsigned char, vector unsigned char); 9475int vec_all_lt (vector bool char, vector signed char); 9476int vec_all_lt (vector signed char, vector bool char); 9477int vec_all_lt (vector signed char, vector signed char); 9478int vec_all_lt (vector bool short, vector unsigned short); 9479int vec_all_lt (vector unsigned short, vector bool short); 9480int vec_all_lt (vector unsigned short, vector unsigned short); 9481int vec_all_lt (vector bool short, vector signed short); 9482int vec_all_lt (vector signed short, vector bool short); 9483int vec_all_lt (vector signed short, vector signed short); 9484int vec_all_lt (vector bool int, vector unsigned int); 9485int vec_all_lt (vector unsigned int, vector bool int); 9486int vec_all_lt (vector unsigned int, vector unsigned int); 9487int vec_all_lt (vector bool int, vector signed int); 9488int vec_all_lt (vector signed int, vector bool int); 9489int vec_all_lt (vector signed int, vector signed int); 9490int vec_all_lt (vector float, vector float); 9491 9492int vec_all_nan (vector float); 9493 9494int vec_all_ne (vector signed char, vector bool char); 9495int vec_all_ne (vector signed char, vector signed char); 9496int vec_all_ne (vector unsigned char, vector bool char); 9497int vec_all_ne (vector unsigned char, vector unsigned char); 9498int vec_all_ne (vector bool char, vector bool char); 9499int vec_all_ne (vector bool char, vector unsigned char); 9500int vec_all_ne (vector bool char, vector signed char); 9501int vec_all_ne (vector signed short, vector bool short); 9502int vec_all_ne (vector signed short, vector signed short); 9503int vec_all_ne (vector unsigned short, vector bool short); 9504int vec_all_ne (vector unsigned short, vector unsigned short); 9505int vec_all_ne (vector bool short, vector bool short); 9506int vec_all_ne (vector bool short, vector unsigned short); 9507int vec_all_ne (vector bool short, vector signed short); 9508int vec_all_ne (vector pixel, vector pixel); 9509int vec_all_ne (vector signed int, vector bool int); 9510int vec_all_ne (vector signed int, vector signed int); 9511int vec_all_ne (vector unsigned int, vector bool int); 9512int vec_all_ne (vector unsigned int, vector unsigned int); 9513int vec_all_ne (vector bool int, vector bool int); 9514int vec_all_ne (vector bool int, vector unsigned int); 9515int vec_all_ne (vector bool int, vector signed int); 9516int vec_all_ne (vector float, vector float); 9517 9518int vec_all_nge (vector float, vector float); 9519 9520int vec_all_ngt (vector float, vector float); 9521 9522int vec_all_nle (vector float, vector float); 9523 9524int vec_all_nlt (vector float, vector float); 9525 9526int vec_all_numeric (vector float); 9527 9528int vec_any_eq (vector signed char, vector bool char); 9529int vec_any_eq (vector signed char, vector signed char); 9530int vec_any_eq (vector unsigned char, vector bool char); 9531int vec_any_eq (vector unsigned char, vector unsigned char); 9532int vec_any_eq (vector bool char, vector bool char); 9533int vec_any_eq (vector bool char, vector unsigned char); 9534int vec_any_eq (vector bool char, vector signed char); 9535int vec_any_eq (vector signed short, vector bool short); 9536int vec_any_eq (vector signed short, vector signed short); 9537int vec_any_eq (vector unsigned short, vector bool short); 9538int vec_any_eq (vector unsigned short, vector unsigned short); 9539int vec_any_eq (vector bool short, vector bool short); 9540int vec_any_eq (vector bool short, vector unsigned short); 9541int vec_any_eq (vector bool short, vector signed short); 9542int vec_any_eq (vector pixel, vector pixel); 9543int vec_any_eq (vector signed int, vector bool int); 9544int vec_any_eq (vector signed int, vector signed int); 9545int vec_any_eq (vector unsigned int, vector bool int); 9546int vec_any_eq (vector unsigned int, vector unsigned int); 9547int vec_any_eq (vector bool int, vector bool int); 9548int vec_any_eq (vector bool int, vector unsigned int); 9549int vec_any_eq (vector bool int, vector signed int); 9550int vec_any_eq (vector float, vector float); 9551 9552int vec_any_ge (vector signed char, vector bool char); 9553int vec_any_ge (vector unsigned char, vector bool char); 9554int vec_any_ge (vector unsigned char, vector unsigned char); 9555int vec_any_ge (vector signed char, vector signed char); 9556int vec_any_ge (vector bool char, vector unsigned char); 9557int vec_any_ge (vector bool char, vector signed char); 9558int vec_any_ge (vector unsigned short, vector bool short); 9559int vec_any_ge (vector unsigned short, vector unsigned short); 9560int vec_any_ge (vector signed short, vector signed short); 9561int vec_any_ge (vector signed short, vector bool short); 9562int vec_any_ge (vector bool short, vector unsigned short); 9563int vec_any_ge (vector bool short, vector signed short); 9564int vec_any_ge (vector signed int, vector bool int); 9565int vec_any_ge (vector unsigned int, vector bool int); 9566int vec_any_ge (vector unsigned int, vector unsigned int); 9567int vec_any_ge (vector signed int, vector signed int); 9568int vec_any_ge (vector bool int, vector unsigned int); 9569int vec_any_ge (vector bool int, vector signed int); 9570int vec_any_ge (vector float, vector float); 9571 9572int vec_any_gt (vector bool char, vector unsigned char); 9573int vec_any_gt (vector unsigned char, vector bool char); 9574int vec_any_gt (vector unsigned char, vector unsigned char); 9575int vec_any_gt (vector bool char, vector signed char); 9576int vec_any_gt (vector signed char, vector bool char); 9577int vec_any_gt (vector signed char, vector signed char); 9578int vec_any_gt (vector bool short, vector unsigned short); 9579int vec_any_gt (vector unsigned short, vector bool short); 9580int vec_any_gt (vector unsigned short, vector unsigned short); 9581int vec_any_gt (vector bool short, vector signed short); 9582int vec_any_gt (vector signed short, vector bool short); 9583int vec_any_gt (vector signed short, vector signed short); 9584int vec_any_gt (vector bool int, vector unsigned int); 9585int vec_any_gt (vector unsigned int, vector bool int); 9586int vec_any_gt (vector unsigned int, vector unsigned int); 9587int vec_any_gt (vector bool int, vector signed int); 9588int vec_any_gt (vector signed int, vector bool int); 9589int vec_any_gt (vector signed int, vector signed int); 9590int vec_any_gt (vector float, vector float); 9591 9592int vec_any_le (vector bool char, vector unsigned char); 9593int vec_any_le (vector unsigned char, vector bool char); 9594int vec_any_le (vector unsigned char, vector unsigned char); 9595int vec_any_le (vector bool char, vector signed char); 9596int vec_any_le (vector signed char, vector bool char); 9597int vec_any_le (vector signed char, vector signed char); 9598int vec_any_le (vector bool short, vector unsigned short); 9599int vec_any_le (vector unsigned short, vector bool short); 9600int vec_any_le (vector unsigned short, vector unsigned short); 9601int vec_any_le (vector bool short, vector signed short); 9602int vec_any_le (vector signed short, vector bool short); 9603int vec_any_le (vector signed short, vector signed short); 9604int vec_any_le (vector bool int, vector unsigned int); 9605int vec_any_le (vector unsigned int, vector bool int); 9606int vec_any_le (vector unsigned int, vector unsigned int); 9607int vec_any_le (vector bool int, vector signed int); 9608int vec_any_le (vector signed int, vector bool int); 9609int vec_any_le (vector signed int, vector signed int); 9610int vec_any_le (vector float, vector float); 9611 9612int vec_any_lt (vector bool char, vector unsigned char); 9613int vec_any_lt (vector unsigned char, vector bool char); 9614int vec_any_lt (vector unsigned char, vector unsigned char); 9615int vec_any_lt (vector bool char, vector signed char); 9616int vec_any_lt (vector signed char, vector bool char); 9617int vec_any_lt (vector signed char, vector signed char); 9618int vec_any_lt (vector bool short, vector unsigned short); 9619int vec_any_lt (vector unsigned short, vector bool short); 9620int vec_any_lt (vector unsigned short, vector unsigned short); 9621int vec_any_lt (vector bool short, vector signed short); 9622int vec_any_lt (vector signed short, vector bool short); 9623int vec_any_lt (vector signed short, vector signed short); 9624int vec_any_lt (vector bool int, vector unsigned int); 9625int vec_any_lt (vector unsigned int, vector bool int); 9626int vec_any_lt (vector unsigned int, vector unsigned int); 9627int vec_any_lt (vector bool int, vector signed int); 9628int vec_any_lt (vector signed int, vector bool int); 9629int vec_any_lt (vector signed int, vector signed int); 9630int vec_any_lt (vector float, vector float); 9631 9632int vec_any_nan (vector float); 9633 9634int vec_any_ne (vector signed char, vector bool char); 9635int vec_any_ne (vector signed char, vector signed char); 9636int vec_any_ne (vector unsigned char, vector bool char); 9637int vec_any_ne (vector unsigned char, vector unsigned char); 9638int vec_any_ne (vector bool char, vector bool char); 9639int vec_any_ne (vector bool char, vector unsigned char); 9640int vec_any_ne (vector bool char, vector signed char); 9641int vec_any_ne (vector signed short, vector bool short); 9642int vec_any_ne (vector signed short, vector signed short); 9643int vec_any_ne (vector unsigned short, vector bool short); 9644int vec_any_ne (vector unsigned short, vector unsigned short); 9645int vec_any_ne (vector bool short, vector bool short); 9646int vec_any_ne (vector bool short, vector unsigned short); 9647int vec_any_ne (vector bool short, vector signed short); 9648int vec_any_ne (vector pixel, vector pixel); 9649int vec_any_ne (vector signed int, vector bool int); 9650int vec_any_ne (vector signed int, vector signed int); 9651int vec_any_ne (vector unsigned int, vector bool int); 9652int vec_any_ne (vector unsigned int, vector unsigned int); 9653int vec_any_ne (vector bool int, vector bool int); 9654int vec_any_ne (vector bool int, vector unsigned int); 9655int vec_any_ne (vector bool int, vector signed int); 9656int vec_any_ne (vector float, vector float); 9657 9658int vec_any_nge (vector float, vector float); 9659 9660int vec_any_ngt (vector float, vector float); 9661 9662int vec_any_nle (vector float, vector float); 9663 9664int vec_any_nlt (vector float, vector float); 9665 9666int vec_any_numeric (vector float); 9667 9668int vec_any_out (vector float, vector float); 9669@end smallexample 9670 9671@node SPARC VIS Built-in Functions 9672@subsection SPARC VIS Built-in Functions 9673 9674GCC supports SIMD operations on the SPARC using both the generic vector 9675extensions (@pxref{Vector Extensions}) as well as built-in functions for 9676the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis} 9677switch, the VIS extension is exposed as the following built-in functions: 9678 9679@smallexample 9680typedef int v2si __attribute__ ((vector_size (8))); 9681typedef short v4hi __attribute__ ((vector_size (8))); 9682typedef short v2hi __attribute__ ((vector_size (4))); 9683typedef char v8qi __attribute__ ((vector_size (8))); 9684typedef char v4qi __attribute__ ((vector_size (4))); 9685 9686void * __builtin_vis_alignaddr (void *, long); 9687int64_t __builtin_vis_faligndatadi (int64_t, int64_t); 9688v2si __builtin_vis_faligndatav2si (v2si, v2si); 9689v4hi __builtin_vis_faligndatav4hi (v4si, v4si); 9690v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi); 9691 9692v4hi __builtin_vis_fexpand (v4qi); 9693 9694v4hi __builtin_vis_fmul8x16 (v4qi, v4hi); 9695v4hi __builtin_vis_fmul8x16au (v4qi, v4hi); 9696v4hi __builtin_vis_fmul8x16al (v4qi, v4hi); 9697v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi); 9698v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi); 9699v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi); 9700v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi); 9701 9702v4qi __builtin_vis_fpack16 (v4hi); 9703v8qi __builtin_vis_fpack32 (v2si, v2si); 9704v2hi __builtin_vis_fpackfix (v2si); 9705v8qi __builtin_vis_fpmerge (v4qi, v4qi); 9706 9707int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t); 9708@end smallexample 9709 9710@node Target Format Checks 9711@section Format Checks Specific to Particular Target Machines 9712 9713For some target machines, GCC supports additional options to the 9714format attribute 9715(@pxref{Function Attributes,,Declaring Attributes of Functions}). 9716 9717@menu 9718* Solaris Format Checks:: 9719@end menu 9720 9721@node Solaris Format Checks 9722@subsection Solaris Format Checks 9723 9724Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format 9725check. @code{cmn_err} accepts a subset of the standard @code{printf} 9726conversions, and the two-argument @code{%b} conversion for displaying 9727bit-fields. See the Solaris man page for @code{cmn_err} for more information. 9728 9729@node Pragmas 9730@section Pragmas Accepted by GCC 9731@cindex pragmas 9732@cindex #pragma 9733 9734GCC supports several types of pragmas, primarily in order to compile 9735code originally written for other compilers. Note that in general 9736we do not recommend the use of pragmas; @xref{Function Attributes}, 9737for further explanation. 9738 9739@menu 9740* ARM Pragmas:: 9741* M32C Pragmas:: 9742* RS/6000 and PowerPC Pragmas:: 9743* Darwin Pragmas:: 9744* Solaris Pragmas:: 9745* Symbol-Renaming Pragmas:: 9746* Structure-Packing Pragmas:: 9747* Weak Pragmas:: 9748* Diagnostic Pragmas:: 9749* Visibility Pragmas:: 9750@end menu 9751 9752@node ARM Pragmas 9753@subsection ARM Pragmas 9754 9755The ARM target defines pragmas for controlling the default addition of 9756@code{long_call} and @code{short_call} attributes to functions. 9757@xref{Function Attributes}, for information about the effects of these 9758attributes. 9759 9760@table @code 9761@item long_calls 9762@cindex pragma, long_calls 9763Set all subsequent functions to have the @code{long_call} attribute. 9764 9765@item no_long_calls 9766@cindex pragma, no_long_calls 9767Set all subsequent functions to have the @code{short_call} attribute. 9768 9769@item long_calls_off 9770@cindex pragma, long_calls_off 9771Do not affect the @code{long_call} or @code{short_call} attributes of 9772subsequent functions. 9773@end table 9774 9775@node M32C Pragmas 9776@subsection M32C Pragmas 9777 9778@table @code 9779@item memregs @var{number} 9780@cindex pragma, memregs 9781Overrides the command line option @code{-memregs=} for the current 9782file. Use with care! This pragma must be before any function in the 9783file, and mixing different memregs values in different objects may 9784make them incompatible. This pragma is useful when a 9785performance-critical function uses a memreg for temporary values, 9786as it may allow you to reduce the number of memregs used. 9787 9788@end table 9789 9790@node RS/6000 and PowerPC Pragmas 9791@subsection RS/6000 and PowerPC Pragmas 9792 9793The RS/6000 and PowerPC targets define one pragma for controlling 9794whether or not the @code{longcall} attribute is added to function 9795declarations by default. This pragma overrides the @option{-mlongcall} 9796option, but not the @code{longcall} and @code{shortcall} attributes. 9797@xref{RS/6000 and PowerPC Options}, for more information about when long 9798calls are and are not necessary. 9799 9800@table @code 9801@item longcall (1) 9802@cindex pragma, longcall 9803Apply the @code{longcall} attribute to all subsequent function 9804declarations. 9805 9806@item longcall (0) 9807Do not apply the @code{longcall} attribute to subsequent function 9808declarations. 9809@end table 9810 9811@c Describe c4x pragmas here. 9812@c Describe h8300 pragmas here. 9813@c Describe sh pragmas here. 9814@c Describe v850 pragmas here. 9815 9816@node Darwin Pragmas 9817@subsection Darwin Pragmas 9818 9819The following pragmas are available for all architectures running the 9820Darwin operating system. These are useful for compatibility with other 9821Mac OS compilers. 9822 9823@table @code 9824@item mark @var{tokens}@dots{} 9825@cindex pragma, mark 9826This pragma is accepted, but has no effect. 9827 9828@item options align=@var{alignment} 9829@cindex pragma, options align 9830This pragma sets the alignment of fields in structures. The values of 9831@var{alignment} may be @code{mac68k}, to emulate m68k alignment, or 9832@code{power}, to emulate PowerPC alignment. Uses of this pragma nest 9833properly; to restore the previous setting, use @code{reset} for the 9834@var{alignment}. 9835 9836@item segment @var{tokens}@dots{} 9837@cindex pragma, segment 9838This pragma is accepted, but has no effect. 9839 9840@item unused (@var{var} [, @var{var}]@dots{}) 9841@cindex pragma, unused 9842This pragma declares variables to be possibly unused. GCC will not 9843produce warnings for the listed variables. The effect is similar to 9844that of the @code{unused} attribute, except that this pragma may appear 9845anywhere within the variables' scopes. 9846@end table 9847 9848@node Solaris Pragmas 9849@subsection Solaris Pragmas 9850 9851The Solaris target supports @code{#pragma redefine_extname} 9852(@pxref{Symbol-Renaming Pragmas}). It also supports additional 9853@code{#pragma} directives for compatibility with the system compiler. 9854 9855@table @code 9856@item align @var{alignment} (@var{variable} [, @var{variable}]...) 9857@cindex pragma, align 9858 9859Increase the minimum alignment of each @var{variable} to @var{alignment}. 9860This is the same as GCC's @code{aligned} attribute @pxref{Variable 9861Attributes}). Macro expansion occurs on the arguments to this pragma 9862when compiling C and Objective-C. It does not currently occur when 9863compiling C++, but this is a bug which may be fixed in a future 9864release. 9865 9866@item fini (@var{function} [, @var{function}]...) 9867@cindex pragma, fini 9868 9869This pragma causes each listed @var{function} to be called after 9870main, or during shared module unloading, by adding a call to the 9871@code{.fini} section. 9872 9873@item init (@var{function} [, @var{function}]...) 9874@cindex pragma, init 9875 9876This pragma causes each listed @var{function} to be called during 9877initialization (before @code{main}) or during shared module loading, by 9878adding a call to the @code{.init} section. 9879 9880@end table 9881 9882@node Symbol-Renaming Pragmas 9883@subsection Symbol-Renaming Pragmas 9884 9885For compatibility with the Solaris and Tru64 UNIX system headers, GCC 9886supports two @code{#pragma} directives which change the name used in 9887assembly for a given declaration. These pragmas are only available on 9888platforms whose system headers need them. To get this effect on all 9889platforms supported by GCC, use the asm labels extension (@pxref{Asm 9890Labels}). 9891 9892@table @code 9893@item redefine_extname @var{oldname} @var{newname} 9894@cindex pragma, redefine_extname 9895 9896This pragma gives the C function @var{oldname} the assembly symbol 9897@var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME} 9898will be defined if this pragma is available (currently only on 9899Solaris). 9900 9901@item extern_prefix @var{string} 9902@cindex pragma, extern_prefix 9903 9904This pragma causes all subsequent external function and variable 9905declarations to have @var{string} prepended to their assembly symbols. 9906This effect may be terminated with another @code{extern_prefix} pragma 9907whose argument is an empty string. The preprocessor macro 9908@code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is 9909available (currently only on Tru64 UNIX)@. 9910@end table 9911 9912These pragmas and the asm labels extension interact in a complicated 9913manner. Here are some corner cases you may want to be aware of. 9914 9915@enumerate 9916@item Both pragmas silently apply only to declarations with external 9917linkage. Asm labels do not have this restriction. 9918 9919@item In C++, both pragmas silently apply only to declarations with 9920``C'' linkage. Again, asm labels do not have this restriction. 9921 9922@item If any of the three ways of changing the assembly name of a 9923declaration is applied to a declaration whose assembly name has 9924already been determined (either by a previous use of one of these 9925features, or because the compiler needed the assembly name in order to 9926generate code), and the new name is different, a warning issues and 9927the name does not change. 9928 9929@item The @var{oldname} used by @code{#pragma redefine_extname} is 9930always the C-language name. 9931 9932@item If @code{#pragma extern_prefix} is in effect, and a declaration 9933occurs with an asm label attached, the prefix is silently ignored for 9934that declaration. 9935 9936@item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname} 9937apply to the same declaration, whichever triggered first wins, and a 9938warning issues if they contradict each other. (We would like to have 9939@code{#pragma redefine_extname} always win, for consistency with asm 9940labels, but if @code{#pragma extern_prefix} triggers first we have no 9941way of knowing that that happened.) 9942@end enumerate 9943 9944@node Structure-Packing Pragmas 9945@subsection Structure-Packing Pragmas 9946 9947For compatibility with Win32, GCC supports a set of @code{#pragma} 9948directives which change the maximum alignment of members of structures 9949(other than zero-width bitfields), unions, and classes subsequently 9950defined. The @var{n} value below always is required to be a small power 9951of two and specifies the new alignment in bytes. 9952 9953@enumerate 9954@item @code{#pragma pack(@var{n})} simply sets the new alignment. 9955@item @code{#pragma pack()} sets the alignment to the one that was in 9956effect when compilation started (see also command line option 9957@option{-fpack-struct[=<n>]} @pxref{Code Gen Options}). 9958@item @code{#pragma pack(push[,@var{n}])} pushes the current alignment 9959setting on an internal stack and then optionally sets the new alignment. 9960@item @code{#pragma pack(pop)} restores the alignment setting to the one 9961saved at the top of the internal stack (and removes that stack entry). 9962Note that @code{#pragma pack([@var{n}])} does not influence this internal 9963stack; thus it is possible to have @code{#pragma pack(push)} followed by 9964multiple @code{#pragma pack(@var{n})} instances and finalized by a single 9965@code{#pragma pack(pop)}. 9966@end enumerate 9967 9968Some targets, e.g. i386 and powerpc, support the @code{ms_struct} 9969@code{#pragma} which lays out a structure as the documented 9970@code{__attribute__ ((ms_struct))}. 9971@enumerate 9972@item @code{#pragma ms_struct on} turns on the layout for structures 9973declared. 9974@item @code{#pragma ms_struct off} turns off the layout for structures 9975declared. 9976@item @code{#pragma ms_struct reset} goes back to the default layout. 9977@end enumerate 9978 9979@node Weak Pragmas 9980@subsection Weak Pragmas 9981 9982For compatibility with SVR4, GCC supports a set of @code{#pragma} 9983directives for declaring symbols to be weak, and defining weak 9984aliases. 9985 9986@table @code 9987@item #pragma weak @var{symbol} 9988@cindex pragma, weak 9989This pragma declares @var{symbol} to be weak, as if the declaration 9990had the attribute of the same name. The pragma may appear before 9991or after the declaration of @var{symbol}, but must appear before 9992either its first use or its definition. It is not an error for 9993@var{symbol} to never be defined at all. 9994 9995@item #pragma weak @var{symbol1} = @var{symbol2} 9996This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}. 9997It is an error if @var{symbol2} is not defined in the current 9998translation unit. 9999@end table 10000 10001@node Diagnostic Pragmas 10002@subsection Diagnostic Pragmas 10003 10004GCC allows the user to selectively enable or disable certain types of 10005diagnostics, and change the kind of the diagnostic. For example, a 10006project's policy might require that all sources compile with 10007@option{-Werror} but certain files might have exceptions allowing 10008specific types of warnings. Or, a project might selectively enable 10009diagnostics and treat them as errors depending on which preprocessor 10010macros are defined. 10011 10012@table @code 10013@item #pragma GCC diagnostic @var{kind} @var{option} 10014@cindex pragma, diagnostic 10015 10016Modifies the disposition of a diagnostic. Note that not all 10017diagnostics are modifiable; at the moment only warnings (normally 10018controlled by @samp{-W...}) can be controlled, and not all of them. 10019Use @option{-fdiagnostics-show-option} to determine which diagnostics 10020are controllable and which option controls them. 10021 10022@var{kind} is @samp{error} to treat this diagnostic as an error, 10023@samp{warning} to treat it like a warning (even if @option{-Werror} is 10024in effect), or @samp{ignored} if the diagnostic is to be ignored. 10025@var{option} is a double quoted string which matches the command line 10026option. 10027 10028@example 10029#pragma GCC diagnostic warning "-Wformat" 10030#pragma GCC diagnostic error "-Wformat" 10031#pragma GCC diagnostic ignored "-Wformat" 10032@end example 10033 10034Note that these pragmas override any command line options. Also, 10035while it is syntactically valid to put these pragmas anywhere in your 10036sources, the only supported location for them is before any data or 10037functions are defined. Doing otherwise may result in unpredictable 10038results depending on how the optimizer manages your sources. If the 10039same option is listed multiple times, the last one specified is the 10040one that is in effect. This pragma is not intended to be a general 10041purpose replacement for command line options, but for implementing 10042strict control over project policies. 10043 10044@end table 10045 10046@node Visibility Pragmas 10047@subsection Visibility Pragmas 10048 10049@table @code 10050@item #pragma GCC visibility push(@var{visibility}) 10051@itemx #pragma GCC visibility pop 10052@cindex pragma, visibility 10053 10054This pragma allows the user to set the visibility for multiple 10055declarations without having to give each a visibility attribute 10056@xref{Function Attributes}, for more information about visibility and 10057the attribute syntax. 10058 10059In C++, @samp{#pragma GCC visibility} affects only namespace-scope 10060declarations. Class members and template specializations are not 10061affected; if you want to override the visibility for a particular 10062member or instantiation, you must use an attribute. 10063 10064@end table 10065 10066@node Unnamed Fields 10067@section Unnamed struct/union fields within structs/unions 10068@cindex struct 10069@cindex union 10070 10071For compatibility with other compilers, GCC allows you to define 10072a structure or union that contains, as fields, structures and unions 10073without names. For example: 10074 10075@smallexample 10076struct @{ 10077 int a; 10078 union @{ 10079 int b; 10080 float c; 10081 @}; 10082 int d; 10083@} foo; 10084@end smallexample 10085 10086In this example, the user would be able to access members of the unnamed 10087union with code like @samp{foo.b}. Note that only unnamed structs and 10088unions are allowed, you may not have, for example, an unnamed 10089@code{int}. 10090 10091You must never create such structures that cause ambiguous field definitions. 10092For example, this structure: 10093 10094@smallexample 10095struct @{ 10096 int a; 10097 struct @{ 10098 int a; 10099 @}; 10100@} foo; 10101@end smallexample 10102 10103It is ambiguous which @code{a} is being referred to with @samp{foo.a}. 10104Such constructs are not supported and must be avoided. In the future, 10105such constructs may be detected and treated as compilation errors. 10106 10107@opindex fms-extensions 10108Unless @option{-fms-extensions} is used, the unnamed field must be a 10109structure or union definition without a tag (for example, @samp{struct 10110@{ int a; @};}). If @option{-fms-extensions} is used, the field may 10111also be a definition with a tag such as @samp{struct foo @{ int a; 10112@};}, a reference to a previously defined structure or union such as 10113@samp{struct foo;}, or a reference to a @code{typedef} name for a 10114previously defined structure or union type. 10115 10116@node Thread-Local 10117@section Thread-Local Storage 10118@cindex Thread-Local Storage 10119@cindex @acronym{TLS} 10120@cindex __thread 10121 10122Thread-local storage (@acronym{TLS}) is a mechanism by which variables 10123are allocated such that there is one instance of the variable per extant 10124thread. The run-time model GCC uses to implement this originates 10125in the IA-64 processor-specific ABI, but has since been migrated 10126to other processors as well. It requires significant support from 10127the linker (@command{ld}), dynamic linker (@command{ld.so}), and 10128system libraries (@file{libc.so} and @file{libpthread.so}), so it 10129is not available everywhere. 10130 10131At the user level, the extension is visible with a new storage 10132class keyword: @code{__thread}. For example: 10133 10134@smallexample 10135__thread int i; 10136extern __thread struct state s; 10137static __thread char *p; 10138@end smallexample 10139 10140The @code{__thread} specifier may be used alone, with the @code{extern} 10141or @code{static} specifiers, but with no other storage class specifier. 10142When used with @code{extern} or @code{static}, @code{__thread} must appear 10143immediately after the other storage class specifier. 10144 10145The @code{__thread} specifier may be applied to any global, file-scoped 10146static, function-scoped static, or static data member of a class. It may 10147not be applied to block-scoped automatic or non-static data member. 10148 10149When the address-of operator is applied to a thread-local variable, it is 10150evaluated at run-time and returns the address of the current thread's 10151instance of that variable. An address so obtained may be used by any 10152thread. When a thread terminates, any pointers to thread-local variables 10153in that thread become invalid. 10154 10155No static initialization may refer to the address of a thread-local variable. 10156 10157In C++, if an initializer is present for a thread-local variable, it must 10158be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++ 10159standard. 10160 10161See @uref{http://people.redhat.com/drepper/tls.pdf, 10162ELF Handling For Thread-Local Storage} for a detailed explanation of 10163the four thread-local storage addressing models, and how the run-time 10164is expected to function. 10165 10166@menu 10167* C99 Thread-Local Edits:: 10168* C++98 Thread-Local Edits:: 10169@end menu 10170 10171@node C99 Thread-Local Edits 10172@subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage 10173 10174The following are a set of changes to ISO/IEC 9899:1999 (aka C99) 10175that document the exact semantics of the language extension. 10176 10177@itemize @bullet 10178@item 10179@cite{5.1.2 Execution environments} 10180 10181Add new text after paragraph 1 10182 10183@quotation 10184Within either execution environment, a @dfn{thread} is a flow of 10185control within a program. It is implementation defined whether 10186or not there may be more than one thread associated with a program. 10187It is implementation defined how threads beyond the first are 10188created, the name and type of the function called at thread 10189startup, and how threads may be terminated. However, objects 10190with thread storage duration shall be initialized before thread 10191startup. 10192@end quotation 10193 10194@item 10195@cite{6.2.4 Storage durations of objects} 10196 10197Add new text before paragraph 3 10198 10199@quotation 10200An object whose identifier is declared with the storage-class 10201specifier @w{@code{__thread}} has @dfn{thread storage duration}. 10202Its lifetime is the entire execution of the thread, and its 10203stored value is initialized only once, prior to thread startup. 10204@end quotation 10205 10206@item 10207@cite{6.4.1 Keywords} 10208 10209Add @code{__thread}. 10210 10211@item 10212@cite{6.7.1 Storage-class specifiers} 10213 10214Add @code{__thread} to the list of storage class specifiers in 10215paragraph 1. 10216 10217Change paragraph 2 to 10218 10219@quotation 10220With the exception of @code{__thread}, at most one storage-class 10221specifier may be given [@dots{}]. The @code{__thread} specifier may 10222be used alone, or immediately following @code{extern} or 10223@code{static}. 10224@end quotation 10225 10226Add new text after paragraph 6 10227 10228@quotation 10229The declaration of an identifier for a variable that has 10230block scope that specifies @code{__thread} shall also 10231specify either @code{extern} or @code{static}. 10232 10233The @code{__thread} specifier shall be used only with 10234variables. 10235@end quotation 10236@end itemize 10237 10238@node C++98 Thread-Local Edits 10239@subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage 10240 10241The following are a set of changes to ISO/IEC 14882:1998 (aka C++98) 10242that document the exact semantics of the language extension. 10243 10244@itemize @bullet 10245@item 10246@b{[intro.execution]} 10247 10248New text after paragraph 4 10249 10250@quotation 10251A @dfn{thread} is a flow of control within the abstract machine. 10252It is implementation defined whether or not there may be more than 10253one thread. 10254@end quotation 10255 10256New text after paragraph 7 10257 10258@quotation 10259It is unspecified whether additional action must be taken to 10260ensure when and whether side effects are visible to other threads. 10261@end quotation 10262 10263@item 10264@b{[lex.key]} 10265 10266Add @code{__thread}. 10267 10268@item 10269@b{[basic.start.main]} 10270 10271Add after paragraph 5 10272 10273@quotation 10274The thread that begins execution at the @code{main} function is called 10275the @dfn{main thread}. It is implementation defined how functions 10276beginning threads other than the main thread are designated or typed. 10277A function so designated, as well as the @code{main} function, is called 10278a @dfn{thread startup function}. It is implementation defined what 10279happens if a thread startup function returns. It is implementation 10280defined what happens to other threads when any thread calls @code{exit}. 10281@end quotation 10282 10283@item 10284@b{[basic.start.init]} 10285 10286Add after paragraph 4 10287 10288@quotation 10289The storage for an object of thread storage duration shall be 10290statically initialized before the first statement of the thread startup 10291function. An object of thread storage duration shall not require 10292dynamic initialization. 10293@end quotation 10294 10295@item 10296@b{[basic.start.term]} 10297 10298Add after paragraph 3 10299 10300@quotation 10301The type of an object with thread storage duration shall not have a 10302non-trivial destructor, nor shall it be an array type whose elements 10303(directly or indirectly) have non-trivial destructors. 10304@end quotation 10305 10306@item 10307@b{[basic.stc]} 10308 10309Add ``thread storage duration'' to the list in paragraph 1. 10310 10311Change paragraph 2 10312 10313@quotation 10314Thread, static, and automatic storage durations are associated with 10315objects introduced by declarations [@dots{}]. 10316@end quotation 10317 10318Add @code{__thread} to the list of specifiers in paragraph 3. 10319 10320@item 10321@b{[basic.stc.thread]} 10322 10323New section before @b{[basic.stc.static]} 10324 10325@quotation 10326The keyword @code{__thread} applied to a non-local object gives the 10327object thread storage duration. 10328 10329A local variable or class data member declared both @code{static} 10330and @code{__thread} gives the variable or member thread storage 10331duration. 10332@end quotation 10333 10334@item 10335@b{[basic.stc.static]} 10336 10337Change paragraph 1 10338 10339@quotation 10340All objects which have neither thread storage duration, dynamic 10341storage duration nor are local [@dots{}]. 10342@end quotation 10343 10344@item 10345@b{[dcl.stc]} 10346 10347Add @code{__thread} to the list in paragraph 1. 10348 10349Change paragraph 1 10350 10351@quotation 10352With the exception of @code{__thread}, at most one 10353@var{storage-class-specifier} shall appear in a given 10354@var{decl-specifier-seq}. The @code{__thread} specifier may 10355be used alone, or immediately following the @code{extern} or 10356@code{static} specifiers. [@dots{}] 10357@end quotation 10358 10359Add after paragraph 5 10360 10361@quotation 10362The @code{__thread} specifier can be applied only to the names of objects 10363and to anonymous unions. 10364@end quotation 10365 10366@item 10367@b{[class.mem]} 10368 10369Add after paragraph 6 10370 10371@quotation 10372Non-@code{static} members shall not be @code{__thread}. 10373@end quotation 10374@end itemize 10375 10376@node C++ Extensions 10377@chapter Extensions to the C++ Language 10378@cindex extensions, C++ language 10379@cindex C++ language extensions 10380 10381The GNU compiler provides these extensions to the C++ language (and you 10382can also use most of the C language extensions in your C++ programs). If you 10383want to write code that checks whether these features are available, you can 10384test for the GNU compiler the same way as for C programs: check for a 10385predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to 10386test specifically for GNU C++ (@pxref{Common Predefined Macros,, 10387Predefined Macros,cpp,The GNU C Preprocessor}). 10388 10389@menu 10390* Volatiles:: What constitutes an access to a volatile object. 10391* Restricted Pointers:: C99 restricted pointers and references. 10392* Vague Linkage:: Where G++ puts inlines, vtables and such. 10393* C++ Interface:: You can use a single C++ header file for both 10394 declarations and definitions. 10395* Template Instantiation:: Methods for ensuring that exactly one copy of 10396 each needed template instantiation is emitted. 10397* Bound member functions:: You can extract a function pointer to the 10398 method denoted by a @samp{->*} or @samp{.*} expression. 10399* C++ Attributes:: Variable, function, and type attributes for C++ only. 10400* Namespace Association:: Strong using-directives for namespace association. 10401* Java Exceptions:: Tweaking exception handling to work with Java. 10402* Deprecated Features:: Things will disappear from g++. 10403* Backwards Compatibility:: Compatibilities with earlier definitions of C++. 10404@end menu 10405 10406@node Volatiles 10407@section When is a Volatile Object Accessed? 10408@cindex accessing volatiles 10409@cindex volatile read 10410@cindex volatile write 10411@cindex volatile access 10412 10413Both the C and C++ standard have the concept of volatile objects. These 10414are normally accessed by pointers and used for accessing hardware. The 10415standards encourage compilers to refrain from optimizations concerning 10416accesses to volatile objects. The C standard leaves it implementation 10417defined as to what constitutes a volatile access. The C++ standard omits 10418to specify this, except to say that C++ should behave in a similar manner 10419to C with respect to volatiles, where possible. The minimum either 10420standard specifies is that at a sequence point all previous accesses to 10421volatile objects have stabilized and no subsequent accesses have 10422occurred. Thus an implementation is free to reorder and combine 10423volatile accesses which occur between sequence points, but cannot do so 10424for accesses across a sequence point. The use of volatiles does not 10425allow you to violate the restriction on updating objects multiple times 10426within a sequence point. 10427 10428@xref{Qualifiers implementation, , Volatile qualifier and the C compiler}. 10429 10430The behavior differs slightly between C and C++ in the non-obvious cases: 10431 10432@smallexample 10433volatile int *src = @var{somevalue}; 10434*src; 10435@end smallexample 10436 10437With C, such expressions are rvalues, and GCC interprets this either as a 10438read of the volatile object being pointed to or only as request to evaluate 10439the side-effects. The C++ standard specifies that such expressions do not 10440undergo lvalue to rvalue conversion, and that the type of the dereferenced 10441object may be incomplete. The C++ standard does not specify explicitly 10442that it is this lvalue to rvalue conversion which may be responsible for 10443causing an access. However, there is reason to believe that it is, 10444because otherwise certain simple expressions become undefined. However, 10445because it would surprise most programmers, G++ treats dereferencing a 10446pointer to volatile object of complete type when the value is unused as 10447GCC would do for an equivalent type in C. When the object has incomplete 10448type, G++ issues a warning; if you wish to force an error, you must 10449force a conversion to rvalue with, for instance, a static cast. 10450 10451When using a reference to volatile, G++ does not treat equivalent 10452expressions as accesses to volatiles, but instead issues a warning that 10453no volatile is accessed. The rationale for this is that otherwise it 10454becomes difficult to determine where volatile access occur, and not 10455possible to ignore the return value from functions returning volatile 10456references. Again, if you wish to force a read, cast the reference to 10457an rvalue. 10458 10459@node Restricted Pointers 10460@section Restricting Pointer Aliasing 10461@cindex restricted pointers 10462@cindex restricted references 10463@cindex restricted this pointer 10464 10465As with the C front end, G++ understands the C99 feature of restricted pointers, 10466specified with the @code{__restrict__}, or @code{__restrict} type 10467qualifier. Because you cannot compile C++ by specifying the @option{-std=c99} 10468language flag, @code{restrict} is not a keyword in C++. 10469 10470In addition to allowing restricted pointers, you can specify restricted 10471references, which indicate that the reference is not aliased in the local 10472context. 10473 10474@smallexample 10475void fn (int *__restrict__ rptr, int &__restrict__ rref) 10476@{ 10477 /* @r{@dots{}} */ 10478@} 10479@end smallexample 10480 10481@noindent 10482In the body of @code{fn}, @var{rptr} points to an unaliased integer and 10483@var{rref} refers to a (different) unaliased integer. 10484 10485You may also specify whether a member function's @var{this} pointer is 10486unaliased by using @code{__restrict__} as a member function qualifier. 10487 10488@smallexample 10489void T::fn () __restrict__ 10490@{ 10491 /* @r{@dots{}} */ 10492@} 10493@end smallexample 10494 10495@noindent 10496Within the body of @code{T::fn}, @var{this} will have the effective 10497definition @code{T *__restrict__ const this}. Notice that the 10498interpretation of a @code{__restrict__} member function qualifier is 10499different to that of @code{const} or @code{volatile} qualifier, in that it 10500is applied to the pointer rather than the object. This is consistent with 10501other compilers which implement restricted pointers. 10502 10503As with all outermost parameter qualifiers, @code{__restrict__} is 10504ignored in function definition matching. This means you only need to 10505specify @code{__restrict__} in a function definition, rather than 10506in a function prototype as well. 10507 10508@node Vague Linkage 10509@section Vague Linkage 10510@cindex vague linkage 10511 10512There are several constructs in C++ which require space in the object 10513file but are not clearly tied to a single translation unit. We say that 10514these constructs have ``vague linkage''. Typically such constructs are 10515emitted wherever they are needed, though sometimes we can be more 10516clever. 10517 10518@table @asis 10519@item Inline Functions 10520Inline functions are typically defined in a header file which can be 10521included in many different compilations. Hopefully they can usually be 10522inlined, but sometimes an out-of-line copy is necessary, if the address 10523of the function is taken or if inlining fails. In general, we emit an 10524out-of-line copy in all translation units where one is needed. As an 10525exception, we only emit inline virtual functions with the vtable, since 10526it will always require a copy. 10527 10528Local static variables and string constants used in an inline function 10529are also considered to have vague linkage, since they must be shared 10530between all inlined and out-of-line instances of the function. 10531 10532@item VTables 10533@cindex vtable 10534C++ virtual functions are implemented in most compilers using a lookup 10535table, known as a vtable. The vtable contains pointers to the virtual 10536functions provided by a class, and each object of the class contains a 10537pointer to its vtable (or vtables, in some multiple-inheritance 10538situations). If the class declares any non-inline, non-pure virtual 10539functions, the first one is chosen as the ``key method'' for the class, 10540and the vtable is only emitted in the translation unit where the key 10541method is defined. 10542 10543@emph{Note:} If the chosen key method is later defined as inline, the 10544vtable will still be emitted in every translation unit which defines it. 10545Make sure that any inline virtuals are declared inline in the class 10546body, even if they are not defined there. 10547 10548@item type_info objects 10549@cindex type_info 10550@cindex RTTI 10551C++ requires information about types to be written out in order to 10552implement @samp{dynamic_cast}, @samp{typeid} and exception handling. 10553For polymorphic classes (classes with virtual functions), the type_info 10554object is written out along with the vtable so that @samp{dynamic_cast} 10555can determine the dynamic type of a class object at runtime. For all 10556other types, we write out the type_info object when it is used: when 10557applying @samp{typeid} to an expression, throwing an object, or 10558referring to a type in a catch clause or exception specification. 10559 10560@item Template Instantiations 10561Most everything in this section also applies to template instantiations, 10562but there are other options as well. 10563@xref{Template Instantiation,,Where's the Template?}. 10564 10565@end table 10566 10567When used with GNU ld version 2.8 or later on an ELF system such as 10568GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of 10569these constructs will be discarded at link time. This is known as 10570COMDAT support. 10571 10572On targets that don't support COMDAT, but do support weak symbols, GCC 10573will use them. This way one copy will override all the others, but 10574the unused copies will still take up space in the executable. 10575 10576For targets which do not support either COMDAT or weak symbols, 10577most entities with vague linkage will be emitted as local symbols to 10578avoid duplicate definition errors from the linker. This will not happen 10579for local statics in inlines, however, as having multiple copies will 10580almost certainly break things. 10581 10582@xref{C++ Interface,,Declarations and Definitions in One Header}, for 10583another way to control placement of these constructs. 10584 10585@node C++ Interface 10586@section #pragma interface and implementation 10587 10588@cindex interface and implementation headers, C++ 10589@cindex C++ interface and implementation headers 10590@cindex pragmas, interface and implementation 10591 10592@code{#pragma interface} and @code{#pragma implementation} provide the 10593user with a way of explicitly directing the compiler to emit entities 10594with vague linkage (and debugging information) in a particular 10595translation unit. 10596 10597@emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in 10598most cases, because of COMDAT support and the ``key method'' heuristic 10599mentioned in @ref{Vague Linkage}. Using them can actually cause your 10600program to grow due to unnecessary out-of-line copies of inline 10601functions. Currently (3.4) the only benefit of these 10602@code{#pragma}s is reduced duplication of debugging information, and 10603that should be addressed soon on DWARF 2 targets with the use of 10604COMDAT groups. 10605 10606@table @code 10607@item #pragma interface 10608@itemx #pragma interface "@var{subdir}/@var{objects}.h" 10609@kindex #pragma interface 10610Use this directive in @emph{header files} that define object classes, to save 10611space in most of the object files that use those classes. Normally, 10612local copies of certain information (backup copies of inline member 10613functions, debugging information, and the internal tables that implement 10614virtual functions) must be kept in each object file that includes class 10615definitions. You can use this pragma to avoid such duplication. When a 10616header file containing @samp{#pragma interface} is included in a 10617compilation, this auxiliary information will not be generated (unless 10618the main input source file itself uses @samp{#pragma implementation}). 10619Instead, the object files will contain references to be resolved at link 10620time. 10621 10622The second form of this directive is useful for the case where you have 10623multiple headers with the same name in different directories. If you 10624use this form, you must specify the same string to @samp{#pragma 10625implementation}. 10626 10627@item #pragma implementation 10628@itemx #pragma implementation "@var{objects}.h" 10629@kindex #pragma implementation 10630Use this pragma in a @emph{main input file}, when you want full output from 10631included header files to be generated (and made globally visible). The 10632included header file, in turn, should use @samp{#pragma interface}. 10633Backup copies of inline member functions, debugging information, and the 10634internal tables used to implement virtual functions are all generated in 10635implementation files. 10636 10637@cindex implied @code{#pragma implementation} 10638@cindex @code{#pragma implementation}, implied 10639@cindex naming convention, implementation headers 10640If you use @samp{#pragma implementation} with no argument, it applies to 10641an include file with the same basename@footnote{A file's @dfn{basename} 10642was the name stripped of all leading path information and of trailing 10643suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source 10644file. For example, in @file{allclass.cc}, giving just 10645@samp{#pragma implementation} 10646by itself is equivalent to @samp{#pragma implementation "allclass.h"}. 10647 10648In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as 10649an implementation file whenever you would include it from 10650@file{allclass.cc} even if you never specified @samp{#pragma 10651implementation}. This was deemed to be more trouble than it was worth, 10652however, and disabled. 10653 10654Use the string argument if you want a single implementation file to 10655include code from multiple header files. (You must also use 10656@samp{#include} to include the header file; @samp{#pragma 10657implementation} only specifies how to use the file---it doesn't actually 10658include it.) 10659 10660There is no way to split up the contents of a single header file into 10661multiple implementation files. 10662@end table 10663 10664@cindex inlining and C++ pragmas 10665@cindex C++ pragmas, effect on inlining 10666@cindex pragmas in C++, effect on inlining 10667@samp{#pragma implementation} and @samp{#pragma interface} also have an 10668effect on function inlining. 10669 10670If you define a class in a header file marked with @samp{#pragma 10671interface}, the effect on an inline function defined in that class is 10672similar to an explicit @code{extern} declaration---the compiler emits 10673no code at all to define an independent version of the function. Its 10674definition is used only for inlining with its callers. 10675 10676@opindex fno-implement-inlines 10677Conversely, when you include the same header file in a main source file 10678that declares it as @samp{#pragma implementation}, the compiler emits 10679code for the function itself; this defines a version of the function 10680that can be found via pointers (or by callers compiled without 10681inlining). If all calls to the function can be inlined, you can avoid 10682emitting the function by compiling with @option{-fno-implement-inlines}. 10683If any calls were not inlined, you will get linker errors. 10684 10685@node Template Instantiation 10686@section Where's the Template? 10687@cindex template instantiation 10688 10689C++ templates are the first language feature to require more 10690intelligence from the environment than one usually finds on a UNIX 10691system. Somehow the compiler and linker have to make sure that each 10692template instance occurs exactly once in the executable if it is needed, 10693and not at all otherwise. There are two basic approaches to this 10694problem, which are referred to as the Borland model and the Cfront model. 10695 10696@table @asis 10697@item Borland model 10698Borland C++ solved the template instantiation problem by adding the code 10699equivalent of common blocks to their linker; the compiler emits template 10700instances in each translation unit that uses them, and the linker 10701collapses them together. The advantage of this model is that the linker 10702only has to consider the object files themselves; there is no external 10703complexity to worry about. This disadvantage is that compilation time 10704is increased because the template code is being compiled repeatedly. 10705Code written for this model tends to include definitions of all 10706templates in the header file, since they must be seen to be 10707instantiated. 10708 10709@item Cfront model 10710The AT&T C++ translator, Cfront, solved the template instantiation 10711problem by creating the notion of a template repository, an 10712automatically maintained place where template instances are stored. A 10713more modern version of the repository works as follows: As individual 10714object files are built, the compiler places any template definitions and 10715instantiations encountered in the repository. At link time, the link 10716wrapper adds in the objects in the repository and compiles any needed 10717instances that were not previously emitted. The advantages of this 10718model are more optimal compilation speed and the ability to use the 10719system linker; to implement the Borland model a compiler vendor also 10720needs to replace the linker. The disadvantages are vastly increased 10721complexity, and thus potential for error; for some code this can be 10722just as transparent, but in practice it can been very difficult to build 10723multiple programs in one directory and one program in multiple 10724directories. Code written for this model tends to separate definitions 10725of non-inline member templates into a separate file, which should be 10726compiled separately. 10727@end table 10728 10729When used with GNU ld version 2.8 or later on an ELF system such as 10730GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the 10731Borland model. On other systems, G++ implements neither automatic 10732model. 10733 10734A future version of G++ will support a hybrid model whereby the compiler 10735will emit any instantiations for which the template definition is 10736included in the compile, and store template definitions and 10737instantiation context information into the object file for the rest. 10738The link wrapper will extract that information as necessary and invoke 10739the compiler to produce the remaining instantiations. The linker will 10740then combine duplicate instantiations. 10741 10742In the mean time, you have the following options for dealing with 10743template instantiations: 10744 10745@enumerate 10746@item 10747@opindex frepo 10748Compile your template-using code with @option{-frepo}. The compiler will 10749generate files with the extension @samp{.rpo} listing all of the 10750template instantiations used in the corresponding object files which 10751could be instantiated there; the link wrapper, @samp{collect2}, will 10752then update the @samp{.rpo} files to tell the compiler where to place 10753those instantiations and rebuild any affected object files. The 10754link-time overhead is negligible after the first pass, as the compiler 10755will continue to place the instantiations in the same files. 10756 10757This is your best option for application code written for the Borland 10758model, as it will just work. Code written for the Cfront model will 10759need to be modified so that the template definitions are available at 10760one or more points of instantiation; usually this is as simple as adding 10761@code{#include <tmethods.cc>} to the end of each template header. 10762 10763For library code, if you want the library to provide all of the template 10764instantiations it needs, just try to link all of its object files 10765together; the link will fail, but cause the instantiations to be 10766generated as a side effect. Be warned, however, that this may cause 10767conflicts if multiple libraries try to provide the same instantiations. 10768For greater control, use explicit instantiation as described in the next 10769option. 10770 10771@item 10772@opindex fno-implicit-templates 10773Compile your code with @option{-fno-implicit-templates} to disable the 10774implicit generation of template instances, and explicitly instantiate 10775all the ones you use. This approach requires more knowledge of exactly 10776which instances you need than do the others, but it's less 10777mysterious and allows greater control. You can scatter the explicit 10778instantiations throughout your program, perhaps putting them in the 10779translation units where the instances are used or the translation units 10780that define the templates themselves; you can put all of the explicit 10781instantiations you need into one big file; or you can create small files 10782like 10783 10784@smallexample 10785#include "Foo.h" 10786#include "Foo.cc" 10787 10788template class Foo<int>; 10789template ostream& operator << 10790 (ostream&, const Foo<int>&); 10791@end smallexample 10792 10793for each of the instances you need, and create a template instantiation 10794library from those. 10795 10796If you are using Cfront-model code, you can probably get away with not 10797using @option{-fno-implicit-templates} when compiling files that don't 10798@samp{#include} the member template definitions. 10799 10800If you use one big file to do the instantiations, you may want to 10801compile it without @option{-fno-implicit-templates} so you get all of the 10802instances required by your explicit instantiations (but not by any 10803other files) without having to specify them as well. 10804 10805G++ has extended the template instantiation syntax given in the ISO 10806standard to allow forward declaration of explicit instantiations 10807(with @code{extern}), instantiation of the compiler support data for a 10808template class (i.e.@: the vtable) without instantiating any of its 10809members (with @code{inline}), and instantiation of only the static data 10810members of a template class, without the support data or member 10811functions (with (@code{static}): 10812 10813@smallexample 10814extern template int max (int, int); 10815inline template class Foo<int>; 10816static template class Foo<int>; 10817@end smallexample 10818 10819@item 10820Do nothing. Pretend G++ does implement automatic instantiation 10821management. Code written for the Borland model will work fine, but 10822each translation unit will contain instances of each of the templates it 10823uses. In a large program, this can lead to an unacceptable amount of code 10824duplication. 10825@end enumerate 10826 10827@node Bound member functions 10828@section Extracting the function pointer from a bound pointer to member function 10829@cindex pmf 10830@cindex pointer to member function 10831@cindex bound pointer to member function 10832 10833In C++, pointer to member functions (PMFs) are implemented using a wide 10834pointer of sorts to handle all the possible call mechanisms; the PMF 10835needs to store information about how to adjust the @samp{this} pointer, 10836and if the function pointed to is virtual, where to find the vtable, and 10837where in the vtable to look for the member function. If you are using 10838PMFs in an inner loop, you should really reconsider that decision. If 10839that is not an option, you can extract the pointer to the function that 10840would be called for a given object/PMF pair and call it directly inside 10841the inner loop, to save a bit of time. 10842 10843Note that you will still be paying the penalty for the call through a 10844function pointer; on most modern architectures, such a call defeats the 10845branch prediction features of the CPU@. This is also true of normal 10846virtual function calls. 10847 10848The syntax for this extension is 10849 10850@smallexample 10851extern A a; 10852extern int (A::*fp)(); 10853typedef int (*fptr)(A *); 10854 10855fptr p = (fptr)(a.*fp); 10856@end smallexample 10857 10858For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}), 10859no object is needed to obtain the address of the function. They can be 10860converted to function pointers directly: 10861 10862@smallexample 10863fptr p1 = (fptr)(&A::foo); 10864@end smallexample 10865 10866@opindex Wno-pmf-conversions 10867You must specify @option{-Wno-pmf-conversions} to use this extension. 10868 10869@node C++ Attributes 10870@section C++-Specific Variable, Function, and Type Attributes 10871 10872Some attributes only make sense for C++ programs. 10873 10874@table @code 10875@item init_priority (@var{priority}) 10876@cindex init_priority attribute 10877 10878 10879In Standard C++, objects defined at namespace scope are guaranteed to be 10880initialized in an order in strict accordance with that of their definitions 10881@emph{in a given translation unit}. No guarantee is made for initializations 10882across translation units. However, GNU C++ allows users to control the 10883order of initialization of objects defined at namespace scope with the 10884@code{init_priority} attribute by specifying a relative @var{priority}, 10885a constant integral expression currently bounded between 101 and 65535 10886inclusive. Lower numbers indicate a higher priority. 10887 10888In the following example, @code{A} would normally be created before 10889@code{B}, but the @code{init_priority} attribute has reversed that order: 10890 10891@smallexample 10892Some_Class A __attribute__ ((init_priority (2000))); 10893Some_Class B __attribute__ ((init_priority (543))); 10894@end smallexample 10895 10896@noindent 10897Note that the particular values of @var{priority} do not matter; only their 10898relative ordering. 10899 10900@item java_interface 10901@cindex java_interface attribute 10902 10903This type attribute informs C++ that the class is a Java interface. It may 10904only be applied to classes declared within an @code{extern "Java"} block. 10905Calls to methods declared in this interface will be dispatched using GCJ's 10906interface table mechanism, instead of regular virtual table dispatch. 10907 10908@end table 10909 10910See also @xref{Namespace Association}. 10911 10912@node Namespace Association 10913@section Namespace Association 10914 10915@strong{Caution:} The semantics of this extension are not fully 10916defined. Users should refrain from using this extension as its 10917semantics may change subtly over time. It is possible that this 10918extension will be removed in future versions of G++. 10919 10920A using-directive with @code{__attribute ((strong))} is stronger 10921than a normal using-directive in two ways: 10922 10923@itemize @bullet 10924@item 10925Templates from the used namespace can be specialized and explicitly 10926instantiated as though they were members of the using namespace. 10927 10928@item 10929The using namespace is considered an associated namespace of all 10930templates in the used namespace for purposes of argument-dependent 10931name lookup. 10932@end itemize 10933 10934The used namespace must be nested within the using namespace so that 10935normal unqualified lookup works properly. 10936 10937This is useful for composing a namespace transparently from 10938implementation namespaces. For example: 10939 10940@smallexample 10941namespace std @{ 10942 namespace debug @{ 10943 template <class T> struct A @{ @}; 10944 @} 10945 using namespace debug __attribute ((__strong__)); 10946 template <> struct A<int> @{ @}; // @r{ok to specialize} 10947 10948 template <class T> void f (A<T>); 10949@} 10950 10951int main() 10952@{ 10953 f (std::A<float>()); // @r{lookup finds} std::f 10954 f (std::A<int>()); 10955@} 10956@end smallexample 10957 10958@node Java Exceptions 10959@section Java Exceptions 10960 10961The Java language uses a slightly different exception handling model 10962from C++. Normally, GNU C++ will automatically detect when you are 10963writing C++ code that uses Java exceptions, and handle them 10964appropriately. However, if C++ code only needs to execute destructors 10965when Java exceptions are thrown through it, GCC will guess incorrectly. 10966Sample problematic code is: 10967 10968@smallexample 10969 struct S @{ ~S(); @}; 10970 extern void bar(); // @r{is written in Java, and may throw exceptions} 10971 void foo() 10972 @{ 10973 S s; 10974 bar(); 10975 @} 10976@end smallexample 10977 10978@noindent 10979The usual effect of an incorrect guess is a link failure, complaining of 10980a missing routine called @samp{__gxx_personality_v0}. 10981 10982You can inform the compiler that Java exceptions are to be used in a 10983translation unit, irrespective of what it might think, by writing 10984@samp{@w{#pragma GCC java_exceptions}} at the head of the file. This 10985@samp{#pragma} must appear before any functions that throw or catch 10986exceptions, or run destructors when exceptions are thrown through them. 10987 10988You cannot mix Java and C++ exceptions in the same translation unit. It 10989is believed to be safe to throw a C++ exception from one file through 10990another file compiled for the Java exception model, or vice versa, but 10991there may be bugs in this area. 10992 10993@node Deprecated Features 10994@section Deprecated Features 10995 10996In the past, the GNU C++ compiler was extended to experiment with new 10997features, at a time when the C++ language was still evolving. Now that 10998the C++ standard is complete, some of those features are superseded by 10999superior alternatives. Using the old features might cause a warning in 11000some cases that the feature will be dropped in the future. In other 11001cases, the feature might be gone already. 11002 11003While the list below is not exhaustive, it documents some of the options 11004that are now deprecated: 11005 11006@table @code 11007@item -fexternal-templates 11008@itemx -falt-external-templates 11009These are two of the many ways for G++ to implement template 11010instantiation. @xref{Template Instantiation}. The C++ standard clearly 11011defines how template definitions have to be organized across 11012implementation units. G++ has an implicit instantiation mechanism that 11013should work just fine for standard-conforming code. 11014 11015@item -fstrict-prototype 11016@itemx -fno-strict-prototype 11017Previously it was possible to use an empty prototype parameter list to 11018indicate an unspecified number of parameters (like C), rather than no 11019parameters, as C++ demands. This feature has been removed, except where 11020it is required for backwards compatibility @xref{Backwards Compatibility}. 11021@end table 11022 11023G++ allows a virtual function returning @samp{void *} to be overridden 11024by one returning a different pointer type. This extension to the 11025covariant return type rules is now deprecated and will be removed from a 11026future version. 11027 11028The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and 11029their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated 11030and will be removed in a future version. Code using these operators 11031should be modified to use @code{std::min} and @code{std::max} instead. 11032 11033The named return value extension has been deprecated, and is now 11034removed from G++. 11035 11036The use of initializer lists with new expressions has been deprecated, 11037and is now removed from G++. 11038 11039Floating and complex non-type template parameters have been deprecated, 11040and are now removed from G++. 11041 11042The implicit typename extension has been deprecated and is now 11043removed from G++. 11044 11045The use of default arguments in function pointers, function typedefs 11046and other places where they are not permitted by the standard is 11047deprecated and will be removed from a future version of G++. 11048 11049G++ allows floating-point literals to appear in integral constant expressions, 11050e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} } 11051This extension is deprecated and will be removed from a future version. 11052 11053G++ allows static data members of const floating-point type to be declared 11054with an initializer in a class definition. The standard only allows 11055initializers for static members of const integral types and const 11056enumeration types so this extension has been deprecated and will be removed 11057from a future version. 11058 11059@node Backwards Compatibility 11060@section Backwards Compatibility 11061@cindex Backwards Compatibility 11062@cindex ARM [Annotated C++ Reference Manual] 11063 11064Now that there is a definitive ISO standard C++, G++ has a specification 11065to adhere to. The C++ language evolved over time, and features that 11066used to be acceptable in previous drafts of the standard, such as the ARM 11067[Annotated C++ Reference Manual], are no longer accepted. In order to allow 11068compilation of C++ written to such drafts, G++ contains some backwards 11069compatibilities. @emph{All such backwards compatibility features are 11070liable to disappear in future versions of G++.} They should be considered 11071deprecated @xref{Deprecated Features}. 11072 11073@table @code 11074@item For scope 11075If a variable is declared at for scope, it used to remain in scope until 11076the end of the scope which contained the for statement (rather than just 11077within the for scope). G++ retains this, but issues a warning, if such a 11078variable is accessed outside the for scope. 11079 11080@item Implicit C language 11081Old C system header files did not contain an @code{extern "C" @{@dots{}@}} 11082scope to set the language. On such systems, all header files are 11083implicitly scoped inside a C language scope. Also, an empty prototype 11084@code{()} will be treated as an unspecified number of arguments, rather 11085than no arguments, as C++ demands. 11086@end table 11087